Comparing Group and Individual Level Assessments of Job Characteristics in Testing the Job...
Transcript of Comparing Group and Individual Level Assessments of Job Characteristics in Testing the Job...
Com paring Group and Individual Level
Assessments of Job Characteristics in Testing
the Job Demand-Con trol Model A Multilevel
Approach
Jan de Jonge14 Gerard J P van Breukelen 2 Jan A Landeweerd 3
an d Fran s J N Nijh u is3
This article describe s a te st of Karase krsquos Job Demand-Control (JD-C) Mode l
using both group and individual level asse ssme nts of job characteristics By group
asse ssm e nts we me an aggre gate d individual data A random sam ple from
ge neral hospitals and nursing hom es include d 16 institutions 64 units and 1489
he alth care worke rs (82 response) Be cause of the hie rarchically nested data
structure ( ie institution s units and individuals) the re search que stions and
hypothe sis were te sted in multile vel regression analyse s (VARCL) The results
revealed both group level and individual level effects with regard to psychological
outcomes and stre sse d the use fulness of multilevel techniques Karasekrsquos JD-C
Mode l was partly confirm ed by finding two interaction effects at group level and
at individual level with regard to job satisfaction and work motivation respe ctively
The discussion focuse s on theore tical methodological and practical implications
of multilevel mode ling with respect to the JD-C Mode l
KEY WORDS job demand-control model multilevel analysis workstre ss health care
INTRODUCTION
In order to redesign jobs or workplaces re searchers we re always in-
te rested in how job characteristics affe ct working pe ople In the last few
decades many organizational re search studie s have shown that the pres-
Hum an Relations Vol 52 No 1 1999
95
0018-7267990100-0095$16001 Oacute 1999 The Tavistock Institute
1Departm ent of Work and O rganizational Psychology Unive rsity of Nijmegen The Nether-
lands2Departm ent of Methodology and Statistics Maastricht University The Ne therlands3Departm ent of He alth O rganization Policies and E conomics Work and He alth Section
Maastricht University The Netherlands4Reque sts for reprints should be addre sse d to Jan de Jonge Departme nt of Work and O r-
ganizational Psychology Unive rsity of Nijmegen PO Box 9104 6500 HE Nijmegen The
Ne therlands
ence or abse nce of certain job characte ristics (e g job demands and job
autonomy) may le ad to attitudinal and behavioural reactions such as job
satisfaction e xhaustion he alth complaints and e ven illness or disability
( e g Hackman amp O ldham 1980 Karasek amp The ore ll 1990 Schnall
Landsbergis amp Baker 1994 Siegrist Pe ter Junge Cre mer amp Seide l 1990
Warr 1987) As a consequence several the oretical models have been de-
ve loped to understand the relationship betwe en such job characte ristics and
employee reactions (cf Kahn amp Byosiere 1992) In ge neral these models
can be described as situation-centered or person-centered (cf Warr 1987)
Situation-centered mode ls focus primarily on e xplanatory factors or e vents
outside a particular worker while person-centered mode ls see k explana-
tions in the person whose actions are be ing studie d (e g cognitions habits
fee lings and nee ds)
A situation-centered mode l on which much of the current job rede sign
research and job stress research is based is the Job De mand-Control (JD-C)
Model deve loped by Karase k (1979) In its basic form the model postulates
that the primary sources of job stress lie within two basic characteristics of
the job itself (1) psychological job demands and (2) job decision latitude
Job demands are defined as psychological stressors that are present in the
work e nvironment (eg workload) Job de cision latitude or job control is
defined as ldquothe working individualrsquos potential control over his tasks and his
conduct during the working dayrdquo (Karasek 1979 pp 289-290)
Psychological strains are a consequence of the joint effects of the de-
mands of a job and the range of job control available to the employee The se
joint e ffects are also called interaction effects The first major prediction of
the JD -C Model is that the strongest ave rsive job-related strain reactions
( like e xhaustion job-relate d anxie ty and health complaints) will occur when
job de mands are high and workerrsquos control is low (ie high strain jobs) The
second prediction of the model which is some time s overlooked is that mo-
tivation le arning and personal growth will occur in situations where both job
demands and workerrsquos control are high (ie active jobs)
Te sts of the JD -C Model can be roughly divided into two cate gorie s
name ly (1) multi-occupation studies and (2) single -occupation studie s (cf
De Jonge 1995 Jones amp Fle tcher 1996) The forme r kind of studies are
usually large -scale e xaminations of national surveys that mainly focus on
cardiovascular disease s These studies freque ntly use an occupation le ve l
of analysis The latte r refers to individual le ve l self-report studie s that focus
on attitudinal outcome s (e g job satisfaction and work motivation) behav-
ioural outcome s (e g abse ntee ism and smoking consumption) and physi-
ological outcome s ( e g blood pre ssure and muscle te nsion) A large
number of studies evaluating the JD-C Model faile d to provide clear and
unam biguous support (De Jonge amp Kompie r 1997 Jone s amp Fle tche r
96 de Jon ge van Breukelen Landeweerd and Nijhu is
1996) Several conceptual as we ll as methodological criticisms have been
expre ssed in the lite rature that might e xplain the inconclusive re sults First
the re is some doubt as to the conceptualization and operationalization of
job demands and job decision latitude particularly in the multi-occupation
studies Job demands as used by Karasek seem to be a mixture of job stres-
sors job complexity and lack of control (Jones amp Fletcher 1996 Kasl
1996) For instance ldquohectic workrdquo may have different meanings depending
on the particular job of the re spondent In addition to this the original
scale of Karasek (1979 1985) includes not only purely descriptive items
but also affe ctive items which may le ad to spurious re lationships with the
depe nde nt variable s (cf Wall Jackson Mullarkey amp Parker 1996) Job
decision latitude reflects the term ldquojob autonomyrdquo as it is used in the job
redesign tradition but the operationalization of this construct also contains
eleme nts like skill leve l skill varie ty and e ven job scope or job comple xity
(Frese 1989 Ganster 1989 1995 Kasl 1996) So the me asurement of
both job demands and job decision latitude may be confused with other
job characte ristics
Second in multi-occupation studie s re sults may be confounded with
for instance socioeconomic status and health behaviour of the workers
Jobs concomitantly high in demands and low in control are likely ove r-
represented by e mployee s of low socioeconomic class and low health be-
haviour who share other risks for cardiovascular diseases (e g Sie grist e t
al 1990) Third the finding of significant inte raction e ffects of job de-
mands and job decision latitude may be affe cted by the use of different
me thods andor diffe re nt kinds of inte raction te rms (e g Kasl 1996
Landsbergis Schnall Warre n Pickering amp Schwartz 1994) More specifi-
cally both analysis of variance and regression analysis have been carried
out In general the first me thod produces re sults that are re lative ly often
in favor of the JD -C Model However it has been argue d that inte ractions
ideally should be tested with moderated regression analysis (Aiken amp West
1991 Landsbergis e t al 1994) Q uite often studies that use the latter do
not yie ld positive re sults (e g Payne amp Fle tche r 1983 Spe ctor 1987)
These differences in re sults can be e xplained by powe r differences betwe en
different statistical me thods Fourth it is also possible that the inconsis-
tentmdashinte ractivemdashfindings could be due to one or more moderator vari-
ables influe ncing re lations betwe e n job characte ristics and outcomes (eg
personality characte ristics or workplace social support) A number of re-
search studies have started in order to investigate these moderating e ffects
some of them with quite promising re sults (see Johnson 1989 Jones amp
Fletcher 1996 Parkes 1991 De Rijk Le Blanc Schaufeli amp De Jonge
1998)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 97
Finally an alternative explanation for the lack of consistency may be
the assessme nt of job characte ristics Basically the JD-C Model is situ-
ation-centered in that it emphasize s the particular role of the work envi-
ronme nt in the prediction of job-related strain It is self-evide nt that in
case s whe re the aim is to rede sign work indepe nde ntly of a particular
worker one nee ds more objective parame ters which are able to show to
have an impact on peoplersquos behaviour (Frese amp Z apf 1994) In other words
job de mands and job control are in the ory characteristics of jobs rather
than characte ristics of people
Job de mands and job control ge nerally have bee n measured by means
of se lf-report que stionnaires Howeve r the main problem with such ques-
tionnaire instrume nts is thatmdashas re flections of the objective work e nviron-
mentmdashthey are more prone to bias than objective instruments Se lf-report
que stionnaires me asure job characte ristics as perceive d by the individual
worker and may the re fore not reflect the objective task accurate ly (Karase k
amp Theore ll 1990) For instance someone can work in a job with high de-
mands and not report feelings of high workload
Consequently the me thodology of the JD -C studies (the multi-occu-
pation studie s in particular) has varied considerably Most studies re ly heav-
ily on the individualrsquos perception or description of job characte ristics (cf
Soumlde rfeldt Soumlderfeldt Jones O rsquoCampo Muntaner O hlson amp Warg 1996)
A few have highlighted the grouprsquos perception or description as a re flection
of the objective environment (cf De Jonge amp Kompier 1997 Schnall e t
al 1994) This point will be discussed in more detail in the next section
THE ASSESSMENT OF JOB CHARACTE RISTICS
The job characteristics of the JD-C Model have bee n measured in two
different ways (1) ldquoobjective rdquo and (2) ldquosubjectiverdquo (Karase k amp Theorell
1990 Kristensen 1995) ldquoO bjectiverdquo and ldquosubjectiverdquo have been put in quo-
tation marks because they are used inconsistently in the literature on work-
stress So-calle d ldquoobjective rdquo job characte ristics may be de fine d as ones
which are assessed independently of the job incumbent (cf Fre se amp Zapf
1988 1994 Spe ctor 1992) Examples are physical and social characteristics
of the work e nvironment or expe rt ratings Accordingly ldquosubjectiverdquo job
characte ristics are de pendent on employe ersquos cognitive and e motional proc-
essing (like appraisals) and the ir ability in coping (Fre se amp Zapf 1988
1994 Lazarus 1995)
Currently the objective method is carried out through (1) direct in-
depe nde nt me asurement and (2) observe rsrsquo ratings (e g Frese amp Zapf
1988 Kristensen 1995 Schnall e t al 1994 Theorell amp Karasek 1996)
The main problem with direct measurement of objective characteristics is
98 de Jon ge van Breukelen Landeweerd and Nijhu is
that some aspe cts are difficult to e xpress in concre te physical te rms (Warr
1987) For e xample job autonomy and job clarity do not provide usable
ge neral marke rs The second approach seems to be objective as me ntioned
above but may be influenced by observersrsquo bias The measurement yie lds
incomple te and partially invalid information due to limited observation
time or space and the e ffects of the observation itse lf (Fre se amp Z apf 1988)
Additionally obse rve rsrsquo ratings see m to suffer from stronge r halo and
stereotyping effects than subjectsrsquo assessments (Frese 1985 Semmer Zapf
amp Gre if 1996 Spector Brannick amp Coove rt 1989)
The subjective method is usually carried out by means of se lf-report
questionnaire s This kind of me asureme nt has several problems as we ll
(e g Frese 1985 Frese amp Z apf 1988 Kasl 1989 1996 Spector 1992
Zapf 1989) such as (1) conceptual overlap betwe en indepe ndent and de-
pendent me asures in that both me asures reflect the same construct (2)
common me thod variance because the information is derive d from the
same source (eg central tendency acquiescence) (3) influence of a third
variable that causes a spurious re lationship (eg a pe rsonal trait) (4) the
potential influencemdashor alterationmdashof the e stimation of job characteristics
due to the presence of an outcome variable (e g health complaints) (5)
possible e ffect of demand characte ristics of the re search context and e x-
perimenter e ffects re sulting in false correlations betwe en job characteristics
and outcome variable s (6) job incumbents may be so used to the ir work
situation that they deny some of the occupational hazards
Spector (1992) argued that more work nee ds to be done e xploring the
accuracy of reports both from the perspective of the job incumbent and
from alte rnative sources to reflect the work conditions O ne way to avoid
the above -me ntioned problems as far as possible is to look for such alte r-
native sources So-calle d group assessm ents seem to be ve ry useful in meas-
uring job characte ristics (Frese 1985 1989 Fre se amp Zapf 1988 Spector
1992) By group assessments we mean that the scores of job incumbents
with the same job and working in ne arly identical workplaces are aggre -
gate d into one ge neral score The group assessments can be described as
the group e stimate s of the re spective job characte ristic for each job incum-
bent These me asures re fe r to that part of a particular job characteristic
that different workers doing ne arly the same job have in common (see also
Semme r e t al 1996) In other words group asse ssments are base d on the
concept of the ideal typical worker ( ie an ave rage worker with sufficient
skills to perform his or he r tasks) According to Fre se and Zapf (1988
1994) group assessments are more objective measures in the sense that
the influence of idiosyncraticmdashindividualmdashperceptions and possibly illusory
answers are reduced In addition to this the e xpertise of workers is taken
into account and problems of brie f pe riods of observation are avoide d Fi-
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 99
nally group assessments see m to be less subject to methodological prob-
le ms such as attenuation because unsystematic error variance is reduced
Thus group data are likely to be more reliable than individual assessme nts
Theoretical justification for considering group assessments as more ob-
je ctive me asures of the two job characte ristics in the JD-C Model can be
found in the literature on social and inte ractional psychology industrial
and organizational psychology and organizational behaviour and e cology
(eg Alderfer 1987 Baron amp Boudreau 1987 Gibson 1979 Hackman
1976 Kraheacute 1992 Schneide r 1987 Warr 1987) From the point of view
of modern interactionism for instance there e xists situational features that
can be consensually regarde d as part of the situation These features are
conceptualized as situational ldquoaffordance srdquo and are constructs that imply
the complimentarity of both work e nvironments and workers What makes
the features ldquoquasi-objectiverdquo is the fact that they be long to the work situ-
ation irrespective of whe the r or not the job incumbent recognizes them as
such For e xample a danger signal te lls a majority of workers to run to
the e me rge ncy e xit Affordance s have some conce ptual similarity with
Hackman rsquos (1976) ldquogroup-supplied stimulirdquoAnother justification is provide d by Schne ide rrsquos (1987) ASA frame-
work This framework sugge sts that employe es will expe rience similar work-
ing conditions due to attraction se lection and attrition processes in an
organization This re sults in similarity in behaviour within a work se tting
In othe r words the pe ople make the workplace (Schneide r 1987) Al-
though Schneider bases his framework primarily at the organizational leve l
it is most likely that his ideas can be e mployed at the group le ve l of analysis
The claim that group assessment ( ie aggre gated data) is more ob-
je ctive is corroborate d by se ve ral e mpirical findings For instance a
me taanalysis of 16 conve rge nce studie s conducte d by Spe ctor (1992)
showe d that aggre gate -leve l corre lations betwee n job characte ristics and
outcome s were similar to individual le ve l correlations Moreover it ap-
peared that the converge nt validity at the aggre gate d le ve l was rather large
and e ven large r than at the individual leve l
PURPOSE
The purpose of this study was to test the JD-C Model using both group
and individual indicators of job characteristics The reason why we used
group and individual assessments of job demands and job autonomy is that
we want to find out whether aggre gated job characte ristics data significantly
add explained variance to individual job characteristics data with regard to
employee he alth In othe r words the key question is whether aggre gated
job characteristics e xplain additional variance in individualsrsquo attitudinal re-
100 de Jon ge van Breukelen Landeweerd and Nijhu is
actions This could mean that some of the explained variance of employe e
health is attributable to some more objective features outside the individual
pe rce ption or asse ssme nt (e g affordance s or group-supplie d stimuli)
which has important theoretical and practical conseque nces For instance
if job-relate d strain captures group le ve l variance the aspirations of the
JD -C Model to be situation-centered would rece ive support In addition
to all this we tried to measure the two job characte ristics (job demands
and job autonomy) more precise ly and e xte nsive ly Lastly the inte raction
effect of demands and autonomy was tested with regre ssion techniques in-
cluding a multiplicative interaction term (cf Aiken amp West 1991 Jaccard
Turrisi amp Wan 1990 Landsbergis et al 1994)
We tested the JD-C Model in a single -occupation sample of health care
workers because the ir professions are ve ry suitable for testing the JD-C
Model for several reasons (e g Fox Dwye r amp Ganster 1993 Ganster amp
Fusilier 1989) First health care workers seem to be subject to stressful work
conditions (ie high job demands) Second job autonomy is an important
tool in present-day care de livery systems Third because of different types of
health care areas and different specialties he alth care workers are a re lative ly
hete rogeneous group Moreover Ganster and Fusilie r (1989) showed that
there is as much variability in health care sample s as there is across a fairly
wide range of occupations Finally the relative homogeneity in social class
restricts the confounding e ffect of socioeconomic status
In line with the JD-C Model it is hypothesized that job demands and
job autonomy have an interaction effect with respect to e mployee he alth
To be more specific we e xpe ct that job autonomy attenuates the adve rse
effects of job demands on employe e health In addition to this we will
compare aggre gated and individual data by means of the variance to be
explaine d We e xpect that give n that aggre gated data correspond to the
more objective work conditions then if the JD-C Model is correct the
use of aggre gated data should reve al stronger interaction e ffects
METHOD
Sam ple
A random sample of 16 institutions was drawn from all gene ral hos-
pitals and nursing home s in the Netherlands (N = 218) Eight general hos-
pitals and e ight nursing home s participated in the study He alth care
workers in four units in each institute we re asked to comple te a que stion-
naire Six types of units were present in the sample an intensive care unit
(ICU) a psychiatric unit an inte rnal unit a surgical unit a somatic unit
and finally a psychogeriatric unit The initial sample consisted of 1806
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 101
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
ence or abse nce of certain job characte ristics (e g job demands and job
autonomy) may le ad to attitudinal and behavioural reactions such as job
satisfaction e xhaustion he alth complaints and e ven illness or disability
( e g Hackman amp O ldham 1980 Karasek amp The ore ll 1990 Schnall
Landsbergis amp Baker 1994 Siegrist Pe ter Junge Cre mer amp Seide l 1990
Warr 1987) As a consequence several the oretical models have been de-
ve loped to understand the relationship betwe en such job characte ristics and
employee reactions (cf Kahn amp Byosiere 1992) In ge neral these models
can be described as situation-centered or person-centered (cf Warr 1987)
Situation-centered mode ls focus primarily on e xplanatory factors or e vents
outside a particular worker while person-centered mode ls see k explana-
tions in the person whose actions are be ing studie d (e g cognitions habits
fee lings and nee ds)
A situation-centered mode l on which much of the current job rede sign
research and job stress research is based is the Job De mand-Control (JD-C)
Model deve loped by Karase k (1979) In its basic form the model postulates
that the primary sources of job stress lie within two basic characteristics of
the job itself (1) psychological job demands and (2) job decision latitude
Job demands are defined as psychological stressors that are present in the
work e nvironment (eg workload) Job de cision latitude or job control is
defined as ldquothe working individualrsquos potential control over his tasks and his
conduct during the working dayrdquo (Karasek 1979 pp 289-290)
Psychological strains are a consequence of the joint effects of the de-
mands of a job and the range of job control available to the employee The se
joint e ffects are also called interaction effects The first major prediction of
the JD -C Model is that the strongest ave rsive job-related strain reactions
( like e xhaustion job-relate d anxie ty and health complaints) will occur when
job de mands are high and workerrsquos control is low (ie high strain jobs) The
second prediction of the model which is some time s overlooked is that mo-
tivation le arning and personal growth will occur in situations where both job
demands and workerrsquos control are high (ie active jobs)
Te sts of the JD -C Model can be roughly divided into two cate gorie s
name ly (1) multi-occupation studies and (2) single -occupation studie s (cf
De Jonge 1995 Jones amp Fle tcher 1996) The forme r kind of studies are
usually large -scale e xaminations of national surveys that mainly focus on
cardiovascular disease s These studies freque ntly use an occupation le ve l
of analysis The latte r refers to individual le ve l self-report studie s that focus
on attitudinal outcome s (e g job satisfaction and work motivation) behav-
ioural outcome s (e g abse ntee ism and smoking consumption) and physi-
ological outcome s ( e g blood pre ssure and muscle te nsion) A large
number of studies evaluating the JD-C Model faile d to provide clear and
unam biguous support (De Jonge amp Kompie r 1997 Jone s amp Fle tche r
96 de Jon ge van Breukelen Landeweerd and Nijhu is
1996) Several conceptual as we ll as methodological criticisms have been
expre ssed in the lite rature that might e xplain the inconclusive re sults First
the re is some doubt as to the conceptualization and operationalization of
job demands and job decision latitude particularly in the multi-occupation
studies Job demands as used by Karasek seem to be a mixture of job stres-
sors job complexity and lack of control (Jones amp Fletcher 1996 Kasl
1996) For instance ldquohectic workrdquo may have different meanings depending
on the particular job of the re spondent In addition to this the original
scale of Karasek (1979 1985) includes not only purely descriptive items
but also affe ctive items which may le ad to spurious re lationships with the
depe nde nt variable s (cf Wall Jackson Mullarkey amp Parker 1996) Job
decision latitude reflects the term ldquojob autonomyrdquo as it is used in the job
redesign tradition but the operationalization of this construct also contains
eleme nts like skill leve l skill varie ty and e ven job scope or job comple xity
(Frese 1989 Ganster 1989 1995 Kasl 1996) So the me asurement of
both job demands and job decision latitude may be confused with other
job characte ristics
Second in multi-occupation studie s re sults may be confounded with
for instance socioeconomic status and health behaviour of the workers
Jobs concomitantly high in demands and low in control are likely ove r-
represented by e mployee s of low socioeconomic class and low health be-
haviour who share other risks for cardiovascular diseases (e g Sie grist e t
al 1990) Third the finding of significant inte raction e ffects of job de-
mands and job decision latitude may be affe cted by the use of different
me thods andor diffe re nt kinds of inte raction te rms (e g Kasl 1996
Landsbergis Schnall Warre n Pickering amp Schwartz 1994) More specifi-
cally both analysis of variance and regression analysis have been carried
out In general the first me thod produces re sults that are re lative ly often
in favor of the JD -C Model However it has been argue d that inte ractions
ideally should be tested with moderated regression analysis (Aiken amp West
1991 Landsbergis e t al 1994) Q uite often studies that use the latter do
not yie ld positive re sults (e g Payne amp Fle tche r 1983 Spe ctor 1987)
These differences in re sults can be e xplained by powe r differences betwe en
different statistical me thods Fourth it is also possible that the inconsis-
tentmdashinte ractivemdashfindings could be due to one or more moderator vari-
ables influe ncing re lations betwe e n job characte ristics and outcomes (eg
personality characte ristics or workplace social support) A number of re-
search studies have started in order to investigate these moderating e ffects
some of them with quite promising re sults (see Johnson 1989 Jones amp
Fletcher 1996 Parkes 1991 De Rijk Le Blanc Schaufeli amp De Jonge
1998)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 97
Finally an alternative explanation for the lack of consistency may be
the assessme nt of job characte ristics Basically the JD-C Model is situ-
ation-centered in that it emphasize s the particular role of the work envi-
ronme nt in the prediction of job-related strain It is self-evide nt that in
case s whe re the aim is to rede sign work indepe nde ntly of a particular
worker one nee ds more objective parame ters which are able to show to
have an impact on peoplersquos behaviour (Frese amp Z apf 1994) In other words
job de mands and job control are in the ory characteristics of jobs rather
than characte ristics of people
Job de mands and job control ge nerally have bee n measured by means
of se lf-report que stionnaires Howeve r the main problem with such ques-
tionnaire instrume nts is thatmdashas re flections of the objective work e nviron-
mentmdashthey are more prone to bias than objective instruments Se lf-report
que stionnaires me asure job characte ristics as perceive d by the individual
worker and may the re fore not reflect the objective task accurate ly (Karase k
amp Theore ll 1990) For instance someone can work in a job with high de-
mands and not report feelings of high workload
Consequently the me thodology of the JD -C studies (the multi-occu-
pation studie s in particular) has varied considerably Most studies re ly heav-
ily on the individualrsquos perception or description of job characte ristics (cf
Soumlde rfeldt Soumlderfeldt Jones O rsquoCampo Muntaner O hlson amp Warg 1996)
A few have highlighted the grouprsquos perception or description as a re flection
of the objective environment (cf De Jonge amp Kompier 1997 Schnall e t
al 1994) This point will be discussed in more detail in the next section
THE ASSESSMENT OF JOB CHARACTE RISTICS
The job characteristics of the JD-C Model have bee n measured in two
different ways (1) ldquoobjective rdquo and (2) ldquosubjectiverdquo (Karase k amp Theorell
1990 Kristensen 1995) ldquoO bjectiverdquo and ldquosubjectiverdquo have been put in quo-
tation marks because they are used inconsistently in the literature on work-
stress So-calle d ldquoobjective rdquo job characte ristics may be de fine d as ones
which are assessed independently of the job incumbent (cf Fre se amp Zapf
1988 1994 Spe ctor 1992) Examples are physical and social characteristics
of the work e nvironment or expe rt ratings Accordingly ldquosubjectiverdquo job
characte ristics are de pendent on employe ersquos cognitive and e motional proc-
essing (like appraisals) and the ir ability in coping (Fre se amp Zapf 1988
1994 Lazarus 1995)
Currently the objective method is carried out through (1) direct in-
depe nde nt me asurement and (2) observe rsrsquo ratings (e g Frese amp Zapf
1988 Kristensen 1995 Schnall e t al 1994 Theorell amp Karasek 1996)
The main problem with direct measurement of objective characteristics is
98 de Jon ge van Breukelen Landeweerd and Nijhu is
that some aspe cts are difficult to e xpress in concre te physical te rms (Warr
1987) For e xample job autonomy and job clarity do not provide usable
ge neral marke rs The second approach seems to be objective as me ntioned
above but may be influenced by observersrsquo bias The measurement yie lds
incomple te and partially invalid information due to limited observation
time or space and the e ffects of the observation itse lf (Fre se amp Z apf 1988)
Additionally obse rve rsrsquo ratings see m to suffer from stronge r halo and
stereotyping effects than subjectsrsquo assessments (Frese 1985 Semmer Zapf
amp Gre if 1996 Spector Brannick amp Coove rt 1989)
The subjective method is usually carried out by means of se lf-report
questionnaire s This kind of me asureme nt has several problems as we ll
(e g Frese 1985 Frese amp Z apf 1988 Kasl 1989 1996 Spector 1992
Zapf 1989) such as (1) conceptual overlap betwe en indepe ndent and de-
pendent me asures in that both me asures reflect the same construct (2)
common me thod variance because the information is derive d from the
same source (eg central tendency acquiescence) (3) influence of a third
variable that causes a spurious re lationship (eg a pe rsonal trait) (4) the
potential influencemdashor alterationmdashof the e stimation of job characteristics
due to the presence of an outcome variable (e g health complaints) (5)
possible e ffect of demand characte ristics of the re search context and e x-
perimenter e ffects re sulting in false correlations betwe en job characteristics
and outcome variable s (6) job incumbents may be so used to the ir work
situation that they deny some of the occupational hazards
Spector (1992) argued that more work nee ds to be done e xploring the
accuracy of reports both from the perspective of the job incumbent and
from alte rnative sources to reflect the work conditions O ne way to avoid
the above -me ntioned problems as far as possible is to look for such alte r-
native sources So-calle d group assessm ents seem to be ve ry useful in meas-
uring job characte ristics (Frese 1985 1989 Fre se amp Zapf 1988 Spector
1992) By group assessments we mean that the scores of job incumbents
with the same job and working in ne arly identical workplaces are aggre -
gate d into one ge neral score The group assessments can be described as
the group e stimate s of the re spective job characte ristic for each job incum-
bent These me asures re fe r to that part of a particular job characteristic
that different workers doing ne arly the same job have in common (see also
Semme r e t al 1996) In other words group asse ssments are base d on the
concept of the ideal typical worker ( ie an ave rage worker with sufficient
skills to perform his or he r tasks) According to Fre se and Zapf (1988
1994) group assessments are more objective measures in the sense that
the influence of idiosyncraticmdashindividualmdashperceptions and possibly illusory
answers are reduced In addition to this the e xpertise of workers is taken
into account and problems of brie f pe riods of observation are avoide d Fi-
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 99
nally group assessments see m to be less subject to methodological prob-
le ms such as attenuation because unsystematic error variance is reduced
Thus group data are likely to be more reliable than individual assessme nts
Theoretical justification for considering group assessments as more ob-
je ctive me asures of the two job characte ristics in the JD-C Model can be
found in the literature on social and inte ractional psychology industrial
and organizational psychology and organizational behaviour and e cology
(eg Alderfer 1987 Baron amp Boudreau 1987 Gibson 1979 Hackman
1976 Kraheacute 1992 Schneide r 1987 Warr 1987) From the point of view
of modern interactionism for instance there e xists situational features that
can be consensually regarde d as part of the situation These features are
conceptualized as situational ldquoaffordance srdquo and are constructs that imply
the complimentarity of both work e nvironments and workers What makes
the features ldquoquasi-objectiverdquo is the fact that they be long to the work situ-
ation irrespective of whe the r or not the job incumbent recognizes them as
such For e xample a danger signal te lls a majority of workers to run to
the e me rge ncy e xit Affordance s have some conce ptual similarity with
Hackman rsquos (1976) ldquogroup-supplied stimulirdquoAnother justification is provide d by Schne ide rrsquos (1987) ASA frame-
work This framework sugge sts that employe es will expe rience similar work-
ing conditions due to attraction se lection and attrition processes in an
organization This re sults in similarity in behaviour within a work se tting
In othe r words the pe ople make the workplace (Schneide r 1987) Al-
though Schneider bases his framework primarily at the organizational leve l
it is most likely that his ideas can be e mployed at the group le ve l of analysis
The claim that group assessment ( ie aggre gated data) is more ob-
je ctive is corroborate d by se ve ral e mpirical findings For instance a
me taanalysis of 16 conve rge nce studie s conducte d by Spe ctor (1992)
showe d that aggre gate -leve l corre lations betwee n job characte ristics and
outcome s were similar to individual le ve l correlations Moreover it ap-
peared that the converge nt validity at the aggre gate d le ve l was rather large
and e ven large r than at the individual leve l
PURPOSE
The purpose of this study was to test the JD-C Model using both group
and individual indicators of job characteristics The reason why we used
group and individual assessments of job demands and job autonomy is that
we want to find out whether aggre gated job characte ristics data significantly
add explained variance to individual job characteristics data with regard to
employee he alth In othe r words the key question is whether aggre gated
job characteristics e xplain additional variance in individualsrsquo attitudinal re-
100 de Jon ge van Breukelen Landeweerd and Nijhu is
actions This could mean that some of the explained variance of employe e
health is attributable to some more objective features outside the individual
pe rce ption or asse ssme nt (e g affordance s or group-supplie d stimuli)
which has important theoretical and practical conseque nces For instance
if job-relate d strain captures group le ve l variance the aspirations of the
JD -C Model to be situation-centered would rece ive support In addition
to all this we tried to measure the two job characte ristics (job demands
and job autonomy) more precise ly and e xte nsive ly Lastly the inte raction
effect of demands and autonomy was tested with regre ssion techniques in-
cluding a multiplicative interaction term (cf Aiken amp West 1991 Jaccard
Turrisi amp Wan 1990 Landsbergis et al 1994)
We tested the JD-C Model in a single -occupation sample of health care
workers because the ir professions are ve ry suitable for testing the JD-C
Model for several reasons (e g Fox Dwye r amp Ganster 1993 Ganster amp
Fusilier 1989) First health care workers seem to be subject to stressful work
conditions (ie high job demands) Second job autonomy is an important
tool in present-day care de livery systems Third because of different types of
health care areas and different specialties he alth care workers are a re lative ly
hete rogeneous group Moreover Ganster and Fusilie r (1989) showed that
there is as much variability in health care sample s as there is across a fairly
wide range of occupations Finally the relative homogeneity in social class
restricts the confounding e ffect of socioeconomic status
In line with the JD-C Model it is hypothesized that job demands and
job autonomy have an interaction effect with respect to e mployee he alth
To be more specific we e xpe ct that job autonomy attenuates the adve rse
effects of job demands on employe e health In addition to this we will
compare aggre gated and individual data by means of the variance to be
explaine d We e xpect that give n that aggre gated data correspond to the
more objective work conditions then if the JD-C Model is correct the
use of aggre gated data should reve al stronger interaction e ffects
METHOD
Sam ple
A random sample of 16 institutions was drawn from all gene ral hos-
pitals and nursing home s in the Netherlands (N = 218) Eight general hos-
pitals and e ight nursing home s participated in the study He alth care
workers in four units in each institute we re asked to comple te a que stion-
naire Six types of units were present in the sample an intensive care unit
(ICU) a psychiatric unit an inte rnal unit a surgical unit a somatic unit
and finally a psychogeriatric unit The initial sample consisted of 1806
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 101
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
1996) Several conceptual as we ll as methodological criticisms have been
expre ssed in the lite rature that might e xplain the inconclusive re sults First
the re is some doubt as to the conceptualization and operationalization of
job demands and job decision latitude particularly in the multi-occupation
studies Job demands as used by Karasek seem to be a mixture of job stres-
sors job complexity and lack of control (Jones amp Fletcher 1996 Kasl
1996) For instance ldquohectic workrdquo may have different meanings depending
on the particular job of the re spondent In addition to this the original
scale of Karasek (1979 1985) includes not only purely descriptive items
but also affe ctive items which may le ad to spurious re lationships with the
depe nde nt variable s (cf Wall Jackson Mullarkey amp Parker 1996) Job
decision latitude reflects the term ldquojob autonomyrdquo as it is used in the job
redesign tradition but the operationalization of this construct also contains
eleme nts like skill leve l skill varie ty and e ven job scope or job comple xity
(Frese 1989 Ganster 1989 1995 Kasl 1996) So the me asurement of
both job demands and job decision latitude may be confused with other
job characte ristics
Second in multi-occupation studie s re sults may be confounded with
for instance socioeconomic status and health behaviour of the workers
Jobs concomitantly high in demands and low in control are likely ove r-
represented by e mployee s of low socioeconomic class and low health be-
haviour who share other risks for cardiovascular diseases (e g Sie grist e t
al 1990) Third the finding of significant inte raction e ffects of job de-
mands and job decision latitude may be affe cted by the use of different
me thods andor diffe re nt kinds of inte raction te rms (e g Kasl 1996
Landsbergis Schnall Warre n Pickering amp Schwartz 1994) More specifi-
cally both analysis of variance and regression analysis have been carried
out In general the first me thod produces re sults that are re lative ly often
in favor of the JD -C Model However it has been argue d that inte ractions
ideally should be tested with moderated regression analysis (Aiken amp West
1991 Landsbergis e t al 1994) Q uite often studies that use the latter do
not yie ld positive re sults (e g Payne amp Fle tche r 1983 Spe ctor 1987)
These differences in re sults can be e xplained by powe r differences betwe en
different statistical me thods Fourth it is also possible that the inconsis-
tentmdashinte ractivemdashfindings could be due to one or more moderator vari-
ables influe ncing re lations betwe e n job characte ristics and outcomes (eg
personality characte ristics or workplace social support) A number of re-
search studies have started in order to investigate these moderating e ffects
some of them with quite promising re sults (see Johnson 1989 Jones amp
Fletcher 1996 Parkes 1991 De Rijk Le Blanc Schaufeli amp De Jonge
1998)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 97
Finally an alternative explanation for the lack of consistency may be
the assessme nt of job characte ristics Basically the JD-C Model is situ-
ation-centered in that it emphasize s the particular role of the work envi-
ronme nt in the prediction of job-related strain It is self-evide nt that in
case s whe re the aim is to rede sign work indepe nde ntly of a particular
worker one nee ds more objective parame ters which are able to show to
have an impact on peoplersquos behaviour (Frese amp Z apf 1994) In other words
job de mands and job control are in the ory characteristics of jobs rather
than characte ristics of people
Job de mands and job control ge nerally have bee n measured by means
of se lf-report que stionnaires Howeve r the main problem with such ques-
tionnaire instrume nts is thatmdashas re flections of the objective work e nviron-
mentmdashthey are more prone to bias than objective instruments Se lf-report
que stionnaires me asure job characte ristics as perceive d by the individual
worker and may the re fore not reflect the objective task accurate ly (Karase k
amp Theore ll 1990) For instance someone can work in a job with high de-
mands and not report feelings of high workload
Consequently the me thodology of the JD -C studies (the multi-occu-
pation studie s in particular) has varied considerably Most studies re ly heav-
ily on the individualrsquos perception or description of job characte ristics (cf
Soumlde rfeldt Soumlderfeldt Jones O rsquoCampo Muntaner O hlson amp Warg 1996)
A few have highlighted the grouprsquos perception or description as a re flection
of the objective environment (cf De Jonge amp Kompier 1997 Schnall e t
al 1994) This point will be discussed in more detail in the next section
THE ASSESSMENT OF JOB CHARACTE RISTICS
The job characteristics of the JD-C Model have bee n measured in two
different ways (1) ldquoobjective rdquo and (2) ldquosubjectiverdquo (Karase k amp Theorell
1990 Kristensen 1995) ldquoO bjectiverdquo and ldquosubjectiverdquo have been put in quo-
tation marks because they are used inconsistently in the literature on work-
stress So-calle d ldquoobjective rdquo job characte ristics may be de fine d as ones
which are assessed independently of the job incumbent (cf Fre se amp Zapf
1988 1994 Spe ctor 1992) Examples are physical and social characteristics
of the work e nvironment or expe rt ratings Accordingly ldquosubjectiverdquo job
characte ristics are de pendent on employe ersquos cognitive and e motional proc-
essing (like appraisals) and the ir ability in coping (Fre se amp Zapf 1988
1994 Lazarus 1995)
Currently the objective method is carried out through (1) direct in-
depe nde nt me asurement and (2) observe rsrsquo ratings (e g Frese amp Zapf
1988 Kristensen 1995 Schnall e t al 1994 Theorell amp Karasek 1996)
The main problem with direct measurement of objective characteristics is
98 de Jon ge van Breukelen Landeweerd and Nijhu is
that some aspe cts are difficult to e xpress in concre te physical te rms (Warr
1987) For e xample job autonomy and job clarity do not provide usable
ge neral marke rs The second approach seems to be objective as me ntioned
above but may be influenced by observersrsquo bias The measurement yie lds
incomple te and partially invalid information due to limited observation
time or space and the e ffects of the observation itse lf (Fre se amp Z apf 1988)
Additionally obse rve rsrsquo ratings see m to suffer from stronge r halo and
stereotyping effects than subjectsrsquo assessments (Frese 1985 Semmer Zapf
amp Gre if 1996 Spector Brannick amp Coove rt 1989)
The subjective method is usually carried out by means of se lf-report
questionnaire s This kind of me asureme nt has several problems as we ll
(e g Frese 1985 Frese amp Z apf 1988 Kasl 1989 1996 Spector 1992
Zapf 1989) such as (1) conceptual overlap betwe en indepe ndent and de-
pendent me asures in that both me asures reflect the same construct (2)
common me thod variance because the information is derive d from the
same source (eg central tendency acquiescence) (3) influence of a third
variable that causes a spurious re lationship (eg a pe rsonal trait) (4) the
potential influencemdashor alterationmdashof the e stimation of job characteristics
due to the presence of an outcome variable (e g health complaints) (5)
possible e ffect of demand characte ristics of the re search context and e x-
perimenter e ffects re sulting in false correlations betwe en job characteristics
and outcome variable s (6) job incumbents may be so used to the ir work
situation that they deny some of the occupational hazards
Spector (1992) argued that more work nee ds to be done e xploring the
accuracy of reports both from the perspective of the job incumbent and
from alte rnative sources to reflect the work conditions O ne way to avoid
the above -me ntioned problems as far as possible is to look for such alte r-
native sources So-calle d group assessm ents seem to be ve ry useful in meas-
uring job characte ristics (Frese 1985 1989 Fre se amp Zapf 1988 Spector
1992) By group assessments we mean that the scores of job incumbents
with the same job and working in ne arly identical workplaces are aggre -
gate d into one ge neral score The group assessments can be described as
the group e stimate s of the re spective job characte ristic for each job incum-
bent These me asures re fe r to that part of a particular job characteristic
that different workers doing ne arly the same job have in common (see also
Semme r e t al 1996) In other words group asse ssments are base d on the
concept of the ideal typical worker ( ie an ave rage worker with sufficient
skills to perform his or he r tasks) According to Fre se and Zapf (1988
1994) group assessments are more objective measures in the sense that
the influence of idiosyncraticmdashindividualmdashperceptions and possibly illusory
answers are reduced In addition to this the e xpertise of workers is taken
into account and problems of brie f pe riods of observation are avoide d Fi-
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 99
nally group assessments see m to be less subject to methodological prob-
le ms such as attenuation because unsystematic error variance is reduced
Thus group data are likely to be more reliable than individual assessme nts
Theoretical justification for considering group assessments as more ob-
je ctive me asures of the two job characte ristics in the JD-C Model can be
found in the literature on social and inte ractional psychology industrial
and organizational psychology and organizational behaviour and e cology
(eg Alderfer 1987 Baron amp Boudreau 1987 Gibson 1979 Hackman
1976 Kraheacute 1992 Schneide r 1987 Warr 1987) From the point of view
of modern interactionism for instance there e xists situational features that
can be consensually regarde d as part of the situation These features are
conceptualized as situational ldquoaffordance srdquo and are constructs that imply
the complimentarity of both work e nvironments and workers What makes
the features ldquoquasi-objectiverdquo is the fact that they be long to the work situ-
ation irrespective of whe the r or not the job incumbent recognizes them as
such For e xample a danger signal te lls a majority of workers to run to
the e me rge ncy e xit Affordance s have some conce ptual similarity with
Hackman rsquos (1976) ldquogroup-supplied stimulirdquoAnother justification is provide d by Schne ide rrsquos (1987) ASA frame-
work This framework sugge sts that employe es will expe rience similar work-
ing conditions due to attraction se lection and attrition processes in an
organization This re sults in similarity in behaviour within a work se tting
In othe r words the pe ople make the workplace (Schneide r 1987) Al-
though Schneider bases his framework primarily at the organizational leve l
it is most likely that his ideas can be e mployed at the group le ve l of analysis
The claim that group assessment ( ie aggre gated data) is more ob-
je ctive is corroborate d by se ve ral e mpirical findings For instance a
me taanalysis of 16 conve rge nce studie s conducte d by Spe ctor (1992)
showe d that aggre gate -leve l corre lations betwee n job characte ristics and
outcome s were similar to individual le ve l correlations Moreover it ap-
peared that the converge nt validity at the aggre gate d le ve l was rather large
and e ven large r than at the individual leve l
PURPOSE
The purpose of this study was to test the JD-C Model using both group
and individual indicators of job characteristics The reason why we used
group and individual assessments of job demands and job autonomy is that
we want to find out whether aggre gated job characte ristics data significantly
add explained variance to individual job characteristics data with regard to
employee he alth In othe r words the key question is whether aggre gated
job characteristics e xplain additional variance in individualsrsquo attitudinal re-
100 de Jon ge van Breukelen Landeweerd and Nijhu is
actions This could mean that some of the explained variance of employe e
health is attributable to some more objective features outside the individual
pe rce ption or asse ssme nt (e g affordance s or group-supplie d stimuli)
which has important theoretical and practical conseque nces For instance
if job-relate d strain captures group le ve l variance the aspirations of the
JD -C Model to be situation-centered would rece ive support In addition
to all this we tried to measure the two job characte ristics (job demands
and job autonomy) more precise ly and e xte nsive ly Lastly the inte raction
effect of demands and autonomy was tested with regre ssion techniques in-
cluding a multiplicative interaction term (cf Aiken amp West 1991 Jaccard
Turrisi amp Wan 1990 Landsbergis et al 1994)
We tested the JD-C Model in a single -occupation sample of health care
workers because the ir professions are ve ry suitable for testing the JD-C
Model for several reasons (e g Fox Dwye r amp Ganster 1993 Ganster amp
Fusilier 1989) First health care workers seem to be subject to stressful work
conditions (ie high job demands) Second job autonomy is an important
tool in present-day care de livery systems Third because of different types of
health care areas and different specialties he alth care workers are a re lative ly
hete rogeneous group Moreover Ganster and Fusilie r (1989) showed that
there is as much variability in health care sample s as there is across a fairly
wide range of occupations Finally the relative homogeneity in social class
restricts the confounding e ffect of socioeconomic status
In line with the JD-C Model it is hypothesized that job demands and
job autonomy have an interaction effect with respect to e mployee he alth
To be more specific we e xpe ct that job autonomy attenuates the adve rse
effects of job demands on employe e health In addition to this we will
compare aggre gated and individual data by means of the variance to be
explaine d We e xpect that give n that aggre gated data correspond to the
more objective work conditions then if the JD-C Model is correct the
use of aggre gated data should reve al stronger interaction e ffects
METHOD
Sam ple
A random sample of 16 institutions was drawn from all gene ral hos-
pitals and nursing home s in the Netherlands (N = 218) Eight general hos-
pitals and e ight nursing home s participated in the study He alth care
workers in four units in each institute we re asked to comple te a que stion-
naire Six types of units were present in the sample an intensive care unit
(ICU) a psychiatric unit an inte rnal unit a surgical unit a somatic unit
and finally a psychogeriatric unit The initial sample consisted of 1806
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 101
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
Finally an alternative explanation for the lack of consistency may be
the assessme nt of job characte ristics Basically the JD-C Model is situ-
ation-centered in that it emphasize s the particular role of the work envi-
ronme nt in the prediction of job-related strain It is self-evide nt that in
case s whe re the aim is to rede sign work indepe nde ntly of a particular
worker one nee ds more objective parame ters which are able to show to
have an impact on peoplersquos behaviour (Frese amp Z apf 1994) In other words
job de mands and job control are in the ory characteristics of jobs rather
than characte ristics of people
Job de mands and job control ge nerally have bee n measured by means
of se lf-report que stionnaires Howeve r the main problem with such ques-
tionnaire instrume nts is thatmdashas re flections of the objective work e nviron-
mentmdashthey are more prone to bias than objective instruments Se lf-report
que stionnaires me asure job characte ristics as perceive d by the individual
worker and may the re fore not reflect the objective task accurate ly (Karase k
amp Theore ll 1990) For instance someone can work in a job with high de-
mands and not report feelings of high workload
Consequently the me thodology of the JD -C studies (the multi-occu-
pation studie s in particular) has varied considerably Most studies re ly heav-
ily on the individualrsquos perception or description of job characte ristics (cf
Soumlde rfeldt Soumlderfeldt Jones O rsquoCampo Muntaner O hlson amp Warg 1996)
A few have highlighted the grouprsquos perception or description as a re flection
of the objective environment (cf De Jonge amp Kompier 1997 Schnall e t
al 1994) This point will be discussed in more detail in the next section
THE ASSESSMENT OF JOB CHARACTE RISTICS
The job characteristics of the JD-C Model have bee n measured in two
different ways (1) ldquoobjective rdquo and (2) ldquosubjectiverdquo (Karase k amp Theorell
1990 Kristensen 1995) ldquoO bjectiverdquo and ldquosubjectiverdquo have been put in quo-
tation marks because they are used inconsistently in the literature on work-
stress So-calle d ldquoobjective rdquo job characte ristics may be de fine d as ones
which are assessed independently of the job incumbent (cf Fre se amp Zapf
1988 1994 Spe ctor 1992) Examples are physical and social characteristics
of the work e nvironment or expe rt ratings Accordingly ldquosubjectiverdquo job
characte ristics are de pendent on employe ersquos cognitive and e motional proc-
essing (like appraisals) and the ir ability in coping (Fre se amp Zapf 1988
1994 Lazarus 1995)
Currently the objective method is carried out through (1) direct in-
depe nde nt me asurement and (2) observe rsrsquo ratings (e g Frese amp Zapf
1988 Kristensen 1995 Schnall e t al 1994 Theorell amp Karasek 1996)
The main problem with direct measurement of objective characteristics is
98 de Jon ge van Breukelen Landeweerd and Nijhu is
that some aspe cts are difficult to e xpress in concre te physical te rms (Warr
1987) For e xample job autonomy and job clarity do not provide usable
ge neral marke rs The second approach seems to be objective as me ntioned
above but may be influenced by observersrsquo bias The measurement yie lds
incomple te and partially invalid information due to limited observation
time or space and the e ffects of the observation itse lf (Fre se amp Z apf 1988)
Additionally obse rve rsrsquo ratings see m to suffer from stronge r halo and
stereotyping effects than subjectsrsquo assessments (Frese 1985 Semmer Zapf
amp Gre if 1996 Spector Brannick amp Coove rt 1989)
The subjective method is usually carried out by means of se lf-report
questionnaire s This kind of me asureme nt has several problems as we ll
(e g Frese 1985 Frese amp Z apf 1988 Kasl 1989 1996 Spector 1992
Zapf 1989) such as (1) conceptual overlap betwe en indepe ndent and de-
pendent me asures in that both me asures reflect the same construct (2)
common me thod variance because the information is derive d from the
same source (eg central tendency acquiescence) (3) influence of a third
variable that causes a spurious re lationship (eg a pe rsonal trait) (4) the
potential influencemdashor alterationmdashof the e stimation of job characteristics
due to the presence of an outcome variable (e g health complaints) (5)
possible e ffect of demand characte ristics of the re search context and e x-
perimenter e ffects re sulting in false correlations betwe en job characteristics
and outcome variable s (6) job incumbents may be so used to the ir work
situation that they deny some of the occupational hazards
Spector (1992) argued that more work nee ds to be done e xploring the
accuracy of reports both from the perspective of the job incumbent and
from alte rnative sources to reflect the work conditions O ne way to avoid
the above -me ntioned problems as far as possible is to look for such alte r-
native sources So-calle d group assessm ents seem to be ve ry useful in meas-
uring job characte ristics (Frese 1985 1989 Fre se amp Zapf 1988 Spector
1992) By group assessments we mean that the scores of job incumbents
with the same job and working in ne arly identical workplaces are aggre -
gate d into one ge neral score The group assessments can be described as
the group e stimate s of the re spective job characte ristic for each job incum-
bent These me asures re fe r to that part of a particular job characteristic
that different workers doing ne arly the same job have in common (see also
Semme r e t al 1996) In other words group asse ssments are base d on the
concept of the ideal typical worker ( ie an ave rage worker with sufficient
skills to perform his or he r tasks) According to Fre se and Zapf (1988
1994) group assessments are more objective measures in the sense that
the influence of idiosyncraticmdashindividualmdashperceptions and possibly illusory
answers are reduced In addition to this the e xpertise of workers is taken
into account and problems of brie f pe riods of observation are avoide d Fi-
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 99
nally group assessments see m to be less subject to methodological prob-
le ms such as attenuation because unsystematic error variance is reduced
Thus group data are likely to be more reliable than individual assessme nts
Theoretical justification for considering group assessments as more ob-
je ctive me asures of the two job characte ristics in the JD-C Model can be
found in the literature on social and inte ractional psychology industrial
and organizational psychology and organizational behaviour and e cology
(eg Alderfer 1987 Baron amp Boudreau 1987 Gibson 1979 Hackman
1976 Kraheacute 1992 Schneide r 1987 Warr 1987) From the point of view
of modern interactionism for instance there e xists situational features that
can be consensually regarde d as part of the situation These features are
conceptualized as situational ldquoaffordance srdquo and are constructs that imply
the complimentarity of both work e nvironments and workers What makes
the features ldquoquasi-objectiverdquo is the fact that they be long to the work situ-
ation irrespective of whe the r or not the job incumbent recognizes them as
such For e xample a danger signal te lls a majority of workers to run to
the e me rge ncy e xit Affordance s have some conce ptual similarity with
Hackman rsquos (1976) ldquogroup-supplied stimulirdquoAnother justification is provide d by Schne ide rrsquos (1987) ASA frame-
work This framework sugge sts that employe es will expe rience similar work-
ing conditions due to attraction se lection and attrition processes in an
organization This re sults in similarity in behaviour within a work se tting
In othe r words the pe ople make the workplace (Schneide r 1987) Al-
though Schneider bases his framework primarily at the organizational leve l
it is most likely that his ideas can be e mployed at the group le ve l of analysis
The claim that group assessment ( ie aggre gated data) is more ob-
je ctive is corroborate d by se ve ral e mpirical findings For instance a
me taanalysis of 16 conve rge nce studie s conducte d by Spe ctor (1992)
showe d that aggre gate -leve l corre lations betwee n job characte ristics and
outcome s were similar to individual le ve l correlations Moreover it ap-
peared that the converge nt validity at the aggre gate d le ve l was rather large
and e ven large r than at the individual leve l
PURPOSE
The purpose of this study was to test the JD-C Model using both group
and individual indicators of job characteristics The reason why we used
group and individual assessments of job demands and job autonomy is that
we want to find out whether aggre gated job characte ristics data significantly
add explained variance to individual job characteristics data with regard to
employee he alth In othe r words the key question is whether aggre gated
job characteristics e xplain additional variance in individualsrsquo attitudinal re-
100 de Jon ge van Breukelen Landeweerd and Nijhu is
actions This could mean that some of the explained variance of employe e
health is attributable to some more objective features outside the individual
pe rce ption or asse ssme nt (e g affordance s or group-supplie d stimuli)
which has important theoretical and practical conseque nces For instance
if job-relate d strain captures group le ve l variance the aspirations of the
JD -C Model to be situation-centered would rece ive support In addition
to all this we tried to measure the two job characte ristics (job demands
and job autonomy) more precise ly and e xte nsive ly Lastly the inte raction
effect of demands and autonomy was tested with regre ssion techniques in-
cluding a multiplicative interaction term (cf Aiken amp West 1991 Jaccard
Turrisi amp Wan 1990 Landsbergis et al 1994)
We tested the JD-C Model in a single -occupation sample of health care
workers because the ir professions are ve ry suitable for testing the JD-C
Model for several reasons (e g Fox Dwye r amp Ganster 1993 Ganster amp
Fusilier 1989) First health care workers seem to be subject to stressful work
conditions (ie high job demands) Second job autonomy is an important
tool in present-day care de livery systems Third because of different types of
health care areas and different specialties he alth care workers are a re lative ly
hete rogeneous group Moreover Ganster and Fusilie r (1989) showed that
there is as much variability in health care sample s as there is across a fairly
wide range of occupations Finally the relative homogeneity in social class
restricts the confounding e ffect of socioeconomic status
In line with the JD-C Model it is hypothesized that job demands and
job autonomy have an interaction effect with respect to e mployee he alth
To be more specific we e xpe ct that job autonomy attenuates the adve rse
effects of job demands on employe e health In addition to this we will
compare aggre gated and individual data by means of the variance to be
explaine d We e xpect that give n that aggre gated data correspond to the
more objective work conditions then if the JD-C Model is correct the
use of aggre gated data should reve al stronger interaction e ffects
METHOD
Sam ple
A random sample of 16 institutions was drawn from all gene ral hos-
pitals and nursing home s in the Netherlands (N = 218) Eight general hos-
pitals and e ight nursing home s participated in the study He alth care
workers in four units in each institute we re asked to comple te a que stion-
naire Six types of units were present in the sample an intensive care unit
(ICU) a psychiatric unit an inte rnal unit a surgical unit a somatic unit
and finally a psychogeriatric unit The initial sample consisted of 1806
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 101
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
that some aspe cts are difficult to e xpress in concre te physical te rms (Warr
1987) For e xample job autonomy and job clarity do not provide usable
ge neral marke rs The second approach seems to be objective as me ntioned
above but may be influenced by observersrsquo bias The measurement yie lds
incomple te and partially invalid information due to limited observation
time or space and the e ffects of the observation itse lf (Fre se amp Z apf 1988)
Additionally obse rve rsrsquo ratings see m to suffer from stronge r halo and
stereotyping effects than subjectsrsquo assessments (Frese 1985 Semmer Zapf
amp Gre if 1996 Spector Brannick amp Coove rt 1989)
The subjective method is usually carried out by means of se lf-report
questionnaire s This kind of me asureme nt has several problems as we ll
(e g Frese 1985 Frese amp Z apf 1988 Kasl 1989 1996 Spector 1992
Zapf 1989) such as (1) conceptual overlap betwe en indepe ndent and de-
pendent me asures in that both me asures reflect the same construct (2)
common me thod variance because the information is derive d from the
same source (eg central tendency acquiescence) (3) influence of a third
variable that causes a spurious re lationship (eg a pe rsonal trait) (4) the
potential influencemdashor alterationmdashof the e stimation of job characteristics
due to the presence of an outcome variable (e g health complaints) (5)
possible e ffect of demand characte ristics of the re search context and e x-
perimenter e ffects re sulting in false correlations betwe en job characteristics
and outcome variable s (6) job incumbents may be so used to the ir work
situation that they deny some of the occupational hazards
Spector (1992) argued that more work nee ds to be done e xploring the
accuracy of reports both from the perspective of the job incumbent and
from alte rnative sources to reflect the work conditions O ne way to avoid
the above -me ntioned problems as far as possible is to look for such alte r-
native sources So-calle d group assessm ents seem to be ve ry useful in meas-
uring job characte ristics (Frese 1985 1989 Fre se amp Zapf 1988 Spector
1992) By group assessments we mean that the scores of job incumbents
with the same job and working in ne arly identical workplaces are aggre -
gate d into one ge neral score The group assessments can be described as
the group e stimate s of the re spective job characte ristic for each job incum-
bent These me asures re fe r to that part of a particular job characteristic
that different workers doing ne arly the same job have in common (see also
Semme r e t al 1996) In other words group asse ssments are base d on the
concept of the ideal typical worker ( ie an ave rage worker with sufficient
skills to perform his or he r tasks) According to Fre se and Zapf (1988
1994) group assessments are more objective measures in the sense that
the influence of idiosyncraticmdashindividualmdashperceptions and possibly illusory
answers are reduced In addition to this the e xpertise of workers is taken
into account and problems of brie f pe riods of observation are avoide d Fi-
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 99
nally group assessments see m to be less subject to methodological prob-
le ms such as attenuation because unsystematic error variance is reduced
Thus group data are likely to be more reliable than individual assessme nts
Theoretical justification for considering group assessments as more ob-
je ctive me asures of the two job characte ristics in the JD-C Model can be
found in the literature on social and inte ractional psychology industrial
and organizational psychology and organizational behaviour and e cology
(eg Alderfer 1987 Baron amp Boudreau 1987 Gibson 1979 Hackman
1976 Kraheacute 1992 Schneide r 1987 Warr 1987) From the point of view
of modern interactionism for instance there e xists situational features that
can be consensually regarde d as part of the situation These features are
conceptualized as situational ldquoaffordance srdquo and are constructs that imply
the complimentarity of both work e nvironments and workers What makes
the features ldquoquasi-objectiverdquo is the fact that they be long to the work situ-
ation irrespective of whe the r or not the job incumbent recognizes them as
such For e xample a danger signal te lls a majority of workers to run to
the e me rge ncy e xit Affordance s have some conce ptual similarity with
Hackman rsquos (1976) ldquogroup-supplied stimulirdquoAnother justification is provide d by Schne ide rrsquos (1987) ASA frame-
work This framework sugge sts that employe es will expe rience similar work-
ing conditions due to attraction se lection and attrition processes in an
organization This re sults in similarity in behaviour within a work se tting
In othe r words the pe ople make the workplace (Schneide r 1987) Al-
though Schneider bases his framework primarily at the organizational leve l
it is most likely that his ideas can be e mployed at the group le ve l of analysis
The claim that group assessment ( ie aggre gated data) is more ob-
je ctive is corroborate d by se ve ral e mpirical findings For instance a
me taanalysis of 16 conve rge nce studie s conducte d by Spe ctor (1992)
showe d that aggre gate -leve l corre lations betwee n job characte ristics and
outcome s were similar to individual le ve l correlations Moreover it ap-
peared that the converge nt validity at the aggre gate d le ve l was rather large
and e ven large r than at the individual leve l
PURPOSE
The purpose of this study was to test the JD-C Model using both group
and individual indicators of job characteristics The reason why we used
group and individual assessments of job demands and job autonomy is that
we want to find out whether aggre gated job characte ristics data significantly
add explained variance to individual job characteristics data with regard to
employee he alth In othe r words the key question is whether aggre gated
job characteristics e xplain additional variance in individualsrsquo attitudinal re-
100 de Jon ge van Breukelen Landeweerd and Nijhu is
actions This could mean that some of the explained variance of employe e
health is attributable to some more objective features outside the individual
pe rce ption or asse ssme nt (e g affordance s or group-supplie d stimuli)
which has important theoretical and practical conseque nces For instance
if job-relate d strain captures group le ve l variance the aspirations of the
JD -C Model to be situation-centered would rece ive support In addition
to all this we tried to measure the two job characte ristics (job demands
and job autonomy) more precise ly and e xte nsive ly Lastly the inte raction
effect of demands and autonomy was tested with regre ssion techniques in-
cluding a multiplicative interaction term (cf Aiken amp West 1991 Jaccard
Turrisi amp Wan 1990 Landsbergis et al 1994)
We tested the JD-C Model in a single -occupation sample of health care
workers because the ir professions are ve ry suitable for testing the JD-C
Model for several reasons (e g Fox Dwye r amp Ganster 1993 Ganster amp
Fusilier 1989) First health care workers seem to be subject to stressful work
conditions (ie high job demands) Second job autonomy is an important
tool in present-day care de livery systems Third because of different types of
health care areas and different specialties he alth care workers are a re lative ly
hete rogeneous group Moreover Ganster and Fusilie r (1989) showed that
there is as much variability in health care sample s as there is across a fairly
wide range of occupations Finally the relative homogeneity in social class
restricts the confounding e ffect of socioeconomic status
In line with the JD-C Model it is hypothesized that job demands and
job autonomy have an interaction effect with respect to e mployee he alth
To be more specific we e xpe ct that job autonomy attenuates the adve rse
effects of job demands on employe e health In addition to this we will
compare aggre gated and individual data by means of the variance to be
explaine d We e xpect that give n that aggre gated data correspond to the
more objective work conditions then if the JD-C Model is correct the
use of aggre gated data should reve al stronger interaction e ffects
METHOD
Sam ple
A random sample of 16 institutions was drawn from all gene ral hos-
pitals and nursing home s in the Netherlands (N = 218) Eight general hos-
pitals and e ight nursing home s participated in the study He alth care
workers in four units in each institute we re asked to comple te a que stion-
naire Six types of units were present in the sample an intensive care unit
(ICU) a psychiatric unit an inte rnal unit a surgical unit a somatic unit
and finally a psychogeriatric unit The initial sample consisted of 1806
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 101
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
nally group assessments see m to be less subject to methodological prob-
le ms such as attenuation because unsystematic error variance is reduced
Thus group data are likely to be more reliable than individual assessme nts
Theoretical justification for considering group assessments as more ob-
je ctive me asures of the two job characte ristics in the JD-C Model can be
found in the literature on social and inte ractional psychology industrial
and organizational psychology and organizational behaviour and e cology
(eg Alderfer 1987 Baron amp Boudreau 1987 Gibson 1979 Hackman
1976 Kraheacute 1992 Schneide r 1987 Warr 1987) From the point of view
of modern interactionism for instance there e xists situational features that
can be consensually regarde d as part of the situation These features are
conceptualized as situational ldquoaffordance srdquo and are constructs that imply
the complimentarity of both work e nvironments and workers What makes
the features ldquoquasi-objectiverdquo is the fact that they be long to the work situ-
ation irrespective of whe the r or not the job incumbent recognizes them as
such For e xample a danger signal te lls a majority of workers to run to
the e me rge ncy e xit Affordance s have some conce ptual similarity with
Hackman rsquos (1976) ldquogroup-supplied stimulirdquoAnother justification is provide d by Schne ide rrsquos (1987) ASA frame-
work This framework sugge sts that employe es will expe rience similar work-
ing conditions due to attraction se lection and attrition processes in an
organization This re sults in similarity in behaviour within a work se tting
In othe r words the pe ople make the workplace (Schneide r 1987) Al-
though Schneider bases his framework primarily at the organizational leve l
it is most likely that his ideas can be e mployed at the group le ve l of analysis
The claim that group assessment ( ie aggre gated data) is more ob-
je ctive is corroborate d by se ve ral e mpirical findings For instance a
me taanalysis of 16 conve rge nce studie s conducte d by Spe ctor (1992)
showe d that aggre gate -leve l corre lations betwee n job characte ristics and
outcome s were similar to individual le ve l correlations Moreover it ap-
peared that the converge nt validity at the aggre gate d le ve l was rather large
and e ven large r than at the individual leve l
PURPOSE
The purpose of this study was to test the JD-C Model using both group
and individual indicators of job characteristics The reason why we used
group and individual assessments of job demands and job autonomy is that
we want to find out whether aggre gated job characte ristics data significantly
add explained variance to individual job characteristics data with regard to
employee he alth In othe r words the key question is whether aggre gated
job characteristics e xplain additional variance in individualsrsquo attitudinal re-
100 de Jon ge van Breukelen Landeweerd and Nijhu is
actions This could mean that some of the explained variance of employe e
health is attributable to some more objective features outside the individual
pe rce ption or asse ssme nt (e g affordance s or group-supplie d stimuli)
which has important theoretical and practical conseque nces For instance
if job-relate d strain captures group le ve l variance the aspirations of the
JD -C Model to be situation-centered would rece ive support In addition
to all this we tried to measure the two job characte ristics (job demands
and job autonomy) more precise ly and e xte nsive ly Lastly the inte raction
effect of demands and autonomy was tested with regre ssion techniques in-
cluding a multiplicative interaction term (cf Aiken amp West 1991 Jaccard
Turrisi amp Wan 1990 Landsbergis et al 1994)
We tested the JD-C Model in a single -occupation sample of health care
workers because the ir professions are ve ry suitable for testing the JD-C
Model for several reasons (e g Fox Dwye r amp Ganster 1993 Ganster amp
Fusilier 1989) First health care workers seem to be subject to stressful work
conditions (ie high job demands) Second job autonomy is an important
tool in present-day care de livery systems Third because of different types of
health care areas and different specialties he alth care workers are a re lative ly
hete rogeneous group Moreover Ganster and Fusilie r (1989) showed that
there is as much variability in health care sample s as there is across a fairly
wide range of occupations Finally the relative homogeneity in social class
restricts the confounding e ffect of socioeconomic status
In line with the JD-C Model it is hypothesized that job demands and
job autonomy have an interaction effect with respect to e mployee he alth
To be more specific we e xpe ct that job autonomy attenuates the adve rse
effects of job demands on employe e health In addition to this we will
compare aggre gated and individual data by means of the variance to be
explaine d We e xpect that give n that aggre gated data correspond to the
more objective work conditions then if the JD-C Model is correct the
use of aggre gated data should reve al stronger interaction e ffects
METHOD
Sam ple
A random sample of 16 institutions was drawn from all gene ral hos-
pitals and nursing home s in the Netherlands (N = 218) Eight general hos-
pitals and e ight nursing home s participated in the study He alth care
workers in four units in each institute we re asked to comple te a que stion-
naire Six types of units were present in the sample an intensive care unit
(ICU) a psychiatric unit an inte rnal unit a surgical unit a somatic unit
and finally a psychogeriatric unit The initial sample consisted of 1806
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 101
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
actions This could mean that some of the explained variance of employe e
health is attributable to some more objective features outside the individual
pe rce ption or asse ssme nt (e g affordance s or group-supplie d stimuli)
which has important theoretical and practical conseque nces For instance
if job-relate d strain captures group le ve l variance the aspirations of the
JD -C Model to be situation-centered would rece ive support In addition
to all this we tried to measure the two job characte ristics (job demands
and job autonomy) more precise ly and e xte nsive ly Lastly the inte raction
effect of demands and autonomy was tested with regre ssion techniques in-
cluding a multiplicative interaction term (cf Aiken amp West 1991 Jaccard
Turrisi amp Wan 1990 Landsbergis et al 1994)
We tested the JD-C Model in a single -occupation sample of health care
workers because the ir professions are ve ry suitable for testing the JD-C
Model for several reasons (e g Fox Dwye r amp Ganster 1993 Ganster amp
Fusilier 1989) First health care workers seem to be subject to stressful work
conditions (ie high job demands) Second job autonomy is an important
tool in present-day care de livery systems Third because of different types of
health care areas and different specialties he alth care workers are a re lative ly
hete rogeneous group Moreover Ganster and Fusilie r (1989) showed that
there is as much variability in health care sample s as there is across a fairly
wide range of occupations Finally the relative homogeneity in social class
restricts the confounding e ffect of socioeconomic status
In line with the JD-C Model it is hypothesized that job demands and
job autonomy have an interaction effect with respect to e mployee he alth
To be more specific we e xpe ct that job autonomy attenuates the adve rse
effects of job demands on employe e health In addition to this we will
compare aggre gated and individual data by means of the variance to be
explaine d We e xpect that give n that aggre gated data correspond to the
more objective work conditions then if the JD-C Model is correct the
use of aggre gated data should reve al stronger interaction e ffects
METHOD
Sam ple
A random sample of 16 institutions was drawn from all gene ral hos-
pitals and nursing home s in the Netherlands (N = 218) Eight general hos-
pitals and e ight nursing home s participated in the study He alth care
workers in four units in each institute we re asked to comple te a que stion-
naire Six types of units were present in the sample an intensive care unit
(ICU) a psychiatric unit an inte rnal unit a surgical unit a somatic unit
and finally a psychogeriatric unit The initial sample consisted of 1806
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 101
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
health care workers from 64 units including nurses nursesrsquo aide s student
nurses activity therapists secretarie s and kitchen staff Self-report ques-
tionnaire s were fille d in and 1489 responde nts returne d the completed
que stionnaire by post (82 response ) In order to keep the work situation
roughly constant across individuals within a give n unit we only included
health care workers with the same job leve l name ly registe red nurses (cf
Thomas 1986) Furthermore only workers who had bee n e mployed for
more than 3 months we re included in the final sample in order to e nsure
valid and reliable observations of the work situation (cf Frese amp Zapf
1994 Katz 1978) These two restrictions reduce the sample to 895 regis-
te red nurses E ighty-four percent of the re spondents were wome n and the
age ranged from 19 to 59 years (M = 307 SD = 74) The me an work
expe rience was 108 years (SD = 65)
Measures
Dem ograph ic variables ( ie gende r and age ) functioned as control vari-
ables These variable s may confound the relationships betwe e n job charac-
te ristics and outcome variable s (e g Karase k amp Theorell 1990 Schaufeli
amp Van Die rendonck 1993 Warr 1987) Be cause type of un it may also play
a confounding role we controlle d for this variable too
Job demands and job autonom y are the predictor variable s In accordance
with the theoretical background we have tried to operationalize these con-
structs more precise ly and more comple te ly There fore these two measures
differ somewhat from Karase krsquos measures The two job characteristics include
both individual data and aggre gate d data (ie mean group scores of the 64
units are used) This means that there are two variable s for each job charac-
teristic one variable with individual scores and one variable with aggre gated
individual scores The se aggre gate -leve l variable s combine judgments across
individual jobs thus removing variance due to individual differences and idi-
osyncratic re sponses In order to minimize bias the subjective indicators of
our two job characteristics contain items with a minimum of cognitive proc-
essing In other words these items are precise ly defined and are as neutral as
possible (cf Frese 1989 Frese amp Z apf 1988 Kasl 1987 Wall et al 1996)
Job dem ands were me asured by an eight-item questionnaire (5-point
response scale ranging from 1 = never to 5 = always) This scale was de-
ve loped by De Jonge Lande wee rd and Nijhuis (1993) and is extensive ly
validate d in Dutch samples We used a relative ly wide range of both quali-
tative and quantitative demanding aspects like working under pressure of
time job comple xity working hard and strenuous work The reliability of
this scale (Cronbachrsquos a ) is 85 An example of the items is ldquoIn the unit
where I work work is carrie d out under pressure of timerdquo
102 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
Job auton om y was measured by means of the Maastricht Autonomy Ques-
tionnaire (MAQ De Jonge 1995) which consists of ten Likert items with a
5-point response scale ranging from 1 = very little to 5 = very much The MAQ
measures the workerrsquos opportunity (or freedom) inherent in the job to deter-
mine a varie ty of task dimensions like method of working amount of work
and work goals The reliability of the MAQ is 81 For instance ldquoThe oppor-
tunity that the work offers to determine the method of working yourselfrdquoIn line with the two major predictions of the JD-C Model we used
four outcome variable s ( ie em otional exhaustion job-related an xiety work
m otivation and job satisfaction)
Em otional exhau stion is a compone nt of the Dutch ve rsion of the
Maslach Burnout Inve ntory the MBI-NL (Schaufe li amp Van Dierendonck
1993 1994) Of the three dimensions of burnout emotional exhaustion is
close st to more traditional strain variable s (cf Maslach 1993 Shirom
1989) The scale consists of e ight items scored on a 7-point scale (ranging
from 0 = ne ve r to 6 = always a = 85)
Job-related an xiety was me asure d by me ans of a compone nt of the
D u tc h O r gan iz a tiona l S tre ss Q ue st ionna ir e ( V O S Re ic he amp Va n
Dijkhuizen 1979) asking respondents how the y ge nerally fe lt at work The
scale consists of four items with a response scale ranging from 1 = never
to 4 = always ( a = 78) The items reflect feelings of anxie ty ne rvousness
tenseness and restlessness re spe ctive ly
Work m otivation was measured by five items in which the respondents
were aske d how inte resting stimulating and challenging the ir work was
(De Jonge et al 1993) The questions we re answered on a 5-point scale
(response scale ranging from 1 = strongly disagre e to 5 = fully agre e)
The re liability of this scale is 87 This scale has bee n we ll-validate d in
Dutch samples of health care workers (cf De Jonge 1995)
Job satisfaction was measured by one item ldquoI am satisfied with my present
jobrdquo The question was answered on a 5-point scale ranging from 1 = strongly
disagree to 5 = fully agree Several researchers have shown that a global rating
of overall job satisfaction is an inclusive me asure of ove rall job satisfaction
(e g Scarpello amp Campbell 1983 Wanous Reichers amp Hudy 1997)
Data Analys is
Much behavioural and social research involve s hie rarchical data struc-
ture s Conventional statistical techniques (e g ordinary regre ssion analysis)
ignore this hie rarchy and may there fore le ad to incorrect re sults (Bryk amp
Raudenbush 1992 Hox 1994 Hox amp Kreft 1994)
In the multi-occupation studies regarding the JD-C Model for e xam-
ple occupational grouping was used as a measure of objective differences
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 103
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
betwe e n jobs re le vant to job rede sign (e g Karase k amp The ore ll 1990
Schwartz Pieper amp Karase k 1988) Although this me thod appears to be
a slight improve me nt on the use of observe rsrsquo ratings it is still very con-
servative and too indirect to isolate the two job characteristics (Fre se amp
Zapf 1988 Ganster amp Schaubroeck 1991 Kasl 1989) More specifically
information is lost if data are aggre gate d to mean group scores There is
a gre at deal of imprecision as a re sult of the inability to deal with variability
in job characte ristics within the group (Ganster amp Fusilier 1989 Landsber-
gis Schurman Israel Schnall Hugentobler Cahill amp Baker 1993) Con-
sequently the statistical analysis may lose power (Hox 1994)
Furthe rmore conve ntional statistical techniques le an heavily on the as-
sumption of inde pendence of observations A common problem with the se
techniques is that the statistical dependence among the scores of employe es
within the same group (due to group characteristics not included in the
mode l) is discounted All observations are regarded as independent when
in fact the re is de pe nde nce (Hox 1994 Vancouver Millsap amp Peters
1994) Violation of the assumption of indepe ndence of observations may
cause too small e stimates of the standard errors of conventional statistical
techniques (Bryk amp Raudenbush 1992) This negative bias in turn may
le ad to spurious ldquosignificantrdquo findings
A final problem is that in small groups the group ave rages will have
large standard e rrors Conventional statistical analysis using group means
will be unusable to separate systematic variation from sampling error (Van-
couve r et al 1994)
Recent deve lopments in statistical theory with regard to the estimation
of hierarchical linear models allow us to take the hierarchy in data into
account (Aitkin amp Longford 1986) In this so-called ldquomultileve l re searchrdquothe data structure in the population is hie rarchical and the data are viewed
as a multistage sample from this hierarchical population For e xample in
organizational re search the population consists of organizations units or
groups within these organizations and e mploye e s within these units or
groups With the he lp of multileve l models we can formulate and test hy-
potheses about relationships occurring at different leve ls and e ven across
le ve ls In the present study a three-leve l model is used (cf Bryk amp Raude n-
bush 1992 Chap 8) First the macrole ve l contains a random sample of
16 institutions Second there are 64 units at the mesoleve l Finally the re
are 895 nurses assumed to be randomly sampled pe r unit (microleve l)
The basic hie rarchical regre ssion mode l for a three-leve l datase t can
be formulate d as a general regre ssion equation with a dependent variable
y and independent variable s x1 to xh in which the subscript i re fe rs to the
microle ve l (nurse) j to the mesoleve l (unit) and k to the macrole ve l ( in-
stitution see Eq 1)
104 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
yijk = b 0jk + b 1jkx1ijk + b 2jkx2ijk + + b hjkxhijk + eijk (1)
The subscript h indexe s the indepe ndent variable x and the corresponding
regre ssion parame ter b Equation (1) is the general microleve l mode l In
the multileve l model the intercept ( b 0jk) and regre ssion coefficients ( b s)
of microle ve l predictors (such as ge nde r age and job demands) may vary
across mesoleve l units (units) and across macroleve l units ( institutions)
randomly andor as a linear function of a me sole ve l or macroleve l fixe d
factor (see Bryk amp Raudenbush 1992) In our mode l we assume that the
intercept ( b 0jk) from Eq (1) can be rewritten as presented in Eq (2)
(2)
in which z1 to z5 represent the dummy variable s for type of unit (me soleve l
fixe d factors) u0jk the random effect of the factor unit within type of unit
and v0k the random effect of the factor institution In this way differences
in health and well-being betwe en units and institutions are accommodated
Because of the exploratory nature of this study explicit hypothe se s
about the interaction betwe en our predictor variable s on the one hand
and the random factors unit and institution on the other (so-called cross-
le ve l effects) we re absent There fore we restricted our analyse s to random
variation betwe en units and institutions in the intercept ( b 0jk) only which
represents random (main) e ffects of the factors unit and institution The
other regression coefficients ( b 1 to b h) are assume d to be constant across
units and institutions which implie s noninteraction betwee n fixe d factors
(xs) and random factors (unit and institution)
This results in a multileve l mode l in which individuals are the unit of
analysis and in which the intercept b 0jk may vary at two le ve ls units and
institutions Additionally some me soleve l and macroleve l fixe d factors can
be introduce d to e xplain variability For our purpose we only included
mesole ve l fixed factors that is the two aggre gate d job characteristics and
the ir interaction term
Take n together multileve l analysis has several advantages in compari-
son with conve ntional statistical techniques First data from more than one
hierarchical le ve l can be included in the analysis So with respect to the
JD -C Model we are able to e stimate the relative importance of individual
and group leve l factors Second the statistical dependence betwe e n indi-
viduals of the same unit (or units of the same institution) is taken into
account through the random variation of szlig0jk across units and institutions
Finally the multileve l model separate s sampling error due to variation be-
twe en units from variation within units
b b0 01
5
0 0 0jkq
q q jk ky z u v= + + +=aring
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 105
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
In this study the multileve l mode ls are estimate d and fitted with the
computer program VARCL (VARiance Component analysis by maximum
L ikelihood Longford 1993) VARCL is one of the most popular computer
programs for analyzing multileve l regression models (cf Hox 1994 Kreft
De Leeuw amp Van der Lee den 1994) Additionally a standard statistical pro-
gram (SPSS) was used in order to obtain the two raw data file s with aggre -
gated and individual data respectively to be used in VARCL (cf Hox 1994)
Mod el Bu ild in g
The strategy for mode l building within VARCL is to follow the hy-
pothesis and expe ctations as described e arlier
1 The first model that is fitted is an e mpty mode l a fully uncondi-
tional mode l without predictors at any leve l apart from the ran-
dom effects of units and institutions This model represents the
(unexplained) variation of the outcome variable s at e ach le ve l
(nurse unit and institution) In case of significant unit and insti-
tution effects we will have to perform multileve l analyse s rather
than ordinary line ar regression analyse s
2 The second model includes all covariate s ( ie gender age and
type of unit) and the individual job characteristics How much of
the total variance can be e xplaine d by these variable s
3 The third mode l again includes the variable s of mode l 2 but now
the aggre gate d job characte ristics are adde d The questions are
first whe ther the aggre gated variable s add explained variance to
model 2 and second how much variance can be explained by both
individual and aggre gate d variable s
In order to test the interaction hypothe sis we performed multileve l
regre ssion analyse s including a multiplicative inte raction term for all out-
come variable s The multiplicative te rm was computed from the grand
mean centered scores of job demands and job autonomy for the individual
variable s and aggre gate d variable s re spective ly (cf Aiken amp We st 1991
Kleinbaum Kupper amp Muller 1988)
RESULTS
Prelim inary An alyses
The means standard deviations and zero-order Pearson correlations
of the study variable s are presented in Table I Note that there are two
covariate s at the microle ve l ( ie gende r and age ) and five dummy variable s
106 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
Ta
ble
IM
ea
ns
Sta
nd
ard
De
via
tio
ns
an
dZ
ero
-Ord
er
Pe
ars
on
Co
rre
lati
on
so
fth
eS
tud
yV
ari
ab
les
(n=
89
5)
Va
ria
ble
MS
D1
23
45
67
89
10
11
12
13
14
15
16
1G
en
de
ra
macrmacr
2A
ge
30
70
73
6macr1
4
3In
div
d
em
3
20
56
09
macr0
6
4In
div
a
ut
27
25
5macr1
3
08
macr3
0
5In
div
d
xab
macr09
34
00
macr03
macr07
0
1
6A
gg
rd
em
3
21
32
20
macr0
8
56
macr2
5
macr02
7A
gg
ra
uto
2
69
28
macr20
0
8
macr30
4
8
macr04
macr52
8A
gg
rd
xab
macr04
10
12
macr0
7
01
macr13
3
3
02
macr26
9S
ati
sfa
ctio
n3
90
86
09
macr0
7
macr18
1
3
11
macr1
5
09
1
5
10
Mo
tiva
tio
n3
80
68
macr01
macr10
macr0
9
20
1
0
macr18
1
8
07
4
9
11
Ex
ha
ust
ion
17
48
7macr0
2macr0
24
5
macr13
macr0
52
5
macr08
macr0
9
macr42
macr2
1
12
An
xie
ty1
47
43
macr04
macr01
19
macr0
1macr0
61
0
04
macr11
macr2
5
macr11
4
8
13
Psy
chia
tric
macrmacr
macr26
1
8
macr16
1
9
macr06
macr28
4
1
macr27
macr1
7
macr04
10
1
6
14
Inte
rna
lmacr
macr1
3
macr06
03
08
0
8
06
16
1
4
06
06
macr02
00
macr14
15
Su
rgic
al
macrmacr
10
macr0
30
7
macr02
06
12
macr0
41
5
macr00
06
06
01
macr14
macr1
3
16
So
ma
tic
macrmacr
15
macr0
8
20
macr2
2
macr11
3
5
macr46
macr1
5
macr05
macr13
0
2macr0
4macr2
1
macr20
macr1
9
17
Psy
cho
-ge
rmacr
macr1
5
macr08
macr0
0macr0
60
1macr0
1macr1
3
09
0
5macr1
0
01
macr06
macr21
macr2
0
macr19
macr2
8
aG
en
de
rw
as
cod
ed
0(m
ale
s)a
nd
1(f
em
ale
s)
bG
ran
dm
ea
nce
nte
red
p
lt0
5(t
wo
-ta
ile
d)
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 107
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
at the me soleve l controlling for the six types of units The Intensive Care
Unit (ICU) is the reference cate gory and has the value 0 on all five dummy
variable s At the me soleve l the predictor variable s job demands and job
autonomy consist of aggre gate d individual data (ie me an group score per
unit was computed cf Hox 1994)
To justify aggre gation of the two job characteristics we have to demon-
strate sufficient homogene ity of within-unit variance To test whe ther there
was agree me nt within the 64 units of the ratings of job demands and job
autonomy we used the estimate of within-group interrate r reliability of James
Demaree and Wolf (1993) This coefficient can be interpreted similarly to
other types of reliability coefficients such as coefficient alpha (cf George
1990) In general the ave rage within-group interrater reliability for job de-
mands and job autonomy was 95 and 96 respectively In summary the reli-
ability estimate s for the two job characteristics indicate d a high leve l of
agreement within units justifying the use of aggre gates in subsequent analyse s
Finally a confirmatory factor analysis (LISRE L 8) was conducted to
show that indeed there are four separate outcome variable s (cf Joumlreskog
amp Soumlrbom 1993) The corresponding LISREL analysis showe d that a four-
factor solution yie lde d an acceptable chi-square re lative to its degrees of
freedom ( c 2(129) = 51339 p lt 001) and relative ly good other fit indices
(NNFI = 93 CFI = 94 AGFI = 91 RMSEA = 06)
Mod el Tests
The first mode l within VARCL is an e mpty model with only one fixed
effect name ly the intercept (the ave rage individual me an) and two random
effects of the factors units and institutions Significance of the random e ffects
of units and institutions within VARCL was te sted by computing the deviance
(D) for the ordinary regre ssion model (a regression model without these ran-
dom effects) The difference between this deviance and the deviance of our
multileve l null model has a c 2-distribution with two de gree s of freedom under
H0 such that the re are ne ithe r unit nor institution effects (e g Bryk amp
Raude nbush 1992 Kleinbaum e t al 1988) For all outcome variable s the
results showed that the difference betwe en the two deviances is significant
which me ans that H0 was reje cted We may conclude that there are differ-
ences betwe en units andor institutions with re spect to all outcome variable s
The variance in these variable s is mainly a function of individual differences
(857 macr944 ) but unit and institution differences together explain some of
the variance (56 macr143 ) So multileve l regression analyse s rather than or-
dinary regre ssion analyse s have to be performed
108 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
Em otion al Exhau stion
The results of the multileve l regression analyse s with e motional e xhaus-
tion as outcome variable are give n in Table II In the second mode l of Table
II all e ight covariate s and the two individual job characteristics are entered
Different models can be compare d with respect to predictive power by
a likelihood ratio te st (Bosker amp Snijders 1990 Bryk amp Raude nbush 1992)
De viance (D) is computed for each model and the difference betwe en the
deviance statistics ( D D) is used to test the hypothese s If one model is a spe-
cial reduced version of the other model this difference has a c 2-distribution
under H0 that the exte nded model does not predict be tter than the reduced
mode l Critical values of the c 2-statistic mean that the reduced model is too
simple a description of the data (eg Kle inbaum et al 1988)
Table II Results of the Multileve l Regression Analyses with Respe ct to
E motional Exhaustion (p-Values Based on Approximate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 176 176 178
Individual leve lb
Ge nde r macr10 macr09
Age macr00 macr00
Job demands 73 71
Job autonomy macr04 macr04
Dem acute aut macr02 01
Group levelb
Type of unita
Psychiatric 55 52
Internal 12 11
Surgical 24 23
Som atic 06 macr01
Psycho-ge riatric 20 17
Job demands 10
Job autonomy macr07
Dem acute aut macr65
V ariance decomposition ( s e2)
Individual leve l 680 561 560
Group level 071 011 010
Institution level 013 002 000
Mode l fit
Deviance (D) 225630 204054 203599
D mode l 1 ( D D) 21576
D mode l 2 ( D D) 455
D df 10 3
R2 249 254
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 109
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
After e ntering the above -me ntioned variable s the ove rall model fit im-
proved ( D D(10) = 21576 p lt 05) This implie s that mode l 2 has a better
fit than mode l 1 and reduces unexplained variance at all three le ve ls The
total mode led or explained proportion of variance (R2) is 249 For a
random intercept mode l this parameter can be e stimate d as the propor-
tional reduction in me an squared prediction e rror due to predictor variable s
(see also Snijders amp Bosker 1994)
It appears howe ver that the inte raction term of job demands and job
autonomy is not significant O nly individual job demands have a significant
positive relationship with e motional exhaustion In other words highe r le v-
els of individual job de mands are associate d with higher leve ls of e motional
exhaustion
In our next mode l aggre gate d variable s are entered in order to explain
differences in emotional e xhaustion Moreover model 3 e xamines whether
aggre gated job characte ristics add variance to model 2 Table II shows that
this model does not have a better fit than model 2 ( D D(3) = 455 p =
ns) indicating that the effects of the aggre gate d variable s are modest The
total mode led variance is 254
Job-Related An xiety
Table III presents the results of the multileve l regre ssion analyse s with
job-related anxie ty as outcome variable Model 2 contains the re sults for
the covariate s and the individual job characteristics Entering the se vari-
ables improved model fit ( D D(10) = 6303 p lt 05) and reduced some
unexplained variance at the individual and group le ve l (mode lle d variance
85 ) The table indicate s that the interaction term of job demands and
job autonomy is not significant Similar to e motional exhaustion mode l 2
shows that the individual job demands have a significant positive association
with anxie ty That is highe r le ve ls of individual job de mands are re lated
to highe r leve ls of job-relate d anxie ty
The aggre gated variable s we re entered in the next mode l (model 3)
Compare d with mode l 2 model 3 does not le ad to an improveme nt in
mode l fit ( D D(3) = 607 p = ns) This means that the remaining unex-
plaine d variance in model 2 cannot be explained by the aggre gate d job
characte ristics The total mode le d variance in mode l 3 is 91
Work Motivation
The results of the multileve l regre ssion analyse s with work motivation
are presented in Table IV Entering the covariates and individual variable s
in mode l 2 improve d mode l fit ( D D(10) = 5891 p lt 05) So the se vari-
110 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
ables contribute significantly to the explanation of work motivation at all
three leve ls the entire mode le d variance is 106
Model 3 provide s a significant improve ment in mode l fit compare d
with model 2 ( D D(3) = 1171 p lt 05) The aggre gated variable s are able
to e xplain some variance that cannot be e xplaine d by the individual vari-
ables The total modele d variance is 128 Finally this mode l shows a
significant positive interaction effect at the individual leve l Added to this
the aggre gated job demands have a significant negative relationship with
work motivation In othe r words higher le ve ls of aggre gate d job demands
are associate d with lower leve ls of work motivation
The technique for examining the interaction betwe en job demands and
job autonomy is plotting the e quations (cf Aiken amp We st 1991) Following
Table III Results of the Multile vel Regression Analyse s with Respe ct to
Job Re late d Anxiety (p-V alues Based on Approxim ate Standard Errors
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 147 154 155
Individual leve lb
Ge nde r macr01 macr01
Age macr00 macr00
Job demands 18 15
Job autonomy 01 00
Dem acute aut macr05 macr03
Group levelb
Type of unita
Psychiatric 23 22
Internal 00 macr01
Surgical 00 macr01
Som atic macr06 macr09
Psycho-ge riatric macr04 macr05
Job demands 10
Job autonomy 01
Dem acute aut macr28
V ariance decomposition ( s e2)
Individual leve l 178 173 172
Group level 011 000 000
Institution level 000 000 000
Mode l fit
Deviance (D) 103160 96857 96250
D mode l 1 ( D D) 6303
D mode l 2 ( D D) 607
D df 10 3
R2 86 91
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 111
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
the method of Cohen and Cohen (1983) value s of the predictor variable s
were chosen one standard deviation below and above the me an Simple
regre ssion lines were then generate d by entering these value s in the re-
gre ssion e quation The results of the computations of these simple regre s-
sion equations are give n in Fig 1
The interaction term at individual le ve l with regard to work motiva-
tion shows that job demands and work motivation are slightly positive ly
re lated at high le ve ls of autonomy At the same time howeve r de mands
and motivation are slightly ne gative ly associate d in the case of low leve ls
of autonomy
Table IV Results of the Multilevel Regression Analyses with Respe ct to
W ork Motivation ( p -V alue s B ase d on A pproxim ate Standa rd E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 377 431 425
Individual leve lb
Ge nde r 10 10
Age macr01 macr01
Job demands 02 07
Job autonomy 21 20
Dem acute aut 14 13
Group levelb
Type of unita
Psychiatric macr31 macr32
Internal macr20 macr16
Surgical macr16 macr08
Som atic macr40 macr24
Psycho-ge riatric macr41 macr34
Job demands macr31
Job autonomy 19
Dem acute aut 36
V ariance decomposition ( s e2)
Individual leve l 403 386 385
Group level 046 034 025
Institution level 021 000 000
Mode l fit
Deviance (D) 179680 173789 172618
D mode l 1 ( D D) 5891
D mode l 2 ( D D) 1171
D df 10 3
R2 106 128
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
112 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
Job Satisfaction
Finally Table V shows the results of the multileve l regre ssion analyse s
with job satisfaction First the model fit improved significantly whe n the
covariate s and individual variable s are e ntered in the model ( D D(10) =
6112 p lt 05) The modeled variance is 104 Next model 3 includes
the three aggre gate d variable s and shows a significant improve ment in
mode l fit ( D D(3) = 1214 p lt 05) Mode l 3 reduced unexplaine d variance
particularly at group and institution leve l the total e xplained variance is
132 Lastly this mode l shows a significant positive interaction effect at
the group leve l Job demands and job autonomy at individual le ve l have a
significant negative and positive association with job satisfaction respec-
tive ly That is higher leve ls of individual job demands are relate d to lower
le ve ls of job satisfaction Conve rse ly highe r le ve ls of job autonomy are as-
sociated with highe r leve ls of job satisfaction
Figure 2 shows the graphical representation of the interaction betwe en
job demands and job autonomy at group le ve l with regard to job satisfaction
It appears that job de mands and job satisfaction are positive ly associated in
the case of high leve ls of job autonomy At low le ve ls of job autonomy how-
ever job demands and job satisfaction are ne gative ly relate d
In multileve l literature our aggre gate d variable s are considered to be
contextual effects (Bosker amp Snijde rs 1991 Raudenbush 1989) The classic
formulation of a contextual effect model involve s a regression e quation in-
cluding both the individual variable (s) and the group variable (s) Howe ver
Fig 1 Graphical re pre sentation of the individual level inte raction among
job demands and job autonom y in the pre diction of work motivation
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 113
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
such a model might suffer from high colline arity leading to poor precision
in model fit (Aitkin amp Longford 1986)
O ur mode ls 3 (Tables IImacrV ) are such contextual e ffect models To test
whether the mode ls are subject to high colline arity we carrie d out ordinary
regre ssion analyse s of mode l 3 with all four outcome variable s Collinearity
was checked by means of the Variance Inflation Factor (V IF) A rule of
thumb for evaluating V IFs is that one should be concerned about any value
larger than 100 (Kleinbaum e t al 1988) The V IFs in our analyse s did
not e xcee d 33 which indicates that no seve re collinearity problems are to
be expe cted All in all it can be concluded that our contextual e ffect models
do not suffer from high collinearity
Table V Results of the Multilevel Regression Analyse s with Re spe ct to
Job Satisfaction ( p -V alue s B ase d on A pp ro ximate Standar d E rro rs
Provide d by V ARCL)
Mode l
1 2 3
Grand me an ( b 0) 388 396 393
Individual leve lb
Ge nde r 23 22
Age macr00 macr00
Job demands macr22 macr18
Job autonomy 16 15
Dem acute aut 15 12
Group levelb
Type of unita
Psychiatric macr54 macr53
Internal macr05 macr05
Surgical macr15 macr09
Som atic macr17 04
Psycho-ge riatric macr12 macr03
Job demands macr22
Job autonomy 35
Dem acute aut 112
V ariance decomposition ( s e2)
Individual leve l 628 604 602
Group level 085 035 028
Institution level 024 021 010
Mode l fit
Deviance (D) 219732 213620 212406
D mode l 1 ( D D) 6112
D mode l 2 ( D D) 1214
D df 10 3
R2 104 132
aReference category ICU
bUnstandard ized regression coefficients
p lt 05
114 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
DISCUSSION
The present study contributes to job redesign and job stress re search
by testing the Job Demand-Control Model using a multileve l analytic ap-
proach To be more specific we used both aggre gate d individual data and
individual data as indicators of the job characteristics job demands and job
autonomy In our vie w the aggre gate d data correspond to the more ob-
je ctive work conditions while the individual data correspond to stressor
perceptions or re sults of appraisal processes In interpre ting the findings
of this study it is important to consider not only the confirmation of the
hypothesis and the answe rs on the questions but also the ir re lative signifi-
cance We will brie fly discuss them
First the re sults partly support the interaction hypothesis of the JD-C
Model by finding two interaction effects and both in the expe cted direc-
tion The significant interaction terms represent 25 of the inte ractions
te sted which me ans no strong support for the JD-C Model The support
for the mode l is quite meaningful however because one significant strong
interaction was found at the aggre gate d (ie group) leve l This finding un-
derlines the position of the JD -C Mode l as a situation-centered mode l with
the interplay of two more objective job characteristics
In contrast no significant interaction e ffects were found for the ad-
verse health outcomes Although the interaction terms were in the right
direction the le ve l of 5 significance just could not be reached (p-values
were about 06 and 07) An e xplanation for this unexpected result may be
some lack of power in this kind of outcome variable s and in detecting
interactions at all (cf Aiken amp West 1991)
Fig 2 Graphical re pre sentation of the group level interaction among job de-
mands and job autonom y in the pre diction of job satisfaction
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 115
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
The modeled variance (R2) for the best fitting mode ls range d from
9 for job-relate d anxie ty to 25 for emotional exhaustion Adde d to this
the job characte ristics account for a higher amount of reduced variance in
adve rse he alth outcome s (e xhaustion 9 and anxie ty 5 ) than in the
two other variable s ( ie about 2 for both motivation and satisfaction)
Although these values are not very high they are rather consistent with
those obtaine d by othe r high quality occupational stress studie s ( e g
Karase k 1989 Semme r et al 1996 Warr 1990)
Se cond an important question was whether aggre gated job charac-
te ristics data significantly add explained variance to individual job charac-
te ristics data with respect to employe e health Though the percentages are
not ve ry high the findings indicate that aggre gated job characte ristics data
are important in e xplaining work motivation and job satisfaction and are
able to e xplain some variance that could not be e xplaine d by the individual
data (about 2macr3 ) From the standpoint of me asureme nt one can argue
that there are things which can be me asured be tter by aggre gated leve l
characte ristics This remarkable point has also bee n noted by Z apf (1989)
For instance job incumbents may be so used to their work situation in
such a way that the y de ny some of the occupational hazards In other
words group le ve l characte ristics may tap (a part of) the context in which
individual workers operate These re sults legitimate the claim of the JD-C
Model to be a environme ntal-oriented model as far as work motivation
and job satisfaction are concerned
Conversely the aggre gated job characteristics we re not important in
explaining emotional e xhaustion and job-related anxie ty while addressing
the individual leve l job characteristics Moreove r they did not significantly
add explained variance to individual job characteristics An e xplanation may
be that emotional e xhaustion and job-relate d anxie ty are for the most part
de termine d by stressor perceptions or re sults of appraisal processes (cf
Lazarus 1995) Another explanation may be that other factors (e g social
cues me thod variance ) must be acknowle dged as potential sources (cf
Spector 1992) So base d upon these findings Karasekrsquos JD -C Model seems
to be not only situation-centered but also contains some person-centered
assumptions In summary the current findings sugge st that work motivation
and job satisfaction are more dependent on the group (ie the ave rage
worker) while emotional exhaustion and to a le sser e xtent anxie ty are
more de pendent on the individual
An intere sting e xplanation for the group leve l and individual le ve l e f-
fects may be that both group and individual assessment (partly) reflect the
same features of the work situation which were re fe rred to as affordances
(see introduction) Affordances see m to be important for all workers be-
cause they can be regarded as be ing part of the situation Another the o-
116 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
retical rationale for these e ffects has been provided by an intergroup theory
(cf Alderfer 1987 Schneider 1985) The logic of this theory is that inter-
actions of employe es at any le ve l of analysis represent the e ffects of group
memberships In its most expande d form this perspective emphasizes the
embedded nature of subsystems In othe r words eve ry person can be em-
bedded in a group (an employee in a unit) e very group can be e mbedded
in other groups (a unit in an organization) and so on Applie d to our study
employee health might be a product of multileve l embeddedness of job
characte ristics That is the individual le ve l of analysis is embedded in and
affected by at least the ne xt leve l of analysis So it see ms that the next
larger unit(s) in which individual job characteristics are embedded will also
have an impact at least on work motivation and job satisfaction
All in all the present study supports the choice of the JD-C Model in
job redesign research as well as job stress research It can be concluded that
perceptions of job characteristics do not only reflect subjective feelings but
also are grounded in some kind of e nvironmental reality as far as job satis-
faction and work motivation are concerned Assuming that aggre gated job
characteristics data are more re lated to the objective work e nvironment than
individual data these results suggest a rede sign of work conditions for the
sake of job satisfaction and work motivation and a change of the individual
worker for the sake of e motional exhaustion and job-related anxie ty
Some we aknesses of the present study can be mentioned First we tried
to de fine nearly identical jobs in order to me et the criterion of the same work
situations However our method of sample restriction (ie 895 registered
nurses left) probably le d to a reduction in variance Despite the fact that our
procedure was variance reducing we did find some evide nce for the interac-
tive JD-C Model Second as noted before aggre gated data are more re liable
than individual data and thus le ss affected by attenuation However aggre -
gated data will have this benefit only in case of high unsystematic e rror vari-
ance In case of low error variance ldquorealrdquo differences between individuals
within units will be ignored using the method of aggre gation Third our pre-
sent multileve l analysis is not totally free of problems For e xample it can
only be applied to each outcome variable separately and conseque ntly ig-
nores relationships betwee n them Multivariate e xtensions like multileve l
structural equation mode ling are needed (e g Hox 1994 McArdle amp Ham-
agam i 1996 Mutheacuten 1994) Fourth we used an exploratory procedure to
derive a parsimonious model So there is a possibility that some decisions
we have made are based on chance The current findings have to be cross-
validated with another large and hierarchical sample Finally it is not possible
to dete rmine whether the assumed causal paths are present on the basis of
our cross-sectional data Although JD-C theory guided our hypothesis about
causal relationships hypothesized causal connections should be interpreted
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 117
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
carefully For this purpose longitudinal studies are required (cf Zapf Dor-
mann amp Fre se 1996)
In conclusion this study shows that both group and individual asse ss-
ments of job characteristics are important in predicting e mployee he alth
Practically our re sults support the notion that researchers may nee d to
focus on work conditions ( ie the concept of the ideal typical worker) and
the individual e mployee simultaneously Furthermore the results imply that
previous individual leve l re search should be supple mented by a considera-
tion of aggre gated le ve l e ffects Further re search however is neede d for
a better understanding of the relationships that we re hypothesized and in-
vestigated and for refineme nts of techniques and measurements
REFERENCES
AIKEN L S amp WE ST S G Multiple regression Testing and interpreting interactions Ne wbury
Park CA Sage Publications 1991
AITKIN M amp LO NGFO RD N Statistical mode ling issue s in school e ffectiveness studie s
Journal of the Royal Statistical Society 1986 Series A 149 1-43
ALDERFE R C P An intergrou p perspe ctive on group dynamics In J Lorsch (Ed) Hand-
book of organ izational behavior Englewood Cliffs NJ Prentice-Hall 1987 pp 190-222
BARO N R M amp BO UDREAU L A An ecological pe rspe ctive on inte grating personality
and social psychology Journal of Person ality and Social Psychology 1987 53 1222-1228
BO SKER R J amp SNIJDE RS T A B Statistische aspe cten van multi-niveau onderzoe k
[Statistical aspe cts of multilevel research] Tijdschrift voor Onderwijsresearch 1990 15(5)
317-329
BO SKER R J amp SNIJDE RS T A B Ce nte r forward Unpublished manuscript 1991
BRYK A S amp RAUDENBUSH S W Hierarchical linear m odels Applications and data analy-
sis m ethods Ne wbury Park CA Sage Publications 1992
CO HEN J amp CO HE N P Applied m ultiple regressio ncorrelation analysis for the behavioral
sciences (2nd e d) Hillsdale Lawre nce E rlbaum Associate s 1983
FOX M L DWYER D J amp GANSTER D C E ffe cts of stre ssful job demands and control
on psychologic al and attitudinal outcome s in a hospital se tting Academy of Management
Journal 1993 36(2) 289-318
FRESE M Stre ss at work and psychosomat ic complaints A causal interpre tation Journ al of
Applied Psychology 1985 70(2) 314-328
FRESE M Theore tical mode ls of control and health In S L Saute r J J Hurre ll Jr and
C L Coope r (Eds) Job control and worker health Chiche ster Wile y amp Sons 1989 pp
107-128
FRESE M amp Z APF D Methodologic al issues in the study of work stre ss O bjective vs
subje ctive me asurem ent of work stre ss and the que stion of longitudinal studie s In C L
Coope r and R Payne (Eds) Cau ses coping and consequences of stress at work Chiche ster
Wile y amp Sons 1988 pp 375-411
FRESE M amp Z APF D Action as the core of work psychology A Ge rm an approach In
H C Triandis M D Dunne tte and L M Hough (E ds) Handbook of industrial and
organ izational psychology (Vol 4) Palo Alto CA Consultin g Psychologists Pre ss 1994
pp 271-340
GANSTE R D C Worke r control and well-being A re view of re search in the workplace In
S L Saute r J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health
Chiche ster Wiley amp Sons 1989 pp 3-23
GANSTE R D C Inte rventions for building he althy organizations Sugge stions from the stre ss
rese arch literature In L R Murphy J J Hurre ll Jr S L Saute r and G P Keita
(Eds) Job stress interventions Washington APA 1995 pp 323-336
118 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
GANSTE R D C amp FUSILIE R M R Control in the workplace In C L Coope r and I
T Robertson (Eds) International review of industrial and organ izational psychology Chich-
ester Wile y 1989 pp 235-280
GANSTE R D C amp SCH AUBRO ECK J Work stre ss and e mploye e health Journal of Man-
agem ent 1991 17(2) 235-271
GEO RGE J M Personality affect and be havior in groups Journal of Applied Psychology
1990 75(2) 107-116
GIBSO N J J The ecological approach to visual perception Boston Houghton Mifflin 1979
HACKMAN J R Group influence s on individuals In M D Dunne tte (E d) Handbook of
industrial and organ izational psychology Chicago Rand McNally 1976 pp 1455-1526
HACKMAN J R amp O LDHAM G R Work redesign Reading MA Addison-We sley Pub-
lishing Com pany 1980
HOX J J Applied m ultilevel analysis Amste rdam TT-Publikatie s 1994
HOX J J amp KREFT I G G Multilevel analysis methods Sociological Methods amp Research
1994 22(3) 283-299
JACCARD J TURRISI R amp WANN C K Interaction effects in m ultiple regression New-
bury Park CA Sage Publications 1990
JAMES L R DEMAREE R G amp WO LF G rwg An asse ssm ent of within-group interrate r
agree ment Journal of Applied Psychology 1993 78(2) 306-309
JOHNSO N J V Control colle ctivity and the psychosocial work e nvironm ent In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 56-74
JONES F amp FLE TCHER B(C) Job control and health In M J Schabracq J A M
Winnubst and C L Coope r (Eds) Handbook of work and health psychology Chiche ster
Wile y amp Sons 1996 pp 33-50
JONGE J DE Job autonomy we ll-being and he alth A study among Dutch health care
worke rs PhD thesis University of Limburg Maastricht 1995
JONGE J DE amp KO MPIER M A J A critical e xamination of the Demand-Con trol-Sup-
port Mode l from a work psychological pe rspe ctive International Journal of Stress Man-
agem ent 1997 4(4) 235-258
JONGE J DE LANDEWEERD J A amp NIJHUIS F J N Constructie e n valide ring van
de vragenlijst te n behoe ve van he t proje ct ldquoautonomie in het werkrdquo [Constructio n and
validation of the que stionnaire for the ldquojob autonomy proje ctrdquo] Studies bedrijfsgezond-
heidszorg num m er 9 Maastricht Unive rsity of Limburg 1993
JOumlRESKO G K G amp SOumlRBO M D LISREL 8 Userrsquos reference guide Chicago Scientific
Software Inte rnational 1993
KAHN R L amp BYOSIE RE P Stre ss in organizations In M D Dune tte and L M Hough
(Eds) Handbook of industrial and organ izational psychology (Vol 3) Palo Alto CA Con-
sulting Psychologists Press 1992 pp 571-650
KARASEK R A Jr Job demands job decision latitude and me ntal strain Implications for
job design Adm inistrative Science Quarterly 1979 24 285-308
KARASEK R A Job content instru m ent Questionnaire and userrsquos guide revision 11 Los An-
ge les Unive rsity of Southe rn California 1985
KARASEK R Control in the workplace and its he alth-re lated aspe cts In S L Saute r J J
Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster Wile y
amp Sons 1989 pp 129-159
KARASEK R A amp THEO RELL T Healthy work Stress productivity and the recon stru ction
of working life New York Basic Books 1990
KASL S V Methodologie s in stre ss and he alth Past difficulties pre sent dilemmas future
dire ctions In S V Kasl and C L Coope r (Eds) Stress and health Issu es in research
m ethodology Chiche ster John Wiley amp Sons 1987 pp 307-318
KASL S V An e pidemiological perspe ctive on the role of control in he alth In S L Saute r
J J Hurre ll Jr and C L Coope r (Eds) Job control and worker health Chiche ster
Wile y amp Sons 1989 pp 161-189
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 119
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
KASL S V The influence of the work environm ent on cardiovascular health A historical
conceptual and me thodologic al perspe ctive Journal of Occupational Health Psychology
1996 1(1) 42-56
KATZ R Job longevity as a situational factor in job satisfaction Adm inistrative Science Quar-
terly 1978 23 204-223
KLEINBAUM D G KUPPER L L amp MULLE R K E Applied regressio n analysis and
other m ultivariable m ethods (2nd e d) Boston PWS-KE NT Publishing Com pany 1988
KRAHEacute B Person ality and social psychology Towards a synthesis Ne wbury Park CA Sage
Publications 1992
KREFT I G G LEEUW J DE amp LEE DEN R VAN DER Review of five multilevel
analysis program s BMDP-5V GENMO D HLM ML3 VARCL The Am erican Statisti-
cian 1994 48(4) 324-335
KRISTE NSE N T S The demand-con trol-support mode l Methodological challenges for fu-
ture research Stress Medicine 1995 11 17-26
L A ND SB E R G IS P A S CH NA L L P L WA R R E N K P IC KE R ING T G amp
SCH WARTZ J E Association between ambulatory blood pre ssure and alte rnative for-
mulations of job strain Scandinavian Journal of Work Environm en t amp Health 1994 20
349-363
LANDSB ERGIS P A SCHU RMAN S J ISRAEL B A SCH NALL P L HUGEN-
TO BLER M K CAHILL J amp BAKE R D Job stre ss and he art disease Evidence
and strate gie s for pre vention New Solutions 1993 Summe r 42-58
LAZ ARUS R S Psychological stre ss in the workplace In R Crandall amp P L Perre weacute(Eds) Occupational stress A handbook Washington Taylor amp Francis 1995 pp 3-14
LO NGFO RD N T VARCL Software for variance com ponent analysis of data with nested ran-
dom effects (m axim um likelihood) Groninge n ie c ProGAMMA 1993
MASLACH C Burnout A multidim ensional perspe ctive In W B Schaufe li C Maslach amp
T Marek (Eds) Profession al burnout recent developm ents in theory and research Ne w
York Taylor amp Francis 1993 pp 19-32
McARDLE J J amp HAMAGAMI F Multile vel models from a multiple group structural
equation perspe ctive In G A Marcoulide s and R E Schumacke r (Eds) Advanced stru c-
tural equation m odeling Issu es and techn iques Mahwah NJ Lawre nce Erlbaum Associ-
ates 1996 pp 89-124
MUTHEacuteN B O Multileve l covariance structure analysis Sociological Methods amp Research
1994 22(3) 376-398
PARKES K Locus of control as moderator An e xplanation for additive versus interactive
findings in the demand-discre tion mode l of work stress British Journal of Psychology
1991 82 291-312
PAYNE R amp FLETCHE R B(C) Job demands supports and constraints as pre dictors of
psychologica l strain among schoolteache rs Journal of Vocational Behavior 1983 22 136-
147
RAUDENBUSH S ldquoCe nte ringrdquo pre dictors in multile vel analysis Choice s and conse que nce s
Multilevel m odelling newsletter 1989 12 10-12
REICHE H M J K I amp DIJKHU IZ EN N VAN Vragen lijst organ isatie stress Test-han-
dleiding [O rganizational stre ss que stionnaire Test-m anual] Nijmege n Unive rsity of Ni-
jmege n 1979
RIJK A E DE BLANC P M LE SCHAUFE LI W B amp JONGE J DE Active coping
and nee d for control as moderators of the Job De mand-Control Mode l Effects on burn-
out Journal of Occupation al and Organizational Psychology 1998 71 1-18
SCARPE LLO V amp CAMPBELL J P Job satisfaction Are all the parts there Personnel
Psychology 1983 36 577-600
SCHAUFE LI W B amp DIERENDO NCK D VAN The construct validity of two burnout
me asure s Journal of Organizational Behavior 1993 14 631-647
SCHAUFE LI W B amp DIE RE NDO NCK D VAN Burnout een begrip ge meten De Ne d-
erlandse ve rsie van de Maslach Burnout Inve ntory [Burnout the me asurem ent of a con-
struct The Dutch ve rsion of the Maslach Burnout Inve ntory] G edrag en G ezondheid
1994 22(4) 153-172
120 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
SCHNALL P L LANDSB ERGIS P A amp BAKER D Job strain and cardiovascular dis-
ease Annual Review of Public Health 1994 15 381-411
SCHNEID ER B O rganizational be havior Annual Review of Psychology 1985 36 573-611
SCHNEID ER B The people make the place Person nel Psychology 1987 40 437-453
SCHWARTZ J E PIE PER C F amp KARASEK R A A proce dure for linking psychosocial
job characte ristics data to health surve ys Am erican Journal of Public Health 1988 78(8)
904-909
SEMME R N Z APF D amp GREIF S ldquoShare d job strainrdquo A ne w approach for asse ssing
the validity of job stre ss me asure ments Journal of Occupational and Organizational Psy-
chology 1996 69 293-310
SHIR O M A Burnout in work organizations In C L Coope r and I T Robertson (Eds)
International review of industrial and organ izational psychology Chiche ster John Wile y amp
Sons 1989 pp 25-48
SIE GRIST J PETE R R JUNGE A CREME R P amp SEIDE L D Low status control
high e ffort at work and ischemic he art dise ase Prospe ctive evidence from blue -collar
me n Social Science and Medicine 1990 31(10) 1127-1134
SNIJDE RS T A B amp BO SKER R J Mode led variance in two-le vel models Sociological
Methods amp Research 1994 22(3) 342-363
SOumlDERFELDT B SOumlDE RFELDT M JONES K O rsquoCAMPO P MUNTANE R C O HL-
SO N C G amp WARG L E Does organization matter A multilevel analysis of the
de mand-control mode l applied to human service s Social Science and Medicine 1997
44(4) 527-534
SPE CTOR P E Inte ractive e ffects of perce ived control and job stre ssors on affective reactions
and health outcome s for clerical worke rs Work and Stress 1987 1 155-162
SPE CTOR P E A conside ration of the validity and me aning of se lf-report measure s of job
conditions In C L Coope r and I T Robe rtson (E ds) International review of industrial
and organ izational psychology New York Wiley amp Sons 1992 pp 123-151
SPE CTOR P E BRANNICK M T amp CO OV ERT M D Job Analysis In C L Cooper
and I T Robe rtson (Eds) International review of industrial and organ izational psychology
Chiche ster Wiley amp Sons 1989 pp 281-328
THEO RELL T amp KARASEK R A Curre nt issues re lating to psychosocial job strain and
cardiovascular disease re search Journal of Occupational Health Psychology 1996 1(1)
9-26
THO MAS J G Source s of social information A longitudinal analysis Hum an Relations 1986
39 855-870
VANCO UV ER J B MILLSA P R E amp PETERS P A Multile vel analysis of organizational
goal congrue nce Journal of Applied Psychology 1994 79(5) 666-679
WALL T D JACKSO N P R MULLARKEY S amp PARKER S K The demands-con trol
model of job strain A more spe cific test Journal of Occupational and Organizational Psy-
chology 1996 69 153-166
WANO US J P REICHE RS A E amp HUDY M J O ve rall job satisfaction How good are
single -item measure s Journal of Applied Psychology 1997 82(2) 247-252
WARR P Work unemploym ent and m ental health O xford Clare ndon Pre ss 1987
WARR P B Decision latitude job de mands and e mploye e we ll-being Work and Stress 1990
4(4) 285-294
Z APF D Selbst- und Frem dbeobachtung in der psychologischen Arb eitsanalyse Goumlttinge n
Hogre fe 1989
Z APF D DO RMANN C amp FRESE M Longitudinal studie s in organizational stress re-
search A review of the literature with reference to me thodologic al issues Journal of
Occupational Health Psychology 1996 1(2) 145-169
BIOGRAPHICAL NOTES
JAN DE JONGE is a Lecture r in Work and O rganizational Psychology at the Unive rsity of
Nijmegen His main interest is research in work and organizational psychology in particular
Group and In dividual Level As ses sm en ts of Job Ch aracteristics 121
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is
job characteristics employe e health and flexibilization of work In 1996 he obtaine d his PhD
Degree for a the sis on job autonomy we ll-being and health
GERARD J P VAN BRE UKELEN is a Lecture r in Methodology and Statistics at the Maas-
tricht Unive rsity
JAN A LANDEWEERD is a Senior Le cture r in Work and O rganizational Psychology at the
Maastricht University
FRANS J N NIJHU IS is since 1995 Profe ssor in Work and Health Psychology at the Maas-
tricht Unive rsity
122 de Jon ge van Breukelen Landeweerd and Nijhu is