The structure of the questionnaire on experiencing and assessing stress at work under review :...

Translation of scales – p. 1 / 33

National Institute Research of Working Conditions

MULTIDIMENSIONAL SCALING.

EXPERIMENTING WITH PROXSCAL.

The structure of the questionnaire on experiencing and assessing stress at work under review

Guy Notelaers

Brussels

2000

Translation of scales – p. 2 / 33

Translation of scales

Scale Dutch name abbreviation

Pace and amount of work Werktempo en hoeveelheid whMental load Geestelijke belasting gbEmotional load Emotionele belasting ebVariety in your work Afwisseling in het werk awOpportunities to learn Leermogelijkheden lmIndependence in your work Zelfstandigheid in het

werkziw

Relations with colleagues Relatie met collega's rmcRelations with direct boss Relatie met directe

leidingrml

Participation Inspraak insprUncertainty about the future

Toekomstonzekerheid ton

Pleasure in your work Plezier in het werk piwInvolvement in the organisation

Betrokkenheid bij deorganisatie

bbo

Recuperation needs Herstelbehoefte herbWorrying Piekeren p

The structure of the questionnaire on experiencing and assessing stress atwork under review – p. 3 / 33

Introduction

NOVA (National Institute for the Research on Working Conditions)is investigating stress on a large scale in companies in Belgium.The measurement model is developed by van Veldhoven PhD.Starting from a facetanalysis, he constructed 25 Mokkenscales and1 Rashscale which have been validated in the Netherlands (vanVeldhoven, 1996) and in Flanders (Notelaers, 1999). In theNetherlands the measurement instrument (VBBA or questionnaire onexperiencing and assessing stress at work) has been usedintensively for the past 5 years. SKB, the foundation who ownsthe questionnaire has already studied approximately 100 000 Dutchemployees. In Belgium we interviewed about 30 000 employees.

The aim of this research is to help organisations and companiesto develop a policy of counteracting stress at work. Theanalysis is kept very simple. We use analysis of variance tofind out if belonging to a certain group makes a difference. Theresult is an overview of the organisation’s position as opposedto a reference population on all the questioned angles (scales)which have quite good measurement properties.1

Taking into account that the items in question are Mokken items,we only have to add up the items to get scalescores. The scoresof the scales are standardised from 0 to 100 where 0 means havingnothing of a particular latent trait and 100 meaning completelyapplicable or fully experiencing the latent treat.2

1 In the annex we will give an overview of Loevinger’s weighted H whichmeasures the degree of homogeneity of the scale constructed in MSP.

2 f.i- pleasure at work (experiencing no pleasure at all at work is zero,having tremendous pleasure at work is 100)

- or work load (0 nothing to do or having no problems with work load,100 being at the edge)

So it is obvious that the number is expressing something about the way thelatent trait is applicable to the respondent.


An interesting point is that the author of the questionnaireproposes a second-order factoranalysis containing six factors :Psychological Demands, Variety, Possibilities to Control, SocialOrganisational, Satisfaction (well-being) and Tension.

This structure provides the starting point for this paper onmultidimensional scaling. First we want to see if the structurevan Veldhoven proposes holds for the Belgian benchmark. Secondlywe are interested in the fact if being a blue collar worker, anemployee, a nurse or a manager makes a difference. For thisreason we like to fit an individual differences scaling model forthe data from the Belgian benchmark. Thirdly we like toinvestigate if the data gathered in the Netherlands and Belgiumby matching on key variables as age group, educational level,industry, being a manager or not and gender, behave differently.This is also an invitation to use a individual differencesscaling model.

Knowing that ALSCAL is supported by Forrest Young in SPSSminimises S-STRESS and thereby gives too much ‘weight’ to theextreme items in the stimulus-configuration we will not use thealgorithm. In this paper we will use PROXSCAL3 that Busingproposed at SOFTSTAT in 1997 and that now has been implemented inSPSS as the counterpart of ALSCAL. This algorithm minimises rawnormalised stress. The result is a far more honest Euclideanspace.4

3 Using PROXSCAL in SPSS turned out to be very hard. The manual of SPSS i.c.Categories resembles more a printed helpfunction than the former manuals ofthe SPPS 6-version. Furthermore we noticed that very few or no papers areprinted in books or articles. This will be quite normal seen the fact thatthe new SPSS release dates from May this year. To have more information Icontacted SPSS but they just gave the references of Forrest Young andDelbeke who have worked on ALSCAL. Rather contradictionary I would say.Subsequently I contacted Busing who was not a great help.

4 If it does not take too much time we will compare the S-Stress with the rawnormalized stress and the deriving stimuli configurations.


Inspecting van Veldhoven’s model : factoranalysis in SPSS

Although the model we introduced, fitted in LISREL we neverobtained more than an approximate RMSEA fit. Van Veldhoven canstick with his model. This is not strange since he comparedsolutions where he should use RMSEA as a base to compare models.Strange is that he never talks about AIC and CAIC which shouldhave been used to evaluate the fit of the model. From the momentwe had enough data5 we tried to see if the structure vanVeldhoven has offered, fits. In a exploratory factor analysiswith SPSS, we just found evidence for a two-factor of the second-order solution.

InitialEigenvalues

ExtractionSums ofSquaredLoadings

Factor

Total % ofVariance

Cumulative%

Total % ofVariance

Cumulative%

1 4.096 29.255 29.255 3.631 25.933 25.9332 2.542 18.160 47.415 1.968 14.058 39.9913 1.124 8.029 55.443 .513 3.662 43.6534 .951 6.794 62.237

For the extraction of the variance, SPSS needed 10 iterations andas we see the first two eigenvalues soak up all variance so tospeak. So there is no evidence at all to accept a third factor.When we run the program again just asking for a two-factorsolution we could plot the following graph on basis of the matrixof loadings.6 The result is a two-dimensional graph like wewould have in the PROXSCAL output. If we had distances we couldsee some resemblance with MDS and in particular the spatial

5 800 to be exactly. These are necessary to estimate the model with weightedleast squares on the basis of an asymptotic covariance matrix.

6 We did not rotate the solution because of the fact that the group stimulusconfiguration in an individual differences model is orthogonal.


interpretation.7 But here we are looking at loadings and the waythey load on the two axis or factors.

-0,4

-0,2

0

0,2

0,4

0,6

0,8

-0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1

wh

gb

eb

aw

lm

ziwbbo

rm c rm l

piw

piw

ton

herb p

Individual determ ination of latent trait

organisational det of laten trait

Seen from the matrix of factorloadings and from this graph themajor part of the scales load on two factors. Only 'Uncertaintyabout the future' doesn’t load more on one or another factor. Ifwe look at the loading itself we see that is does not load highon either factor. If we interpret the latent variables of thesecond order we can say that jobstrain can be decomposed into twoparts : organisational features (relation with colleagues, withthe boss, autonomy, participation, involvement, pleasure at work)and individual features like mental stress, emotional work load,worrying , need for recuperation. This may provide a conclusionto a never ending debate about stress : jobstrain is due to theorganisation and due to the individual who is working at theorganisation.

7 There is an immediate link with MDS. In metric CMDS a matrix of distancesD between n entities is given and the aim is to find a low-dimensionalconfiguration of the entities such that the distances are approximated in aleast-square sense. When these distances are Euclidean distances, thecoordinates contained in X do represent the principal coordinates whichwould be obtained when doing PCA on a matrix A (of n entities observedw.r.t. p variables). This approach is called principal coordinates analysis.(Delbeke, Van Deun)


Why MDS?

MDS has its origin in psychometrics, where it was meant to helpunderstand people's judgments of the simularity of members orobjects. Torgerson proposed the first MDS method and coined theterm. MDS has now become a widely used general data analysistechnique. (Young, F.) Knowing that PCA and PAF are twoalgorithms that need at least interval data to give robustsolutions and knowing that the measurement level is ordinal weshould not even bother to calculate in SPSS. Contrary to PCA orPAF in SPSS multidimensional scaling is particulary suited forthe analysis of ordinal data.(Delbeke, L. Van Deun, K)

In multidimensional scaling we try to find a configuration ofpoints in which the distance between these points match as closeas possible the proximities between the objects.(Busing, 1998 :2) This configuration allows to gain insight in the underlyingstructure of relations between entities by providing ageometrical representation of these relations.(Delbeke, L. VanDeun, K). This representation is very straightforward. Eachobject or event is represented by a point in a multidimensionalspace. The points are arranged in this space so that thedistances between a pair of points have the strongest possiblerelation to the similarities among a pair of objects. That is,two similar objects are represented by two points that are closetogether, and two dissimilar objects are represented by twopoints that are far apart. (Young, F)

Does the kind of work the respondents do influence their perception of the questionnaire?

Introduction

For our purpose we use the data as described earlier. In ourdataset we were able to recode several variables giving


information about the work the respondent does. Out of the 9categories we use at NOVA we take 4 for this paper :

Source 1 : blue collar workerSource 2 : white collar workerSource 3 : nurses en doctorsSource 4 : managers

In SPSS we would prefer to calculate correlation matrices forfour sources but without the data of the example given inPROXSCAL section of categories we have to allow the algorithm tocalculate data as the basis four distances and disparities.8 Wecalculate from these dissimalarities a Weighted EuclideanIndividual Differences Model.

How many dimensions?

When we inspect the iteration history we see that raw normalisedstress immediately drops from .19 with a simplex start to .03after the first iteration. The raw stress value itself is notvery informative. A large value does not necessarily indicate abad fit. (Borg&Groenen : 34) The reason for this is that it isscale dependent. When we take the square root of it we get theStress-1 (Kruskal). A different form of implicit normalisation isthe Stress 2 (Borg&Groenen : 203). When we normalize it, it isno longer scale dependent and we get the stress value that isused by the PROXSCAL algorithm. Compared to the other existingscaling programs, PROXSCAL has a number of importantadvantages(Busing, 1997 : 68). In contrast to the subgradientmethod used in PROXSCAL, the gradient methods used in MDSCAL andKYST (Kruskal), MINISSA (Lingoes) do not guarantee monotoneconvergence.9 So, convergence of the algorithm is guaranteed,8 The input matrices are given in the annex of this paper.9 In the case of ordinal data we want to recover the order of theproximities and not the proximities or a linear transformation ofthe proximities. This means that another procedure has to befollowed than that of singular value decomposing, namely aprocedure which is invariant under monotonic transformations.


but not necessarily to the global minimum of the function. Infact, convergence to local minima is often observed whendimensionality is relative low, but especially 1.(Busing : 1998 :4) Compared to ALSCAL10 there is in PROXSCAL no prior conversionof the data needed, and the unsquared similarities are directlyapproximated by unsquared distances (Busing, 1997: 2). Thenormalised raw stress however is not used often in the literatureand there exist no rules of thumb to interpret the quality of theminimisation procedure.11 Busing (1997 : 71) calculates the totalfit by subtracting total stress from 1. For the Stress-1 we canborrow a rule of thumb from Kruskal(1964a) : .20=poor, .1=fair, .05=good, 0.025 excellent and0=perfect (Borg&Groenen, 38).

Concerning selection of the appropriate number of dimensions inthree-way mds, from a set of solutions ranging from maxdim downto mindim dimensions there are no universally accepted criteria,only various strategies (Arabie,Caroll&DeSarbo : 1987 : 35).Coxon (1982) puts it clear that the more dimensions12, the easierthe minimisation of stress is obtained. Furthermore, knowingthat stress is influenced by the number of parameters in themodel, a good rule of thumb for determining the number ofdimensions, is I>4R (where I are the number of stimuli and R thenumber of dimensions. (Stroms, 1997).13 Applying this rule meansthat we have to look for a 1 to 3 dimensional solution.

(Delbeke, L. Van Deun, K) Kruskal and Shepard proposed tominimize a fit measure called Stress by an iterative algorithm (Seeannex)(Delbeke, L. Van Deun, K). Proxscal is working with themetric projection problem. For the exact algorithm Busing refers toCommandeur&Heiser (1993) but the reader can read a draft of it inBusing's paper (Busing, F. Commandeur, J. Heiser, W. 1996 : 2-3).

10 In the minimizing process of the stress loss function of Takana, Young andDeLeeuw (S-Stress) the difference between proximities and distances israised to the power two (Borg&Groenen : 204). The consequence of thisprocedure is that opposites in the model are given more weight.

11 We will see immediately that with the PROXSCAL outcome we can make a scree-plot with the same properties as these of the other algorithms.

12 Here we cannot speak of dimensions as in the uni-dimensional scaling (IRT)because we cannot assure that every dimension represents only one latenttrait. (Storms citing Rosenberg)


We can compare our stress-values with those from a Monto-CarloStudy to see if our structure is not a random one. Thesesimulations show that if n grows, then the expected Stress alsogrows while its variance becomes smaller, if m grows, then theexpected Stress becomes smaller. Spence en Ogilvie (1973:515)suggest to reject the null hypothesis of randomness when all thestressvalues in five dimensions were at least three standarddeviations below the appropriate values of their table of meanStress Values14. When we look carefully at the figures in theappendix we see that we have some structure in our data.

But we have to be careful. Borg and Groenen (1997:40) warn usthat this so called null-hypothesis test is a small hurdle totake. Holding up with the M-Space from Spence&Graef is alsodangerous because it tends to underestimate the dimensionality ofa solution. It is far more statistical (from the inferentialpoint of view) to use MULTISCALE which has an inferential pointof view about MDS. We will not go in this algorithm from

13The fact that we are dealing with non-metric MDS does not matterhere. The difference between metric and non-metric does notinfluence the configuration but it does influence Stress very heavyin the sense that ordinal data increase Stress.

14 Stress to compare = calculated stress - 3*.0091 (Storms)

1 2 3 4 5 60

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

Norm alized Raw StressStress-IStress-IIS-StressDispersion Accounted For (D.A.F.)Tucker's Coefficient of Congruence

Dim ension


Ramsey. We can conclude on the basis of the rule of thumbs fromStorms, from Kruskal and the scree-plot that we have indeed atwo-dimensional solution that is not a random one. This meansthat we have more stimuli than 4 times the dimensonality, we havea moderate fit according to Kruskal&Wish and we can see a veryclear elbow at dimension 2.

Common group Space

The common group serves one major thing and that is to interpretthe dimensions of a solution so that the individual spaces canget meaningful. We are not allowed to rotate the common groupplot because it would degenerate the given solution in PROXSCALthat generates an orthogonal solution. When we calculate thecorrelation between the dimensions we get a medium correlation(r=.4).

When we look at the plotted Mokkenscales we can label ourdimensions as jobstrain perceived at the individual level and asjobstrain perceived as having an organisational cause. It isstrange though that Mental load is not as in the factorialsolution where it is a individual component of jobstrain. Hereit is seen as an organizational feature of jobstrain. The majorimportant remark is that our solution is rotated. If we wouldflip the common groupspace with 90° we would obtain our factorialsolution. Here we use this insight to label the Y-axis as non–organisational.


Individual Job Strain

3210-1-2

Non-Organizational Jobstrain

3

2

1

0

-1

-2

Piekeren

Herstelbehoefte

Betrokkenheid bij dePlezier in het werk

Toekomstonzekerheid

Inspraakrelatie met directe

Relatie met collega'Zelfstandigheid in hLeermogelijkheden

Afwisseling in het w

Emotionele belasting

Geestelijke belastin

werktempo en hoeveel

We do not need to interpret more because the common group spaceshould not itself be interpreted if it turns out to be a simplecompromise between the configurations of groups of subjects withvery different patterns of individual weights.(Coxon, 1982a :191)

Subject Weight's space


More important than interpreting the stimulus configuration ofthe common group space is the interpretation of the dimensionweights. The weights are interpreted as the importance, orsalience of each dimension to each individual.(Young, F)

When we draw a 45° line in our configuration we have a cut-offpoint for comparing the usage of the dimensions for our sourcessince the line represents equal usage of the dimensions. Beinga blue collar worker reflects more the usage of the individualdimension of perceiving jobstrain than being in the othersources. The blue collar line is 17° away form the 45° line.Being a white collar worker or nurse or doctor is 10° away fromthe 45° line. These categories identify more with the individualperception of jobstrain. Managers tend to see jobstrain more asan organisational feature. However they differ slightly morefrom the 45° line.

Dimension Weights

Dimension 1

.6.5.4.3.2.10.0-.1

Dime

nsio

n 2

.6

.5

.4

.3

.2

.1

0.0

-.1

SRC_4

SRC_3SRC_2

SRC_1

These findings are in accordance with what we expected. Thedifferences are however less profound than we expected them tobe. It is interesting to note that blue collar workers aresimilar to nurses in perceiving stress. Earlier research we did


at NOVA showed that even the levels are quite alike except foremotional tension, autonomy and need to recuperate. It is alsointeresting to see that the more we go from blue collar workertowards managers the more the accountability of the seconddimension arises. Far away from every one else are the managers.They are the only source who give more weight to theorganisational dimension. They are probably the only group whocan define their work themselves.

We can also look at the weight in another way. Coxon says thatthe squared subject's weight on a dimension is equal to theproportion of variance in the subject's data that can beaccounted for by that dimension (Coxon, 1982a : 194).

Table Accountability of dimension for sources

Source individual organisationalBleu Collar 34,8 10,6White Collar 30,9 15Health Care 30,5 15,2Manager 19,6 26,2

The feature that the squared distance from the origin of thesubject space to a subject’s point in that space is approximatelyequal to the proportion of variance in the subject’s dataaccounted by the full INDSCAL (+/- 36%) solution(Coxon, 1982a :194), enables us to compare the proportion of the explainedvariance with the proportion of the explained variance we have


found with PAF(39%).15 The resemblance is quite obvious. Thereis only a minor difference in the proportions of explainedvariance between the two techniques.

Private space

As we have seen : the two groups differing mostly are managersand blue collar workers. In this section we will stick withthose two and represent only for those groups the subject space.The two private spaces are related : they may be derived from thereference group space by a simple process of differentiallystretching or shrinking the axes of the group space by the squareroot of the subject’s weights.16 (Coxon, 1982a : 191) This isdone in the figures17 in this section.

When we compare the figures we see that the range of the usedspace on Y is bigger for managers than for blue collar workerswhile the range of the used space on X is bigger for blue collarworkers than it is for managers. The consequence of theweighting procedure is that for blue collar people the distancebetween scales, who are more identified by the X axe, becomesbigger while the distance between scales of the organisationaltype gets smaller. For manager it is the reverse. The result isthat scales of the individual type become more alike (similar)for the managers while they become more dissimilar for bluecollar workers.

Compared to the common group space we see that the individualdimension's range is reduced for the blue collar workers from +/-4 to 2,8. The organisational dimension's range is more reduced15 Earlier we said that there is a link between PCA and MDS. The choice to

apply PFA is legitimate because PCA is not really a reductionist techniqueand the proportion explained variance from a PFA is more comparable with theINDSCAL one.

16 To obtain subject I’s private space, we take the coordinates of the pstiumulus points on the 1st dimension of the groupspace (xji) and rescalethem by the square root of subject I’s weight for this dimension (wi

a 1/2 ).17 What really happens in an individual differences model is not given in the

output of the PROXSCAL in SPSS : streching and shrinking of the axis.


(from 3,75 to 2,2). This means that the stimuli who are moreidentified with the organisational dimension become moresimilar. The more individual perceived scales become moredissimilar or get more discriminatory power than those of theorganisational dimension. The consequence is that as the angleof weights is showing, the space of blue collar workers seemsmore unidimensional : jobstrain is seen as an individualfeature. Worrying, Recuperation needs, Pace and amount of workand Emotional load are more a result of the individual. This isalso the case for Opportunities to learn, Independence in yourwork, Variety in your work and Participation. With these scales the difference is quite small and could be neglected.

The situation of the managers is quite opposite to the situationof the blue collar workers. For them the organisationaldimension is reduced from 3,8 to 2,8 and the individualperception of jobstrain is reduced from 4 to 2.4. When wecompare the rescaling of both the private spaces, we seeimmediately that it has little impact on the managers space. Butfor them Recuperation needs, Independence in your work, Emotionalload and worrying are more alike than for blue collar workers.

Bleu collar space

-1

-0,8

-0,6

-0,4

-0,2

0

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

-1,4 -1,2 -1 -0,8 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8

Individual

non

orga

niza

tion

al

p

herb

bbopiw

ton

insprrm lrm c

ziwlm

awgb

wh eb


Those scales are less important in the perception of jobstrainfor managers. More important (although the weight is not sodifferent from the 45° line) is that Mental load, Relations withcolleagues, Relations with your direct boss and Involvement inthe organisation are seen as being more determined by theorganisation. A big difference with the blue collar space isthat the managers space does not tend as hard as the blue collarspace to be unidimensional.

In the appendix we have calculated the distances between thescales for the two private spaces in order to compare them byinvestigating which scale becomes more (dis)similar. In doing sowe don't look at small movements. We put a threshold at .3. Inthe table a '-' means that the distance between two scalesdiminishes smaller for blue collar workers and a '+' means thatthe distance becomes smaller for managers. The result is that wehave an idea about how much more similar scales are in theprivate spaces we have just drawn. For managers Mental load iscloser to Participation and worrying than for BC. Emotional loadfor manager is more related to Participation. These results donot undermine at all our findings from other research. Havingmore Participation enlarges Mental load and gives certainly more

-1,2-1

-0,8-0,6-0,4-0,20

0,20,40,60,81

1,21,41,61,82

-1,2 -1 -0,8 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6

ton

herb

eb

p

wh

gbrm c

rm l

bbo

insprlm

ziwaw

piw


to worry about. That these similarities lead to more emotionaltension is quite normal. Furthermore worrying is more related toRelations with colleagues and Relations with your direct boss andsurprisingly at first sight to Pleasure in your work than it isfor BC. We could see this as quite a negative relationship butwe are convinced that good relationship with colleagues anddirect hierarchy gives more jobsatisfaction. The fact thatmanagers tend to worry more has to do with the fact that theylike their work more then BC and we cannot put aside the factthey design the work that has to be done.

For BC every scale except Recuperation needs, worrying andEmotional load becomes closer to Uncertainty about the futurebecause the salience of the second dimension is quite small.This can be explained by decomposing stress. In thedecomposition we see immediately that Uncertainty about thefuture is the worst fitting scale for 3 sources but the amplitudeis most pronounced for BC. One could be tempted to delete thescale from the solution but in order to question the structure ofvan Veldhoven that is not in our interest.

Van Veldhoven’s structure revised?

It is quite obvious that we are able to reject the sixdimensional structure that van Veldhoven proposed in 1997. Notonly the stressvalues point in that direction but the INDSCALsolution itself makes a lot of sense. However, if we look moreclosely at the common group space and the private space, we haveto question a few issues. Why Variety in your work,Opportunities to learn, Independence in your work, Participationare so closely together? Van Veldhoven calls them variety andautonomy. Why are Pleasure in your work and Involvement in theorganisation situated at the same y-coordinate in the samequadrant? Van Veldhoven calls those well-being. Emotionaltension and workload are also very close to one anohter. In vanVeldhoven’s structure


they are called psychological demands. Actually, if we arehonest the only two scales that do not fit like they shouldaccording to van Veldhoven are Uncertainty about the future andMental load. Reading Kruskal&Wish (44) brings up the idea thatwe could draw clusters. When we look at this solution we areurged to follow Guttman's advice to take another technique forrepresenting this dataset. Kruskal&Wish do not choose thiscompetitive point of view. They wonder if the clusterlikeinterpretation reveals more about the dimensions. We can followtheir idea. When we see that Recuperation needs and worryinghave almost the same x-coordinate just like Relations withcolleagues and Relations with your direct boss or Pleasure inyour work and Involvement in the organisation, we are able to saythat van Veldhoven's structure is reflected in the formerconfiguration but that he has not seen that there are two latenttraits who can explain the relationship among them and that is

3210-1-2

3

2

1

0

-1

-2

Piekeren

Herstelbehoefte

Betrokkenheid bij dePlezier in het werk

Toekomstonzekerheid

Inspraakrelatie met directe

Relatie met collega'Zelfstandigheid in hLeermogelijkheden

Afwisseling in het w


Geestelijke belastin

werktempo en hoeveel


organisational perception of jobstrain and individual perceptionof jobstrain.

Can we compare Holland and Flanders?

Introduction

We already said that we have gathered data from Holland andBelgium by matching on certain key variables. While comparingvan Veldhoven and Notelaers, we were surprised by the meanscalescores of both groups. The difference was so huge that vanVeldhoven and Notelaers formulated a hypothesis saying that theusage of the scales by the respondents must be different. Thefollowing conclusion was that while comparing the data we were infact comparing totally different things.

This hypothesis could be tested by performing INDSCAL on thedata. If the respondents use the scales differently than wewould expect a difference in the weights on the two dimensions.18

A two-dimensional solution

From the decomposition of Stress in the appendix we can concludethat our solution in two dimensions gives an acceptable fit.However, the common group space is not comparable with the formercommon group plot for the functions we have investigated. Ininterpreting the dimensions we can opt for the same labels asbefore : individual perception and organisational perception ofjobstrain. Not going too deep in the configuration itself wenotice that it looks more like we expected. Opportunities tolearn, Variety in your work, Independence in your work andParticipation are seen as organisational and not as individualcomponents. Pleasure in your work, and Involvement in the

18 We must warn the reader that the data are not completely compatible. Thescales Uncertainty about the future and Mental load are deleted. Comparedto the former solution this means that the worst fitting scales whileminimizing stress are deleted.


organisations are a mixture of both components. Emotional load,worrying and Recuperation needs are seen as individual relatedscales. Apparently, the absence of Mental load and Uncertaintyabout the future influence the configuration very much. Althoughwe can also speak of a more clusterlike solution where the readerrecognizes the structure van Veldhoven proposes, we see that atwo-dimensional solution is again rejecting a six factorsolution.

For this paper we are more interested in the salience of thedimensions since we would like to know whether Holland andFlanders differ substantially from one another. Therefore weprint the table of dimension weights.

Dimension WeightsDimens

ionSource 1 2Hollan

d.511 .436

Flanders

.466 .486

The weights are quite alike. This would lead to the conclusionthat we reject the hypothesis of van Veldhoven and Notelaersstating that the questionnaire is used differently in Flandersand Holland. Intuitively we are tempted to do so but we cannotconclude that this is the case because we are not able to testthe hypothesis. We could use MULTISCALE or some directionalstatistics to compare the vectors. We could also use PINDIS tocompare the space groups for Flanders and Holland. Thisalgorithm gives us the opportunity to make the groupspaceorthogonal and with the Langeheine simulations significance testsfor Procrustean Individual Differences Scaling we would have atool to formally test our hypothesis. Alternatively we could usethe ANOVA-procedure that Coxon and Jones suggest in Measurementand Meanings.


For this paper we can not use MULTISCALE. It would bring us toofar away from the objective of this paper. The same hold forPINDIS. However the usage of directional statistics and thesuggestion of Coxon could be done. We were not able to do sobecause of the fact that the books we needed were not availableWhich is a pity as far this paper is concerned. The topic willhowever be discussed in the paper on Latent Class Analysis.

Conclusion

Van Veldhoven proposed in his Phd a six-dimensional structure forthe Mokkenscales he has developped. Experiments with factoranalysis in SPSS confronted his findings. But the efforts to fitan alternative two-dimensional structure failed in LISREL. Oureffort here in this paper was worth the shot. If we can becertain about one thing than it is that the scales in thequestionnaire can be reduced to two dimensions. The firstdimension translates the individual perception of jobstrain whilethe second dimension tells us about the organisational perceptionof jobstrain. Along these dimensions we have groups of scalesthat tend to cluster in the six-dimensional way like vanVeldhoven proposes.

Knowing that people do different work is asking if doing otherkind of jobs matter with respect to the perception of jobstrain.For managers and blue collar workers we can say that it doesmatter. blue collar workers tend more to see jobstrain as anindividual thing than managers do. It is striking that whitecollar employees do not differ at all from people in health-care.The answer to the question if they differ significantly we cannotgive. We have to draw the same conclusion while testing thehypothesis whether the data from a matched sample from Hollandand Flanders differ significantly. From the weights one couldconclude that there is no difference in the salience of thedimension between the two samples. But we did not have enoughresources to test the hypothesis in an inferential way.


Nevertheless the usage of the INDSCAL model was very fruitful.We experienced that working with PROXSCAL leveled out thedisadvantages within MINISSA or ALSCAL. The algorithm is veryflexible but support is insufficient, in the same way the manualis not substantial enough. The fact that few publications existsusing PROXSCAL in SPSS 10.0 brought us back to Stress 1 as aanchor to evaluate our configuration and the minimisation ofdistances and proximities beneath it.

Literature – p. 24 / 33

Literature

Arabie, P. et.al., Three-way scaling and clustering, Quantitative Applications in the Social Sciences. A Sage University Paper, 65. 1987.

Busing, F. Commandeur, J. Heiser, W. PROXSCAL. A multidimensional scaling program for Individual Differences Scaling with Constraints. Department of Data Theory. Leiden University, 1996.

Busing, F. PROXSCAL : A multidimensional scaling program. In : Bandilla, A. Faulbaum, F. Softstat 97. Advances in StatisticalSoftware. The 9th Conference on the Scientific Use of Statistical Software, 1997, Heidelberg.

Busing, F. Proxscal. User's guide for version 6.3 Data Theory Scaling System Group. Leiden University. 1998.

Coxon, A.P.M. Key Texts in Multidimensional scaling. London, 1982.

Coxon, A.P.M. The User's guide to multidimensional scaling. London, 1982.

Delbeke, L. Van Deun, K. Multidimensional Scaling. Open Distance Learning, http://www.uni-bonn.de/mathpsyc/doc/delbeke/delbeke.htm

Kruskal, J. Wish, M. Multidimensional Scaling. Quantitative Applications in the Social Sciences. A Sage University Paper, 11. 1987.

SPSS. Categories. User's Guide SPSS, version 10.0. 2000.

Storms, G. MDS. Course at KU-Brussel. 1997.

http://www.uni-bonn.de/mathpsyc/doc/delbeke/delbeke.htm

Literature – p. 25 / 33

Young, Forrest. Multidimensional Scaling. University of North Carolina. Http://forrest.psych.unc.edu/teaching/p208a/mds/mds.html

http://forrest.psych.unc.edu/teaching/p208a/mds/mds.html

Annexes – p. 26 / 33

Annex 1 Psychometric Quality of Scales

Schalen Loevinger’s weighted HNederland

DMnumber of violationsin P matrix above 10% inspected per item

Loevinger’s weightedHVlaanderen

DMnumber of violations in P matrix above 10% inspected per item

Deviation of Flanders against Nederland

Werktempo en hoeveelheid

.5 0 .46 0 -.04


.44 0 .4 0 -.04

Lichamelijke inspanning

.6 0 .63 0 .03

Afwisseling in het werk

.48 0 .48 0 0

Leermogelijkheden

.63 0 .68 0 .05

Zelfstandigheid in het werk

.53 0 .53 0 0

Relatie collega’s

.46 0 .61 0 .15

Relatie directe leiding

.57 0 .66 0 .12

Inspraak .55 0 .6 0 .04Toekomstonzekerheid

.81 0 .83 0 .02

Plezier in het werk

.5 0 .54 0 .04

Betrokkenheid .45 0 .47 0 .02


bij de organisatie/ het bedrijfHerstelbehoefte

.49 0 .54 0 .05

Piekeren .77 0 .71 0 -.06 .38/14=.02

7


Schalen Cronbach’s ± Nederland

Cronbach’s± Vlaanderen

Deviation of Flanders against Nederland

Werktempo en hoeveelheid

.87 .85 -.02


.78 .77 .01

Lichamelijke inspanning

.88 .9 .02

Afwisseling in het werk

.82 .83 .01

Leermogelijkheden

.83 .86 .03

Zelfstandigheid in het werk

.88 .91 .03

Relatie collega’s

.9 .91 .01

Relatie directe leiding

.83 .92 .07

Inspraak .88 .9 .02Toekomstonzekerheid

.94 .95 .01

Plezier in het werk

.77 .83 .06

Betrokkenheidbij de organisatie/ het bedrijf

.74 .77 .03

Herstelbehoefte

.87 .89 .02

Piekeren .78 .77 -.01.29/14=.02


Annex 2 Factoranalysis

Total Variance ExplainedInitialEigenval

ues

Extraction Sums

ofSquaredLoadings

Factor Total % ofVariance

Cumulative %

Total % ofVariance

Cumulative %

1 4.096 29.255 29.255 3.581 25.578 25.578 2 2.542 18.160 47.415 1.951 13.936 39.513 3 1.124 8.029 55.4434 .951 6.794 62.2375 .882 6.301 68.5386 .794 5.671 74.2097 .661 4.720 78.9298 .574 4.101 83.0309 .496 3.542 86.57210 .464 3.316 89.88811 .451 3.219 93.10712 .399 2.848 95.95613 .296 2.112 98.06814 .270 1.932 100.000

Extraction Method: Principal Axis Factoring.

Factor MatrixFactor

1 2 Pace and amount

of work-.267 .613

Mental load .106 .366 Emotional load 8.470E-02 .674 Variety in your

work.684 .428

Opportunities tolearn

.760 .218

Independence inyour work

.652 9.329E-02

Relations withcolleagues

.425 -.299

Relations withdirect boss

.632 -.268

Participation .756 .129


Uncertainty aboutthe future

-.224 -7.861E-02

Pleasure in yourwork

.621 -6.393E-02

Involvement inthe organisation

.556 -4.670E-02

Recuperationneeds

-.350 .521

worrying -4.339E-02 .536 Extraction Method: Principal Axis Factoring.a 2 factors extracted. 6 iterations required.


Annex 3 Output of PROXSCAL for four sources

ProximitiesProximities

Source: Bleu Collarwerktempo en

hoeveelheid

Geestelijke

belasting

Emotionele

belasting

Afwisseling in

hetwerk

Leermogelijkhe

den

Zelfstandighei

d inhetwerk

Relatiemet

collega's

relatiemet

directeleiding

InspraakToekomstonzekerheid

Plezierin hetwerk

Betrokkenheidbij deorganis

atie

Herstelbehoeft

e

Piekeren

werktempo en

hoeveelheid

.

Geestelijke

belasting

1065.468

.

Emotionele

belasting

947.842 1749.493

.

Afwisseling in

hetwerk

864.237 1169.5561028.48

3.

Leermogelijkhe

den

1024.0631405.10

1975.517 677.510 .

Zelfstandighei

d inhetwerk

954.088 1295.4901011.63

7674.347 715.094 .

Relatiemet

collega's

1160.175809.383 1784.13

81172.57

01353.83

21203.34

7.

relatiemet

directeleiding

1188.043893.969 1758.52

61124.43

11282.11

91156.66

9593.587 .

Inspraak965.653 1478.30

9851.577 751.178 718.421 661.314 1382.24

31281.78

8.

Toekomstonzekerheid

1210.3381320.48

31586.89

11343.03

51462.63

61392.81

91323.28

01356.89

91447.72

9.

Plezierin hetwerk

1447.2541045.80

71984.17

81287.63

11476.94

51380.55

1877.293 841.279 1545.81

11494.39

5.


atie

1217.6681144.94

51633.78

61082.29

61195.26

71138.47

6970.389 871.992 1219.78

81410.24

9911.199 .

Herstelbehoeft

e

953.178 1635.0471079.33

31347.44

61421.36

91361.65

21722.66

81738.40

41301.75

81583.51

32018.64

71732.94

9.

Piekeren1245.67

91924.76

4970.138 1349.61

11317.37

91373.00

01977.29

71934.91

11238.28

51793.21

62157.88

81855.18

71234.87

5.

Proximities


Source: White Collarwerktempo en

hoeveelheid

Geestelijke

belasting

Emotionele

belasting

Afwisseling in

hetwerk

Leermogelijkhe

den

Zelfstandighei

d inhetwerk

Relatiemet

collega's

relatiemet

directeleiding


Plezierin hetwerk


atie

Herstelbehoeft

e

Piekeren

werktempo enhoeveelheid

.

Geestelijkebelast

ing

1167.298

.

Emotionele

belasting

861.394 1763.937

.

Afwisselingin hetwerk

902.169 942.718 1279.640

.

Leermogelijkheden

968.038 1204.6251195.72

1647.091 .

Zelfstandigheid in

hetwerk

944.910 1062.0581254.07

0666.362 757.890 .

Relatie metcolleg

a's

1204.658754.599 1736.15

9904.096 1050.56

6907.244 .

relatie metdirect

eleidin

g

1231.131865.555 1716.84

5898.205 994.028 877.661 591.202 .

Inspraak

935.478 1389.7351007.52

7869.140 782.602 747.501 1237.17

41103.85

2.

Toekomstonzekerhei

d

1199.3981791.59

21173.86

31565.18

81524.95

41503.61

41764.22

61798.81

21432.71

5.

Plezier inhetwerk

1520.640934.429 2052.19

31085.46

81272.42

51194.12

9860.767 868.382 1504.76

22054.57

2.

Betrokkenheid bij

deorganisatie

1199.0781143.30

61570.40

5947.224 967.726 979.996 1024.32

8925.401 1057.95

61728.83

01044.93

0.

Herstelbehoe

fte

1032.8401819.16

81032.48

11512.77

21484.79

41512.56

81827.53

01844.44

31406.27

51440.71

82170.50

51780.25

1.

Piekeren

1259.7401898.85

41197.28

01562.62

01542.36

51596.63

61927.39

81932.07

31477.71

61576.19

02185.21

81793.47

81200.05

0.

Proximities

Source: Health Carewerktempo en

hoeveel

Geestelijke

belasti

Emotionele

belasti

Afwisseling in

het

Leermogelijkhe

den

Zelfstandighei

d in

Relatiemet

collega

relatiemet

directe


Plezierin hetwerk

Betrokkenheidbij de

Herstelbehoeft

e

Piekeren

Annexes – p. 33 / 33heid ng ng werk het

werk's leiding organis

atiewerktempo en

hoeveelheid

.

Geestelijke

belasting

711.687 .

Emotionele

belasting

421.418 896.149 .

Afwisseling in

hetwerk

574.222 599.428 624.761 .

Leermogelijkhe

den

611.271 720.751 611.329 426.794 .

Zelfstandighei

d inhetwerk

623.013 843.816 559.197 512.077 526.582 .

Relatiemet

collega's

759.647 529.738 894.710 561.252 669.113 704.701 .

relatiemet

directeleiding

778.403 617.656 897.816 574.943 657.824 701.191 421.132 .

Inspraak622.839 936.640 540.161 586.588 525.562 451.093 809.454 722.022 .

Toekomstonzekerheid

788.252 1174.489736.528 970.860 893.262 901.277 1155.94

81149.04

7855.598 .

Plezierin hetwerk

1033.359631.951 1196.93

4785.655 928.006 1008.74

3547.359 597.709 1108.52

31418.11

7.


atie

818.435 717.340 930.207 670.484 713.048 770.601 574.604 550.621 805.687 1134.941673.424 .

Herstelbehoeft

e

628.852 1031.973662.554 922.536 913.724 898.811 1090.36

61121.70

3895.501 949.112 1368.12

51128.19

2.

Piekeren861.507 1263.31

3799.779 1069.46

31046.52

21011.81

01317.83

41326.78

11019.99

1972.147 1562.02

51305.63

7816.279 .

Proximities

Source: Managerswerktempo en

hoeveelheid

Geestelijke

belasting

Emotionele

belasting

Afwisseling in

hetwerk

Leermogelijkhe

den

Zelfstandighei

d inhetwerk

Relatiemet

collega's

relatiemet

directeleiding


Plezierin hetwerk


atie

Herstelbehoeft

e

Piekeren

werktempo en

hoeveelheid

.

Geestelijke

belasting

475.656 .

Annexes – p. 34 / 33Emotion

elebelasti

ng

353.590 698.346 .

Afwisseling in

hetwerk

427.348 323.984 596.221 .

Leermogelijkhe

den

397.746 388.925 521.981 266.131 .

Zelfstandighei

d inhetwerk

412.667 389.452 532.957 297.227 271.634 .

Relatiemet

collega's

416.220 315.681 622.995 287.574 302.078 304.292 .

relatiemet

directeleiding

491.984 327.394 677.864 303.404 312.512 338.694 255.421 .

Inspraak421.578 414.139 526.344 305.075 268.225 286.618 345.039 314.959 .

Toekomstonzekerheid

676.864 1000.606493.081 913.918 818.153 814.292 903.077 970.420 834.499 .

Plezierin hetwerk

661.747 381.686 869.231 402.270 491.243 484.394 404.009 357.729 518.590 1161.098

.


atie

537.341 433.857 682.076 383.519 380.378 415.696 389.401 327.574 364.339 978.040 440.578 .

Herstelbehoeft

e

480.387 787.436 501.615 732.015 695.748 714.329 737.420 800.073 700.419 723.465 962.400 803.828 .

Piekeren567.453 793.783 578.744 713.392 690.411 709.467 743.491 794.033 703.809 777.684 917.260 791.162 546.557 .

Iteration HistoryIterat

ionNormal

izedRaw

Stress

Improvement

0 .19241 .0380 .15442 .0263 .01173 .0228 .00354 .0209 .00195 .0196 .00136 .0187 .00097 .0180 .00078 .0174 .0006


9 .0169 .000510 .0164 .000411 .0161 .000412 .0157 .000313 .0154 .000314 .0152 .000315 .0149 .000216 .0147 .000217 .0145 .000218 .0143 .000219 .0142 .000120 .0141 .000121 .0140 .0001

a Stress of initial configuration: simplex start. b The iteration process has stopped because Improvement has become less than the convergence criterion.

Decomposition of Normalized Raw StressSource MeanSRC_1 SRC_2 SRC_3 SRC_4

Object werktempo enhoeveelheid

.0098 .0098 .0035 .0276 .0127

Geestelijkebelast

ing

.0178 .0208 .0068 .0219 .0168

Emotionele

belasting

.0258 .0182 .0135 .0339 .0229

Afwisselingin het

werk

.0087 .0082 .0066 .0155 .0097

Leermogelijk

.0052 .0115 .0042 .0242 .0113


hedenZelfstandigheid in

hetwerk

.0057 .0130 .0098 .0091 .0094

Relatie metcolleg

a's

.0123 .0052 .0037 .0231 .0111

relatie metdirect

eleidin

g

.0057 .0064 .0063 .0152 .0084

Inspraak

.0166 .0166 .0042 .0101 .0119

Toekomstonzekerhei

d

.0215 .0236 .0155 .0325 .0233

Plezier inhetwerk

.0114 .0118 .0069 .0201 .0126

Betrokkenheid bij

deorganisatie

.0152 .0155 .0110 .0113 .0133

Herstelbehoe

fte

.0173 .0234 .0077 .0190 .0168

Piekeren

.0128 .0204 .0122 .0182 .0159

Mean .0133 .0146 .0080 .0201 .0140


Decomposing Raw Normalized Stress

-0,02

-0,015

-0,01

-0,005

0

0,005

0,01

0,015

wh gb eb aw lm ziw

rmc

rml

insp

rto

npi

wbb

ohe

rb piw

source4source 3source 2source 1


Annex 4 Iterative procedure for minimisation ofStress (Kruskal&Wish)


Annex 5 Output PROXSCAL for four sources

blue collar

Pace and amountof work

Mentalload

Emotional load

Variety inyour work

Opportunities to learn

Independencein your work

Relations with colleagues

Relations with direct boss

Participation


Pleasure in your work

Involvement in the organisation

Recuperation needs

worrying

Pace and amount ofwork

0

Mental load

1,02 0Emotionalload

0,51,59 0Variety in your work

0,630,831,05 0


0,671,110,93 0,27 0

Independence in your work

0,56 10,91 0,17 0,14

0

Relationswith colleagues

0,940,231,62 0,67 0,930,84 0

Relationswith direct boss

1,120,421,64 0,62 0,870,790,18 0

Participation

0,681,29 0,8 0,45 0,190,291,12 1,0

60

Uncertainty about the future

1,121,741,25 1,69 1,781,671,84 1,9

81,77

0


1,70,712,22 1,19 1,411,360,63 0,5

71,612,45

0

Involvement in theorganisation

1,230,151,61 0,59 0,7 0,70,65 0,470,872,240,82

0

Recuperation needs

1,062,180,53 1,59 1,461,452,13 2,1

71,321,272,742,15

0

worrying 1,52,540,97 1,75 1,511,582,41 2,3

71,322,132,932,170,89

0


Managers Pace and amount of work

Mental load

Emotional load

Variety inyour work


Independencein your work



Participation



Involvement in theorganisation

Recuperation needs

worrying

Pace and amount of work

0

Mental load 0,72

0

Emotional load

0,521,52

0

Variety in your work

0,720,751,07

0

Opportunitiesto learn

0,84

1 1,040,26

0


0,710,891,080,150,15 0


10,261,5 0,5

70,80,72 0


1,050,491,590,560,750,680,13 0

Participation 0,830,66

0,40,160,260,96 0,91

0


2,271,981,452,12,23 2,12,18 2,3

92,21

0


1,730,842,151,11,261,240,69 0,5

51,422,85

0


1,450,461,750,690,770,770,75 0,5

20,852,780,71

0

Recuperation needs

0,992,030,571,651,551,552,04 2,1

51,511,262,712,32 0

worrying 1,332,2 0,9 1,5

41,391,392,09 2,0

91,192,242,612,01 1,0

10

blue collar minus managers

Pace and amountof

Mental load

Emotional load

Variety in your

Opportunities to

Independencein

Relations with colle

Relations with

Participation

Uncertainty abou

Pleasurein your

Involvement in the

Recuperation needs


work work learn

yourwork

agues direct boss

t the future

work organisation

Pace andamount of work

0

Mental load

0,3 0

Emotional load

-0,02 0,07 0

Variety in your work

-0,09 0,08-0,02 0


-0,17 0,11-0,11 0,01 0


-0,15 0,11-0,17 0,02 -0,01

0 0


-0,06 -0,03 0,12 0,1 0,13 0,12 0


0,07 -0,07 0,05 0,06 0,12 0,11 0,05 0

Participation

-0,15 0,63 0,8 0,05 0,03 0,03 0,16 0,15 0


-1,15 -0,24 -0,2 -0,41

-0,45

-0,4 -0,34 -0,4-0,44 0

Pleasurein your work

-0,03 -0,13 0,07 0,09 0,15 0,12 -0,06 0,02 0,19 -0,4 0


-0,22 -0,31-0,14 -0,1 -0,07

-0,1 -0,1 -0,1 0,02 -0,54

0,11

Recuperation

0,07 0,15-0,04 -0,06

-0,09

-0,1 0,09 0,02-0,19 0,01 0,03-0,17 0


needsworrying 0,17 0,34 0,07 0,21 0,12 0,19 0,32 0,28 0,13 -

0,110,32 0,16-0,12

Annex 6 Output Holland-Flanders

Iteration History

Iteration

NormalizedRaw

Stress

Improvement

0 .18481 .0594 .12542 .0417 .01773 .0341 .00764 .0293 .00485 .0260 .00336 .0234 .00257 .0214 .00208 .0197 .00179 .0183 .0014

10 .0173 .001111 .0164 .000812 .0158 .000613 .0153 .000514 .0149 .000415 .0146 .000316 .0144 .000217 .0142 .000218 .0140 .000219 .0139 .000120 .0138 .000121 .0137 .0001

a Stress of initial configuration: simplex start.


b The iteration process has stopped because Improvement has become less than the convergence criterion.

Final CoordinatesDimension

1 2Pace and amount of

work-.481 .670

Emotional load .664 .806Variety in your work-.848 -.755


-1.315

-.986

Independence in yourwork

-1.051

-.381

Relations withcolleagues

.399 -.148

Relations with directboss

.249 -.286

Participation -1.538

-.399

Pleasure in your work1.567 -.416Involvement in the

organisation .233 -1.519

Recuperation needs1.521 1.137worrying .600 2.275


2.01.00.0-1.0-2.0

3

2

1

0

-1

-2

Piekeren

Herstelbehoefte

Betrokkenheid bij de

Plezier in het werkInspraakrelatie met directe Relatie met collega'

Zelfstandigheid in h

LeermogelijkhedenAfwisseling in het w

Emotionele belastingwerktempo en hoeveel

Dimension Weights

Dimension 1

.6.5.4.3.2.10.0-.1

Dime

nsio

n 2

.5

.4

.3

.2

.1

0.0

-.1

SRC_2

SRC_1


Annex 7 Structure of Questionnaire

Scales Factor 1

Factor2

Factor 3Factor4

Factor 5Factor6

Pace and amount of work

Psy . Demand

Mental load Psy . Demand

Emotional load Psy .


DemandVariety in your work Variet

yOpportunities to learn

Variety

Independence in yourwork

Pos to control


Soc-Org

Relations with your direct boss

Soc-Org

Participation Pos to control


Soc-Org


Satisfaction

Involvement in the organisation Satisfaction

Recuperation needs TensionConcerns Tension

The structure of the questionnaire on experiencing and assessing stress at work under review :...

Documents

Transcript of The structure of the questionnaire on experiencing and assessing stress at work under review :...