The validity of Mobley's (1977) model of employee turnover

34
ORGANIZATIONALBEHAVIORAND HUMAN PERFORMANCE 34, 141-174 (1984) The Validity of Mobley's (1977) Model of Employee Turnover PETER W. HOM, RODGER W. GRIFFETH, AND C. LOUISE SELLARO Kent State University Unlike earlier tests of an oversimplified version of this model, the validity of W. Mobley's (1977, Journal of Applied Psychology, 62, 237-240) original turnover model was fully investigated. Constructs that were neglected in prior studies were assessed and previously examined constructs were operational- ized with more reliable measures. Measures of all constructs in MoNey's model were obtained from a survey of 192 hospital employees. Turnover data were collected a year following survey administration. Following the theoret- ical causal ordering of Mobley's constructs, each construct was regressed on all causally prior constructs. In general, each construct was accurately pre- dicted by the linear combination of predictors representing its causal deter- minants. In the majority of instances, the best predictor of a model construct was the construct's immediate causal antecedent. Further, an alternative model was evaluated and compared with Mobley's model using path analysis. Employee turnover has been one of the most widely investigated or- ganizational phenomenon. A voluminous literature on turnover has been generated (see reviews by Bluedorn, 1982b; Mobley, 1982; Mobley, Grif- feth, Hand, & Meglino, 1979; Muchinsky & Turtle, 1979; Price, 1977). However, this research has been predominantly bivariate, static, non- cumulative, and based on no or incomplete conceptual models (Mobley & Meglino, 1979). Such atheoretical and simplistic research may explain the generally low predictions of employee termination. For example, ob- served correlations between job dissatisfaction--the most often used predictor--and turnover seldom exceed .40 (Locke, 1976). Responding to the poor prediction of turnover and the failure of traditional research to advance an understanding of the psychology of the withdrawal process, Mobley (1977) developed a model to explain the process by which a dissatisfied employee arrives at a decision to leave the organization. Specifically, Mobley identified several intervening variables that me- This study was supported by a Kent State University research grant awarded to the senior author and the Department of Administrative Sciences. We express our appreciation to Buddy Myers, our chairman, for his generous financial support of this project. We also express our thanks to Robert Boduch and Paul Antolik for scheduling the survey sessions and collecting data from the hospital's personnel files and to Arlene Darrah for typing innumerable versions of this manuscript. Requests for reprints should be addressed to Peter Horn or Rodger Griffeth, Department of Administrative Sciences, Kent State University, Kent, OH 44242. 141 0030-5073/84 $3.00 Copyright © 1984 by Academic Press, Inc. All rightsof reproductionin any formreserved.

Transcript of The validity of Mobley's (1977) model of employee turnover

ORGANIZATIONAL BEHAVIOR AND HUMAN PERFORMANCE 34, 141-174 (1984)

The Validity of Mobley's (1977) Model of Employee Turnover

PETER W. H O M , RODGER W. GRIFFETH, AND C. LOUISE SELLARO

Kent State University

Unlike earlier tests of an oversimplified version of this model, the validity of W. Mobley's (1977, Journal of Applied Psychology, 62, 237-240) original turnover model was fully investigated. Constructs that were neglected in prior studies were assessed and previously examined constructs were operational- ized with more reliable measures. Measures of all constructs in MoNey's model were obtained from a survey of 192 hospital employees. Turnover data were collected a year following survey administration. Following the theoret- ical causal ordering of Mobley's constructs, each construct was regressed on all causally prior constructs. In general, each construct was accurately pre- dicted by the linear combination of predictors representing its causal deter- minants. In the majority of instances, the best predictor of a model construct was the construct ' s immediate causal antecedent. Further, an alternative model was evaluated and compared with Mobley's model using path analysis.

Employee turnover has been one of the most widely investigated or- ganizational phenomenon. A voluminous literature on turnover has been generated (see reviews by Bluedorn, 1982b; Mobley, 1982; Mobley, Grif- feth, Hand, & Meglino, 1979; Muchinsky & Turtle, 1979; Price, 1977). However, this research has been predominantly bivariate, static, non- cumulative, and based on no or incomplete conceptual models (Mobley & Meglino, 1979). Such atheoretical and simplistic research may explain the generally low predictions of employee termination. For example, ob- served correlations between job dissatisfaction--the most often used predictor--and turnover seldom exceed .40 (Locke, 1976). Responding to the poor prediction of turnover and the failure of traditional research to advance an understanding of the psychology of the withdrawal process, Mobley (1977) developed a model to explain the process by which a dissatisfied employee arrives at a decision to leave the organization.

Specifically, Mobley identified several intervening variables that me-

This study was supported by a Kent State University research grant awarded to the senior author and the Department of Administrative Sciences. We express our appreciation to Buddy Myers, our chairman, for his generous financial support of this project. We also express our thanks to Robert Boduch and Paul Antolik for scheduling the survey sessions and collecting data from the hospital 's personnel files and to Arlene Darrah for typing innumerable versions of this manuscript. Requests for reprints should be addressed to Peter Horn or Rodger Griffeth, Department of Administrative Sciences, Kent State University, Kent, OH 44242.

141

0030-5073/84 $3.00 Copyright © 1984 by Academic Press, Inc. All rights of reproduction in any form reserved.

142 HOM, GRIFFETH, AND SELLARO

diate the effect of job satisfaction on resignation. The first component in this model is employee evaluation of the existing job, the result of which is job satisfaction or dissatisfaction. Mobley suggested several models that might explain how dissatisfaction arises, but he favored no particular satisfaction model. Dissatisfaction, however created, stimulates thoughts of quitting, which, in turn, results in an evaluation of the expected utility of searching for alternatives (and the costs of quitting). If the dissatisfied employee believes an acceptable alternative can be found and perceives that the costs of quitting are not prohibitive, the employee then decides to look elsewhere for work. If that search succeeds, the job seeker then evaluates the available employment alternatives and compares them with the present job. If the comparison favors an alternative, the employee decides to resign and eventually leaves the job.

A simplified version of Mobley's (1977) model was tested by Mobley, Horner, and HoUingsworth (1978). Their abbreviated version hypothe- sized job satisfaction to influence thoughts of quitting, intention to search, and intention to quit. Thoughts of quitting was predicted to directly affect intention to search, which, in turn, was predicted to directly affect inten- tion to quit. Further, the probability of finding an acceptable alternative was proposed to affect intentions to search and to quit. They found va- lidity for this reduced model, although probability of finding an acceptable alternative had no direct effects on search and quitting intentions.

Miller, Katerberg, and Hulin (1979), Bannister and Griffeth (1980), Coverdale and Terborg (1980), Griffeth (1979), and Mowday, Koberg, and McArthur (1980) have since found additional support for the reduced form of Mobley's 1977 model. Moreover, the Miller et al. results suggest that the Mobley et al. (1978) version predicted turnover more accurately than did other approaches (cf. Horn, Katerberg, & Hulin, 1979).

Rather than merely replicate another test of the oversimplified version, the present study more fully investigates the validity of Mobley's (1977) original model. Constructs that were neglected in previous studies are examined. Prior studies have used the "probability of finding acceptable alternatives" as a substitute summarizing those components in the model referring to the search and evaluation of alternatives. Yet this surrogate variable failed to demonstrate its hypothesized effects, possibly because it is an inadequate, deficient representation of these numerous, distinct constructs (Miller et al., 1979; Mobley et al., 1978). Instead, this study directly assesses evaluation of the expected utility of searching and the costs of quitting, search for alternatives, evaluation of alternatives, and comparison of alternatives with the present job. Further, constructs that were examined in past tests of Mobley's model are operationalized with greater precision here, perhaps improving model validity as a conse- quence. Thoughts of quitting, intention to search, and intention to quit

EMPLOYEE TURNOVER 143

are each measured with multiple items, unlike earlier research which has relied on less reliable one-item measures (Miller et al., 1979; Mobley, 1982).

Recently, Mobley, Griffeth, Hand, and Meglino (1979) proposed a more comprehensive model that considers organizational, individual, and economic-labor market determinants of the turnover decision pro- cess depicted in the Mobley 1977 model. This latest model is an expanded version of the earlier model (Mobley, 1982). Mobley, Griffeth, Hand, and Meglino's operational definitions of the three central constructs in this 1979 model (job satisfaction, attraction-expected utility of the present job, and attraction-expected utility of alternatives) are used to opera- tionalize job satisfaction, evaluation of alternatives, and comparison of alternatives with the present job in the present test of Mobley's 1977 model. Consequently, evidence generated in this complete test of Mob- ley's earliest model may prove relevant for determining the validity of portions of the more elaborate 1979 version.

Besides testing the complete 1977 model, this study also explores the addition of a new construct to Mobley's model. Most turnover studies have overlooked individual nonwork influences on turnover, with most research centered on family responsibilities (Mobley, 1982; Mowday, Porter, & Steers, 1982; Steers & Mowday, 1981). Social determinants of turnover are proposed in recent turnover models: transfer of spouse (Mobley, 1977); centrality of nonwork values and nonwork consequences of quitting (Mobley, 1982; Mobley, Griffeth, Hand, & Meglino, 1979); and membership integration and kinship responsibility (Bluedorn, 1982a; Price & Mueller, 1981). An alternative social determinant is Fishbein's referent expectations (Fishbein & Ajzen, 1975, 1981). Employee percep- tion of the expectations of significant others (including referents other than family members) for him or her to seek alternative employment or to quit the present work role may be more immediate and proximal de- terminants of job seeking and quitting behaviors, respectively, thantra- ditional antecedents such as marital status, co-worker friendships, or family size (cf. Arnold & Feldman, •982; Price & Mueller, 1981). These traditional social factors represent more distal causes of turnover, al- though they may underlie the referents' behavioral prescription. For ex- ample, the employee's spouse may expect the employee to quit because the spouse is being transferred to another geographic location. Supporting this hypothesis are findings by Horn et al. (1979), Horn and Hulin (1981), Newman (1974), Parker and Dyer (1976), and Stahl and McNichols (1981) suggesting that perceived normative prescriptions exert stronger and more consistent effects on turnover than do the causally remote social antecedents of turnover (for a review of their turnover effects, see Mobley, 1982; Muchinsky & Tuttle, 1979; Porter & Steers, 1973).

144 HOM, GRIFFETH, AND SELLARO

In summary, the purpose of the present investigation is threefold. First, the present study is designed to test the complete and original Mobley (1977) model. Constructs overlooked in existing studies are assessed and previously examined constructs are measured with greater reliability. Second, parts of the expanded Mobley, Griffeth, Hand, and Meglino (1979) model are tested by relying on its definitions of their central con- structs to operationalize constructs in the 1977 model. Finally, this study considers the addition to Mobley's basic model of the perceived social pressure from referent others placed on the employee to perform with- drawal actions.

METHOD

Procedure

A survey was administered to 192 employees in a large hospital in a Northwest Pennsylvania city at the beginning of the year. Small groups of employees were surveyed on company time. Respondents were asked to identify themselves on the questionnaire in order to match survey responses to subsequent turnover. Researchers guaranteed confiden- tiality of individual responses and stressed the voluntariness of partici- pation.

Respondent Characteristics

Males constituted 21% and Caucasians 95% of the participants. Av- erage age was 31 years, and average tenure was 5.1 years. The average number of dependents (counting the respondent) was 1.94, and 52% were currently married. In this sample, 66% had formal education beyond high school, 90% were full-time employees, and 35% were nurses.

Measurement o f Mobley' s Model

Evaluation o f existing job. Several theories are available to explain job satisfaction: met expectations (Porter & Steers, 1973); contribution/in- ducement ratio (March & Simon, 1958); va lue-percep t discrepancy (Locke, 1976); and expectancy (Vroom, 1964) theories. Mobley (1977) cited no particular theory of satisfaction to represent evaluation of the existing job. He recommended comparat ive studies be undertaken, testing the relative validity of these satisfaction models. Two satisfaction theories were examined.

Porter and Steers' (1973) met expectation model proposes that dissat- isfaction results from the job 's failure to meet the employee's initial job expectations. Each respondent was asked to recall his or her initial ex- pectations about the job before entering the organization and to compare expectations about 46 job outcomes with the actual work experience. The

EMPLOYEE TURNOVER 145

respondent rated the extent to which experience about these job out- comes has been worse than, better than, or about what he or she expected from the job. Similar retrospective measures have been used by Arnold and Feldman (1982) and Michaels and Spector (1982); their one-item mea- sures moderately correlated with satisfaction.

Equity theory states that dissatisfaction occurs when an employee com- pares the ratio of his or her organizational rewards to personal contri- butions to the job with the ratio of rewards to contributions of some referent and finds the ratios different (Adams, 1965). To test this model, Dittrich and Carrell's (1979) Organizational Fairness Questionnaire was used to assess five dimensions of perceived unfairness: pay relative to co-workers (oL = .76, eleven items); pace of work activity (e~ = .78, seven items); pay relative to others' pay outside the organization (c~ = .66, four items); administration of raises and promotions (t~ = .84, five items); and supervisory discipline (o~ = .75, four items). These scales have been found to be strongly correlated with job satisfaction and several scales have been significantly related to absenteeism and turnover (see Dittrich & Carrell, 1979). The five scales were summed to represent overall per- ception of inequity.

Job satisfaction. Some early tests of MoNey's model used satisfaction measures which tapped additional constructs besides affect toward the job. For example, Coverdale and Terborg (1980) used the Hoppock scale which includes an item reflecting thoughts of quitting. Following MoNey, Griffeth, Hand, and Meglino's (1979) conceptualization of satisfaction as the purely affective reaction to the job, Dunham and Herman's (1975) Faces scale was used to precisely assess affect toward five major aspects of the job (work, co-workers, supervision, pay, promotions). This scale has demonstrated convergent and discriminant validity with other well- established satisfaction measures (see Dunham, Smith, & Blackburn, 1977). Also, Miller et al. (1970) used the GM Faces in their test of Mob- ley's model. Further, the sum of the five Faces scales correlated .59 (p < .05) with the Hoppock scale in the present sample. The five scales were summed to reflect overall job satisfaction.

Thoughts o f quitting. Using a Likert-type response format, respondents responded to "I think often about quitting my job ." They also answered "How often do you think of quitting your job" by estimating the fre- quency of this thought. This second item resembles the Miller et al. (1979) and Coverdale and Terborg's (1980) one-item measures of thoughts of quitting.

To allow for the possibility that some employees consider nonwork alternatives to their jobs, all questions dealing with alternatives were phrased in terms of some "alternative role (an activity or job other than the present job) ." The alternative role was broadened to encompass non-

146 HOM, GRIFFETH, AND SELLARO

work alternatives because an earlier analysis of exit interviews with em- ployees resigning the year prior to this study revealed a substantial number of these leavers had chosen nonwork roles. Moreover, a sizable proportion of the sample were nurses, and several studies (Brief, 1976; Watson, 1979; Weisman, 1982) reported that nurses often quit their jobs for nonw0rk roles--ei ther to resume their formal education or to assume primary family responsibilities. Further, Mobley's (1977) model itself con- siders one type of nonwork alternative to the present job: withdrawal from the labor market.

Expected utility of searching and cost of quitting. Attitude toward the act of searching was measured to represent the expected utility of searching. According to Fishbein and Ajzen's (1975) theory, the attitude toward searching for alternatives should be equivalent to the perceived consequences of searching (e.g., chances of finding an alternative, costs of search), which Mobley (1977) conceptualizes as the expected utility of searching. The respondent evaluated "For me, searching for an alterna- tive role to my present job during this year is" on three semantic differ- ential scales (awful-nice, pleasant-unpleasant, bad-good; a = .91).

Attitude toward the act of quitting (during this year) was similarly mea- sured with the same semantic differential scales (cx = .90). This measure represents both the perceived costs and benefits of quitting the job (Hom & Hulin, 1981). Hom and Hulin (1981) found that this measure correlated .51 (p < .05) with a direct measure of the perceived consequences of quitting the present job. Corroborating Mobley's (1977) theory, this atti- tude-toward-act measure was correlated with the perception of the costs of quitting ( "How costly would it be for you to quit your present job" ; r = - .22 , p < .05), ease of mobility ("It is easy for me to leave my present job" ; r = .38, p < .05), and personal investment in the job ("My investment in my job is too great for me to think of quitting"; r = - .40, p < .05) (cf. Baysinger & Mobley, 1982; Farrell & Rusbult, 1981). Both attitudinal measures were combined; they correlated .73 (p < .05) in this sample.

Intention to search for alternatives. The respondent rated "Within this year, I intend to search for an alternative role to my present job" and "What are the chances that you will search for an alternative role (an- other activity or job) to your present job during this year?" The second item closely corresponds to Coverdale and Terborg's (1980) single-item measure of search intention.

Search for alternatives. Respondents indicated if they were actively seeking an alternative role, how much time they spent looking, how often they searched for an alternative, and how much effort they expended in their search. Similar self-report measures of search activity have been used in other studies of job search. For example, Wahba (1980; cited in

EMPLOYEE TURNOVER 147

Bluedorn, 1982b) asked respondents if they were currently looking for a job and Bluedorn (1982a) had respondents report on how many times they looked for another job during the last 3 months. Also, Black (1981) measured job search by asking the worker if he or she had undertaken any explicit search behavior (e.g., filing a job application). Further, Barron and Mellow (1979) had individuals estimate the number of hours per week (during a 4-week period) they spent looking for work.

Evaluation of alternative. Respondents first identified the alternative role (another job, homemaker, student, and so forth) that they would most likely pursue if they should quit. Next, MoNey, Griffeth, Hand, and Meglino's (1979) expectancy formulation of this construct (cf. MoNey, Hand, Baker, & Meglino, 1979) was followed. Using Matsui and Ikeda's (1976) procedure, respondents listed five important role outcomes that they could possibly receive from this alternative role. Eliciting idiosyn- cratic outcomes should result in higher validity for the expectancy model than using a standard set of role outcomes (Horn, 1980; Parker & Dyer, 1976). Respondents then rated the attractiveness of these outcomes using a five-point unipolar undesirable-desirable scale. (Outcome valence is best defined in terms of desirability (Ilgen, Nebeker, & Pritchard, 1981; Schwab, Olian-Gottlieb, & Heneman, 1979) and when numerically scaled with only positive numbers (Schwab et al., 1979).) They also estimated the chances that they can receive these outcomes from this alternative role using a 5-point unipolar scale anchored with numerical descriptions of various probability levels (cf. MoNey, Hand, Baker, & Meglino, 1979). The desirability ratings were multiplied by the corresponding likelihood ratings and the five products were summed. This sum was next multiplied by the expectancy that the respondent could obtain this alternative role; this subjective probability was rated with a 5-point scale having numerical probability statements as anchors (cf. MoNey, Hand, Baker, & Meglino, 1979).

This broadened operational definition of the alternative role to embrace nonwork alternatives may enhance model validity. Hom (1980) found that a within-person version of the expectancy model (contrasting the relative instrumentalities of the present job and an alternative job for various job outcomes)- -or "comparison of alternatives with the present job" in MoNey's model--predicted turnover worse for those employees consid- ering nonwork alternatives to their present work roles. In other words, a narrow conceptualization that all departing employees choose alterna- tive sources of employment might lower model validity.

Comparison o f alternative to the present job. An attraction-expected utility index for the present job similar to the index for the alternative was constructed (Mobley, Griffeth, Hand, & Meglino, 1979). The re- spondent listed five personally important job outcomes that are consid-

148 HOM, GRIFFETH, AND SELLARO

ered when making a choice between jobs. Then she or he rated the extent to which the present job was instrumental for these outcomes and their desirabilities. Next, the valence ratings were multiplied by the instru- mentality ratings and the five products were summed. This sum was then multiplied by the expectancy that the employee can remain in the present job.

Mobley, Griffeth, Hand, and Meglino (1979) conceived the attraction of the present job to differ from job satisfaction. An employee may be dissatisfied with his or her current job duties but nevertheless remains in the job because the job is expected to improve or lead to more satisfying organizational roles in the future (e.g., promotion, transfer, graduation from apprenticeship) (Baysinger & Mobley, 1982; Mobley, 1982). Job at- traction is thus future oriented whereas satisfaction is present oriented (Mobley, Griffeth, Hand, & Meglino, 1979). Consistent with Mobley, Grif- feth, Hand, and Meglino's (1979) theory, this job attraction index was correlated with the perceived chances that the employee will be promoted this year (r = . 11, p < . 10) and the perceived relevance of the present position to the employee's career plans (r = .34, p < .01) (Graen & Ginsburgh, 1977), two additional questions asked in the questionnaire. This index was subtracted from the attraction index for the alternative role. If the resulting difference is positive, then the employee is motivated to quit (cf. Parker & Dyer, 1976; Schneider, 1976).

Intention to quit. Respondents answered two kinds of questions: one kind regarding occupational resignation and the other kind regarding or- ganizational resignation. They were asked if they intended to leave their profession at or before the end of the year and if they intended to remain in their profession but leave the hospital at or before the end of the year. There were two questionnaire items per type.

This study therefore examined the ability of the model to predict both occupational and organizational resignation. Measuring intention to leave the occupation is important because occupational withdrawal may occur in this sample, 35% of which were nurses. National estimates of RNs practicing in their profession range from 70% (Fralic, 1980) to 80% (Weisman, 1982), indicating many nurses leave the nursing profession. Although some attrition is temporary (e.g., to care for preschool children; Gaertner, 1981), attrition from the nursing profession is largely perma- nent. A longitudinal study of the nursing career found 40% of RNs either not working or employed in a nonnursing occupation 10 years after grad- uation from baccalaureate nursing programs (National League for Nursing, 1979).

Turnover. Turnover data were collected from hospital personnel rec- ords at the end of the year. Since Mobley's (1977) model concerns vol- untary turnover, it was essential to determine whether the turnover that

EMPLOYEE TURNOVER 149

occurred was voluntary or involuntary. This determination was made from confidential exit interviews the personnel department conducted with departing employees. Nine employees in the sample had quit and all were classified as voluntary cases (reasons cited for leaving: furthering education; obtaining alternative employment; leaving the area). Only 136 respondents provided complete data on all Mobley measures; of these 136, eight respondents had resigned. The behavioral base rate was thus 5.9% for the prediction of turnover by the full model. Turnover was coded 1 for quitters and 0 for stayers.

Extension of Mobley's Model

To assess the subjective norm, the employee's perception of the overall social pressure placed on him or her to perform a withdrawal act was measured (Fishbein & Ajzen, 1975). The respondent rated "People who are important to me and whose opinions I value think I should search for an alternative role (an activity other than my present job) during this year" using a 7-point unlikely-likely scale. Similarly, "People who are important to me and whose opinions I value think I should quit my present job at or before the end of this year" was rated using an identical response scale.

It must be noted that social determinants of turnover can theoretically be represented in existing constructs in MoNey, Griffeth, Hand, and Meg- lino's (1979) model: centrality of nonwork values and roles (MoNey, 1982; MoNey, Griffeth, Hand, & Meglino, 1979) and nonwork consequences of quitting (MoNey, Griffeth, Hand, & Meglino, 1979). An employee for whom nonwork values are central (e.g., family orientation) may quit for another job that does not interfere with the attainment of these values or for a full-time nonwork role (e.g., full-time homemaker) if no extant po- sition of employment permits satisfaction of these central values (i.e., the nonwork role is more desirable than any work role, Mobley, 1982). Further, nonwork consequences of quitting might include beneficial or harmful effects on the employee's referents of her or his turnover (such as the employee spending more time in family activities and functions; cf. Hom, 1979). However, these influences are already incorporated in the present operationalizations of the costs of quitting (the attitude-to- ward-quitting measure reflects both the perceived work and nonwork consequences of quitting; cf. Horn, 1979) and of the evaluation of the alternative (generalized here to include perceived attraction of nonwork roles).

Nonetheless, Fishbein's subjective norm should not be redundant with any of MoNey's constructs nor with the present operationalizations of MoNey's constructs. Fishbein and Ajzen (Ajzen & Fishbein, 1980; Fish- bein & Ajzen, 1975, 1981) distinguish between the perceived social con-

150 HOM, GRIFFETH, AND SELLARO

sequences of an action (belief that performing the act harms or benefits a referent or elicits a certain reaction from a referent) and their own conceptualization of the perceived behavioral prescriptions of relevant referents. In their theory, behavioral beliefs, which may include beliefs about a referent's response to the behavior's enactment or about the behavioral effects on a referent, determine attitude toward the act. In comparison, beliefs that specfic referents think the individual should (or should not) perform the behavior represent normative beliefs, which un- derlie the subjective norm. For example, an employee may quit in order to assume greater responsibility for child rearing (reducing this burden for the spouse is one outcome of quitting) or to accommodate the spouse's career. In contrast, the spouse may think the employee should continue working in the present job. Accordingly, a referent might benefit by the employee's resignation (a nonwork consequence of quitting that can con- ceivably be represented in Mobley's model), but that referent may pre- scribe that the employee remain in the job (a turnover determinant that is excluded from Mobley's theory).

Statistical Analyses

Hierarchical regression analyses. Following the Mobley et al. (1978) procedure, each construct at a particular causal stage in the model was regressed on all constructs located in prior stages. Unlike their method, the present study emphasized interpretation of the increased R 2 that re- suited from including a particular predictor in the regression equation as an indication of its importance because of the problems of multicolli- nearity and instability of regression coefficients (Miller et al., 1979). Two standards were used to decide empirical support for the model. First, the amount of unique criterion variance explained by the construct that theo- retically directly precedes the criterion (this contribution should also be statistically significant) should be larger than the unique contributions to criterion explanation made by the causally distant predictors. (This stan- dard was not applied to the prediction of job satisfaction since Mobley's model makes no prediction about the relative importance of different causes of satisfaction.) Second, the criterion should be accurately (and significantly) predicted by the combined predictors that represent its theoretical antecedents.

Hierarchical regression analyses estimated the contribution of Fish- bein's subjective norm to the explanatory power of Mobley's model. The subjective norm to search for alternatives was added to that regression equation consisting of the theoretical precursors of search intention and the additional criterion variance it explained was ascertained. Similarly, the independent contribution to the prediction of quitting intention made by the subjective norm to quit was assessed by its inclusion in the regres-

EMPLOYEE TURNOVER 151

sion equation comprising the Mobley predictors of quitting intention. Fi- nally, the two subjective norm measures were individually inserted in the regression equation (containing all components in Mobley's model) de- signed to predict turnover. Their unique contributions to turnover pre- diction were separately estimated. The subjective norm measures were thus used to predict only search and quitting intentions and turnover and not other Mobley constructs. This procedure was followed because Fish- bein and Ajzen (1975) argued that subjective norm is a major determinant of intention and behavior (the other major determinant being attitude toward the act).

Path analysis. As an additional test of Mobley's model, path analysis was performed. Path analysis assumes a recursive model, in which all causal linkages run in only one direction and is inappropriate for testing a nonrecursive model which proposes instantaneous reciprocal causation between variables (Asher, 1976; Duncan, 1975; Kerlinger & Pedhazur, 1973). Mobley's model may appear to be a nonrecursive model, but it is best described as a cyclical recursive model (cf. James, Mulaik, & Brett, 1982) that includes feedback loops in which a variable (e.g., job satisfac- tion) causally prior to another variable (e.g., search for alternatives) may be affected at a later time by that variable. For example, if the search for alternatives (initially caused by dissatisfaction) is unsuccessful, the present job may be reevaluated and job dissatisfaction reversed (Mobley, 1982). Although variables must be measured at different points in time to examine this model's cyclical causation (Baysinger & Mobley, 1982; Bill- ings & Wroten, 1978; Duncan, 1975), a simpler recursive version of the model may nonetheless be tested with path analysis for data collected at a single point in time.

First, the path model based on Mobley's (1977) model is directly tested. Path coefficients for this path model are estimated using standardized regression coefficients. Each criterion is regressed only on those predic- tors Mobley theorized as having direct causal effects on that criterion. To test this model against data, two criteria for goodness of fit are used. Using Heise's (1969) matrix algebraic operations, the path coefficients are used to recompute the correlations among all variables. If the dis- crepancies between reproduced and observed correlations are small (dif- fering no more than .05; Billings & Wroten, 1978; Jermier & Schriesheim, 1978), then the model fits the data. Further, a large-sample ×2 test (L; Kim & Kohout, 1975) is calculated, comparing the amount of unexplained criterion variance for the restricted model (which specifies the absence of certain causal paths) to the amount of unexplained variance for the unrestricted model (which has direct paths from each "earlier" variable to each "later" variable). A nonsignificant L value indicates that the restricted model (e.g., Mobley's model) explains similar amounts of cri-

152 HOM, GRIFFETH, AND SELLARO

terion variance as does the full-scale recursive model (in which each criterion is regressed on all causally prior variables).

Should the path model derived from Mobley's model fail to adequately fit the data, alternative path models are explored following Heise's (1969) theory-trimming procedure (cf. Bluedorn, 1982a; Michaels & Spector, 1982). In this procedure, a full-scale recursive model is examined and paths are deleted if their path coefficients are not statistically significant (Billings & Wroten, 1978). Regression analyses are repeated with each criterion regressed only on the significant predictors, reestimating the path coefficients for paths retained in this "trimmed" model. The ability of this more parsimonious path model to accurately reproduce the original correlations and to explain as much criterion variance as does the full- scale model is then ascertained.

Once a path model is identified that reasonably fits the data, direct and indirect causal effects (Alwin & Hauser, 1975) of each causal variable (Z) on an endogenous variable (Y) are computed by "decomposing" their correlation (ryz). This is achieved by algebraically expanding the following equation:

ryz = ~Pyxi rxiz i

where Pyxi is the path parameter from another causal variable X i to de- pendent variable Y, rx. z is the correlation between Xi and Z, and the set of X i variables are all other causes of Y (Kenny, 1979). An indirect e fec t is an effect of a causal variable on an endogenous variable that is trans- mitted through some other intervening variable(s) (Alwin & Hauser, 1975). This indirect effect was calculated by multiplying the path coeffi- cients of the causal paths (linking the cause and effect variables through a chain of mediating variables) making up a particular causal pathway.

HIERARCHICAL REGRESSION ANALYSES

Hierarchical Regression Analytic Results

Table 1 presents the correlations between components of Mobley 's model. There were five significant predictors of turnover: thoughts of quitting (r = .23); expected utility of search and of quitting (r = .28); intention to search (r = .3 i); search behavior (r = .30); and intention to quit (r = .24). Quitting intention was not the best predictor of turnover, although its correlation with turnover was not statistically different from the correlations of the four other significant predictors.

Table 2 presents the regression analyses testing Mobley's model. Con- structs exemplifying the evaluation of the job in toto accurately predicted job satisfaction (R = .76, p ~< .05). Perceived inequity (AR z = .081, p ~<

T A B L E 1 INTERCORRELATIONS AMONG MOBLEY MODEL COMPONENTS

Variable 1 2 3 4 5 6 7 8 9 l 0 11

1. Met expecta t ions (.92) - - .70 - . 5 7 - . 3 4 - . 0 7 - . 3 3 - , 4 1 - . 4 1 - . 2 3 - . 12 2. Inequity - . 5 8 * (.71) - . 6 4 .44 ,26 .05 .28 .32 .32 .18 .09 3. Job satisfaction .70* - . 6 4 * (.67) - . 6 4 - . 3 8 - . 0 8 - . 4 3 - . 4 7 - . 4 7 - . 2 6 - . 1 3 4. Thoughts of quitting - . 5 7 * .46* - . 6 4 * (.87) .59 .12 .39 .68 .71 .39 .21 5. Expec ted utility o f

search and quitting - . 4 0 * ,30* - . 3 9 * .59* (.85) .20 .44 .44 .63 .35 ,28 6. Evaluat ion of al ternative - .01 - .03 .05 .04 ,20* - - ,77 .24 .19 .11 .06 7. Compar i son of al ternative

with job .36* .26* - .35* .35* .44* .73* - - .45 .42 .23 .14 8. Intent ion to quit - . 4 4 * ,35* - . 5 2 * .66* ,53* .12 .43* (.75) ,73 .40 .18 9. Intent ion to search - . 4 3 * .30* - . 4 8 * .70* .67* .17" .41" .75* (.91) .55 .25

10. Search for al ternatives - . 3 2 * .29* - . 3 2 * ,42* ,35" .07 .27* .44* .55* (.78) .30 11. Turnover - . 0 8 .12 - , 0 8 .23* .28* .09 .16" .24* .3l* .30* - -

0

0 <

Note. Observed correlat ions are shown below the diagonal, Reproduced correlat ions based on the pa th model are shown above the diagonal . Internal cons is tency reliabilities are shown in pa ren theses on the main diagonal.

* p ~< .05.

k~ tao

154 HOM, GRIFFETH, AND SELLARO

.05) and met expectations (&R 2 = . 168, p ~< .05) had independent effects on satisfaction. Thoughts of quitting was accurately predicted (R = .66, p ~< .05) and its immediate causal antecedent, job satisfaction, made the largest unique contribution to its prediction (2tR 2 = .088, p ~< .05). The expected utility of search and of quitting was predicted with accuracy (R = .60, p ~< .05) and, as hypothesized, thoughts of quitting made the greatest unique contribution to its prediction (rid/2 = .174, p ~< .05). Intention to search was strongly predicted (R = .77, p ~< .05) and the expected utility of search and quitting made the largest independent con- tribution to its prediction (2tR 2 = .099, p ~< .05). Consistent with Mobley's model, search for alternatives was predicted with accuracy (R = .57, p ~< .05) and the largest unique contribution to the prediction of this con- struct was made by search intention (ha22 = .120, p ~< .05). Evaluation of the alternative was predicted poorly (R = .27, ns) and the direct an- tecedent, search behavior, made no unique contribution to its prediction. Instead, the causally remote expected utility of search and quitting (&R 2 = .023, p ~< .08) explained marginally significant independent criterion variance. The comparison of the alternative with the present job was strongly predicted (R = .84, p ~< .05) and evaluation of the alternative had the largest unique effect on this criterion (~tR 2 = .457, p ~< .05). Intention to quit was predicted accurately (R = .79, p ~< .05). Although comparison of the alternative with the job accounted for significant unique variance in turnover intention (2tR 2 = .015, p ~< .05), contrary to prediction, search intention accounted for more unique criterion variance (&R 2 = .092, p ~< .05). Finally, turnover was predicted with moderate accuracy (R = .39, p ~< .05). Yet resignation intention made no indepen- dent contribution to its prediction; only search behavior explained sig- nificant independent variance in turnover (Z~d~ 2 = .026, p ~< .05).

Hierarchical regression analyses were performed to determine if in- cluding the subjective norm improves the predictive power of Mobley's model. Adding the subjective norm to search to the regression equation consisting of the Mobley determinants of search intention produced a significant increment in R 2 in the prediction of search intention beyond that already accounted for by Mobley's components (2tR 2 = .047, p .05). However, combining the subjective norm to quit with the Mobley antecedents of quitting intention brought about no additional explained variance in quitting intention (&R 2 = .004, ns) over and above that ex- plained by Mobley's predictors. Including the subjective norm to search in the equation designed to predict turnover generated no significant in- crement in explained turnover variance (&R 2 = .003, ns). On the other hand, the subjective norm to quit did account for significant turnover variance beyond that accounted for by all of Mobley's predictors (zkR 2 = .030, p ~< .05).

T A B L E 2

AMOUNT OF UNIQUE CRITERION VARIANCE EXPLAINED BY PREDICTORS AND THEIR STANDARDIZED REGRESSION COEFFICIENTS

Criteria

Predictor

Expected Evaluation Intention utility of Thoughts

Intention Alternative of Search to search and of Job Turnover to quit vs job alternative behavior search quitting quitting satisfaction

z2kR 2 [3 AR 2 [3 AR 2 I 3 AR 2 [3 A R 2 [3 AR 2 [3 AR 2 [3 AR 2 [3 ,dxR 2 [3

Intention to quit .000 .02 . . . . . . . . . . . . . . . . Comparison of

alternative © with job ,-<

Evaluation of alternative

Search for alternatives

intention to Z search ©

Expected utility ~< of search and of quitting ,010 .14 .001 - . 0 4 .009* .13" .023 .21 .002 .06 .099* .39* . . . . . .

Thoughts of quitting .001 .05 .015" .21" .000 .01 ,006 - . 1 2 .000 .02 .088* .45* .174" .56* . . . .

Job satisfaction ,008 .15 .001 - . 0 6 .017" - . 2 2 * .009 .16 .002 .07 .004 - . 1 0 .001 .05 .088* - . 4 5 * - - - - Evaluation of job:

Inequity .004 .08 .000 .01 .000 .02 .000 - . 0 1 .011 .14 .005 -- .09 .000 .01 .001 .04 .081" - . 3 5 " Met expectations .003 .08 .000 .01 .005 - . 1 0 .001 - . 0 4 .002 - . 0 6 .000 .00 .006 - . 1 1 .024* - . 2 2 ' .168" .50* R .39* .79* .84* .27 .57* .77* .60* .66* .76*

.000 .02 .015" .23* . . . . . . . . . . . . . .

.000 .00 .006 - . 1 3 .457* .70* . . . . . . . . . . . .

.026" .20* .000 .01 .601 .05 .000 - . 0 1 . . . . . . . . . .

.004 .12 .092* .52* .000 .03 .010 .17 .120" .54* . . . . . . . .

* p ~< .05.

156 HOM, GRIFFETH, AND SELLARO

Discussion of Hierarchical Regression Analyses

Mobley's original model received empirical verification from the hier- archical regression analyses, although the evidence was not unequivocal. All model constructs except two (evaluation of the alternative and turn- over) were strongly predicted by combinations of their theoretical causal antecedents. With three exceptions (the predictions of turnover, intention to quit, and evaluation of the alternative), the predictor explaining the most unique criterion variance was that criterion's immediate precursor according to Mobley's theory. Theoretical and methodological problems may account for the less supportive findings from these regression anal- yses.

Mobley's 1977 model assumes that the employee's primary (or sole) alternative to the present position is another full-time job elsewhere. Yet dissatisfied employees considering nonwork roles may forego the search process postulated by Mobley (1977) or undertake different kinds of search activities to find nonwork alternatives. Search activity thus failed to best predict the evaluation of the alternative in this study (see Table 2) because evaluation of a nonwork alternative (this study generalized the operational definition of the alternative to encompass nonwork roles) may not follow search behavior (or our particular empirical realization of this construct which may not have adequately reflected the kind of search activities performed to obtain nonwork roles). Further, these employees may form an intention to quit prior to organizational entry. For example, an individual intends to work in a particular job for only a certain duration in order to accumulate sufficient funds to attend school or a married woman decides in advance that she will eventually resign in order to have children. This may explain the failure of comparison of the alternative with the job to best predict intention to quit (see Table 2); the intention to leave the labor market (and thus a particular job) preceded organiza- tional experience (and therefore before the comparison between the present job with the nonwork alternative can take place). In short, Mob- ley's model may be valid for employees who consider only alternative sources of employment, but be invalid for those employees contemplating permanent or temporary withdrawal from the labor force.

This explanation was tested using moderated regression analyses (Ar- nold, 1982; Peters & Champoux, 1979). Each predictor in a given regres- sion equation (predicting a particular model construct from its supposed determinants) was multiplied by a modera tor - -a dichotomous variable indicating whether the employee's desired alternative was a work or non- work role. The resulting product term was then added to that equation, including the moderator term. If a product term explained significant

EMPLOYEE TURNOVER 157

criterion variance beyond that explained by the predictors, then the type of alternative the employee desired moderated the causal effect of an antecedent (a constitutent making up this significant product term).

Few moderating effects were detected. Of three uncovered significant moderating effects, one product term explained 1% and two product terms 2% additional criterion variance. More importantly, the causal ef- fects of comparison of the alternative with the job on quitting intention and of search behavior on evaluation of the alternative were not mod- erated by the kind of desired alternative. Peripheral attachment to the labor force, therefore, appeared not to be at fault for the failure of Mob- ley's model in these instances. It is worth noting that MoNey (1982; Mobley, Griffeth, Hand, & Meglino, 1979) recently introduced attraction of nonwork roles as an additional turnover determinant. It remains to be determined whether conceptualizing nonwork alternatives separately from work alternatives (perhaps the process of comparing the present job with nonwork alternatives [a career choice] differs from the process of comparing the job with alternative jobs [a job choice]) or subsuming this variable under a comprehensive category, "alternative roles" (as was done in this study), is best for model validity.

A methodological shortcoming may explain the poor predictability of evaluation of the alternative (R = .27, ns) by its supposed precursors-- especially search behavior, the hypothesized direct precursor. This study did not distinguish between alternatives that were obtained directly by some employees through an active search and alternatives that were only being entertained or contemplated by other employees who conducted no search. Search behavior was not the best predictor of the present (expectancy-valence) representation of evaluation of the alternative be- cause some employees were merely describing an alternative that they regarded attractive but that they had not actually sought. Nevertheless, MoNey's model proposed that the alternative(s) being evaluated and compared with the present job is (are) a concrete job offer(s) discovered during a job hunt. To test this reason for the poor predictability of eval- uation of the alternative, an alternate indicant of this construct was ex- amined. This indicant asked respondents to rate the attractiveness of an alternative role that they actually found during a recent search.

The attractiveness rating of this uncovered alternative was unrelated (r = .01, ns) to the expectancy-valence indicant used to test MoNey's model. Moreover, MoNey's theoretical determinants of evaluation of the alternative moderately and significantly predicted the desirability of this acquired alternative (R = .47, p < .05). Further, the best predictor was search behavior, the immediate antecedent of evaluation of the alterna- tive. Search behavior was the sole predictor making a significant inde-

158 HOM, GRIFFETH, AND SELLARO

pendent contribution to the prediction of the evaluation of the discovered alternative (zSd? z = . 12, p < .05).

Similarly, why comparison of the alternative with the job was not the most potent predictor of quitting intention (see Table 2), as dictated by Mobley's theory, may originate from failing to specify that the alternative being compared with the present job be one that was actually searched. The expectancy-valence index representing evaluation of the alternative had been included as a component of the index representing comparison of the alternative with the job. This latter index may thus indicate that an employee considers an alternative to be more desirable than the present job. Yet if that employee was not describing an actual alternative that had been found through search activities (e.g., an actual job offer), but instead an alternative that he or she merely regards as desirable, then the employee may not intend to quit even though the comparison favors some alternative over the present position.

Ambiguity in conceptual and operational definitions of perceived al- ternatives plagues numerous turnover studies. Some studies required re- spondents to estimate the general availability of alternatives in the labor market (Arnold & Feldman, 1982; Martin, 1979; Price & Mueller, 1981) or to describe the overall quality of external alternatives (Bluedorn, 1982a; Farrell & Rusbult, 1981) without making reference to any specific alternative. Further, turnover studies that have identified specific alter- natives and required their descriptions have lacked precision in their mea- surement. That is, the expected value of a general class of alternatives (e.g., alternative employment or full-time student) is usually measured rather than the expected value of a particular member of this broad class (e.g., a specified job or school). For example, Horn (1980), Mobley, Hand, Baker, and Meglino (1979), Parker and Dyer (1976), Schneider (1976), Stahl and McNichols (1981), and Stahl (1980) had military personnel de- scribe a civilian job without specifying (or having them specify) a partic- ular job as their alternative to military service. In addition, these studies failed to specify that the alternatives being portrayed are those actually sought or obtained by the respondents. Such imprecise specification of perceived alternatives may explain their poor contribution to the predic- tion of turnover (cf. Michaels & Spector, 1982; Mobley, 1982). Further turnover research should directly assess employees' perceptions of al- ternative jobs that are direct products of their search efforts (or are re- ceived unsolicited from other employers) and compare the attractiveness of actual job offers with that of the present job (cf. studies of organiza- tional choice [Wanous, Keon, & Latack, 1983], in which respondents describe specific job offers). Operationalizing alternatives more precisely may enhance prediction of turnover as well as provide a more valid test of Mobley's model.

EMPLOYEE TURNOVER 159

PATH ANALYSES

Besides testing Mobley's model using path analysis, an alternative path model was developed and compared with Mobley's formulation. This theoretical alternative was based on the present empirical findings and previous research. The prior hierarchical regression analyses clearly sug- gest that this study did not operationalize the perceived alternative in the way Mobley's theory had intended. Although this study's respondents apparently described an alternative that was neither sought nor found during an active search (it was, however, regarded a desirable alternative that they would conceivably pursue if they should quit), this perceived alternative nevertheless may influence the turnover process. Indeed, in- dividuals may be in a constant state of receptivity to ambient information about alternatives without necessarily being engaged in an active search (Baysinger & Mobley, 1982). This receptivity may be stimulated by a need for information about external alternatives in order to have com- parative standards by which to evaluate the present job (Schneider, 1976). In fact, the "greener grass" phenomenon may be based on the perception of highly visible and attractive alternatives (Bluedorn, 1982b; Schneider, 1976). Also saliency of perceived alternatives may forestall development of behavioral commitment to the organization (Pfeffer & Lawler, 1980).

Moreover, Youngblood, Laughlin, Mobley, and Meglino's (1980) lon- gitudinal study of Marine Corps recruits found that attrites were more attracted to a civilian job than were stayers. Because the freedom and time to search for alternatives are extremely limited in a military insti- tution, especially during recruit training, it is unlikely that these recruits described an alternative job that was obtained during a job search. In fact, a survey administered shortly after their arrival to the recruit depot found that recruits who ultimately left perceived civilian work to be more desirable than the stayers. Measured at organizational entry before any job search can be undertaken, this perceived alternative still influenced training attrition. Therefore, awareness of alternatives, which are not necessarily searched nor obtained by employees, may affect their with- drawal (although the process of civilian withdrawal may differ from the process of military withdrawal). Based on the preceding arguments, this alternative model specifies that the present operationalizations of evalu- ation and comparison of the alternative be placed causally prior to inten- tion to search and search activity; the alternative portrayed was not de- rived from the respondent's search.

This alternative model also suggests a revision of Mobley's model; intention to quit is hypothesized to precede search intention and behavior. Fishbein's (Fishbein & Ajzen, 1975; Fishbein & Jaccard, 1973) concep- tual analysis of the stability of behavioral intention suggests this particular

160 HOM, GRIFFETH, AND SELLARO

causal order. The stability of intention is determined by the extent to which performance of the act is dependent on the prior occurrence of events, some of which are not entirely Under the individual' s control. For example, in carrying out an intention to quit, the employee must first perform other actions (discover job vacancies, apply for specific jobs, pass various selection hurdles, receive attractive job offers). But the greater the number of intervening steps, the stronger the likelihood that completion of (or failure to complete) any single step would result in new information which alters the initial intention (e.g., during the job inter- view, the job seeker learns that the alternative job is not as attractive as originally believed). The original intention to leave (by a specific date) may thus be reversed or postponed until a later date (the employee de- cides to resign the following year) if the expected event (such as receiving acceptable job offers), upon which the original intention is predicated, does not occur. In short, the success of the job search 1 mediates the effect of intention to quit on turnover.

The hypothesis that intention to quit causally precedes the actual search for alternatives has also been advanced elsewhere. Steers and Mowday's (1981) model proposed that intention to quit influences turn- over indirectly (as well as directly) by causing an initial search for better alternative jobs. Bowen (I982) argued that it is counterintuitive to expect employees to know that they have attractive job offers before intending to quit, as Mobley's model seems to imply. Further, Michaels and Spector (1982) suggest that perceived alternative employment acts directly on turnover rather than act through intention to quit. Thus, an employee intending to quit may most likely quit when another job becomes available (i.e., job search follows the intention to quit). Finally, Arnold and Feldman (1982) found that intention to search was more closely related to turnover than was intention to leave. Their revised turnover model hypothesized that organizational commitment (which includes a compo- nent representing intention to quit) influences turnover through intention to search.

In summary, the major difference between the alternative model and Mobley's model is the causal ordering of certain constructs. This com- peting model predicts that evaluation of the alternative, comparison of

1 This argument differs from the argument that the ability to quit moderates the relation between intention to quit and turnover (cf. Bowen, 1982). Unlike this latter argument, the former argument implies that quitting is predominantly under volitional control and that the absence of attractive alternatives does not prevent an employee from quitting. Rather, an individual can always quit but may not want to under these circumstances, Thus, the relationship between the initial intention to quit and turnover may be attentuated because this intention changed in light of an unsuccessful job search (this later intention would be strongly related to turnover ff it could be measured) and not because an unfavorable labor market prevented resignation from occurring.

EMPLOYEE TURNOVER 161

the alternative with the present job, and intention to quit precede inten- tion to search and search activity. Further, intention to quit is theorized to succeed evaluation and comparison of the alternative since perception of available and attractive external alternatives may stimulate the inten- tion to quit.

Path Analytic Results

Mobley's model was first tested by regressing each criterion only on those predictors Mobley specified as having direct causal paths to that criterion. This series of regression analyses generated path coefficients used to reproduce the empirical correlations among the 11 variables (see Table 1) following Heise's (1969) method. The path model based on Mob- ley's theory inaccurately reproduced the original correlations. The av- erage of the absolute differences between the empirical correlations and the correlations implied by this model was .188. Further, 37 of the 54 recomputed correlations 2 differed from the corresponding original cor- relations by more than .05 with 32 discrepancies exceeding. 10. Kim and Kohout's (1975) X 2 test was also significant; L(44) = 87.09 (p < .05). Mobley's model explained significantly less criterion variance than did the full-scale model (see Table 2).

Mobley's theory was also tested using Heise's (1969) theory-trimming procedure. Another path model was empirically developed in which vari- ables were causally ordered according to Mobley's theory, but paths were allowed to exist (and estimated) which were not hypothesized by Mobley (1977). First, the full-scale model was examined (see Table 2) and causal paths with nonsignificant (p > .05)path coefficients were omitted. Regression analyses were then repeated, regressing each criterion only on predictors found to have significant direct effects in the full-scale model. The recalculated path coefficients in this trimmed model were next used to reproduce the observed correlations. For this path model, the average absolute difference between actual and recomputed correla- tions was .055; twenty discrepancies exceeded .05 and thirteen equalled or exceeded . 10 (the largest discrepancy was .20). Further, the ×2 test was nonsignificant, L(39) = -3.58 (us), indicating that this restricted model explained criterion variance as well as did the unrestricted model.

This path model based on Mobley's causal ordering of constructs was compared with a model specifying an alternative order: evaluation and comparison of the alternative and quitting intention precede search in-

2 The correlation between the two purely exogenous variables (met expectations and inequity) is given and cannot be analyzed in terms of the causal components of the path model. Their causes lie outside the causal system (Duncan, 1975; Kerlinger & Pedhazur, 1973) and this correlation therefore cannot be reproduced, which leaves 54 (not 55) corre- lations for reproduction by the path model

T A B L E 3

PATH COEFFICIENTS FROM THE P A T H ANALYSIS OF A REVISED M O B L E Y M O D E L

Cr i t e r i a

E x p e c t e d Eva lua t ion util i ty o f

Sea rch In ten t ion In ten t ion Al te rna t ive of s e a r c h and T h o u g h t s Job Predic tor Turnover b e h a v i o r to sea rch to quit v s j o b a l te rna t ive qui t t ing o f qui t t ing sa t i s fac t ion

Search for O

a l ternat ives Pll,,0 = .23" Intent ion to

s e a r c h P10,9 ~ .55* G') Intention to

quit Pgs = .43" Compar i son o f 0~

al ternat ive with job P87 = .22*

Evalua t ion o f > a l ternat ive P76 = .71"

Expec t ed utility of search and quit t ing Pu.5 = .20* p ~ = .30* P75 = .,18" p ~ = .20*

Thoughts o f

quit t ing P94 = .24* Psa = .59* P54 = .59* O Job

sat isfact ion P73 = - .31" t'43 = - . 4 7 " Inequi ty P32 = - .35* Met

expecta t ions P41 = - .24* P31 = ,50" Dis turbance

term PU,i = .93 Pl0,h = .84 p ~ = .56 PSf = .71 P?e = .54 P~a = .98 PSc = .81 P4b = .75 P3a = .65 R .36* .55* .83* .70* .84* .20* .59* .66* .76*

* p <~ .05 .

EMPLOYEE TURNOVER 163

tention and behavior. This competing model was also developed ac- cording to Heise's (1969) procedure. Each variable was regressed on all other variables considered in this alternative model to be causally prior. All nonsignificant causal paths were deleted from this full-scale model and regression analyses were repeated with the significant predictors. The differences between the observed and the reproduced correlations resulting from the trimmed model averaged .044; 18 discrepancies ex- ceeded .05 and 7 exceeded. 10. The ×2 test was nonsignificant; L(38) = - 5 . 2 4 ( n s ) .

A large discrepancy (. 18), however, was found for this model's repro- duction of the correlation between expected utility of search and quitting and turnover, suggesting that a path was needed between expected utility and turnover (cf. Bohrnstedt & Felson, 1983; Schmitt, Coyle, White, & Rauschenberger, 1978). Although this path is unexpected, it is a plausible one. The expected utility of quitting may be most salient when the em- ployee is confronted with an actual choice between the present job and an alternative. It is most likely then that the employee carefully weighs the cost of quitting and decides whether the alternative job can offset the investments in the present job which are relinquished by resignation. Similarly, Horn (1980) and Parker and Dyer (1976) found that employees quit only if the alternative is much more attractive than the present job, suggesting an inertia factor or costs of quitting which must be surmounted for turnover to occur. Apparently, insufficient justification exists for the employee to resign if the alternative is merely more attractive than the present position. Further, Baysinger and Mobley (1982) suggested that turnover occurs only if the alternative employment is adequately superior to current employment to warrant the costs of quitting and setting up in the new employment. Addition of this path to the alternative model re- suited in an average discrepancy between actual and reproduced corre- lations of .036. Moreover, 13 differences exceeded .05 and only 3 ex- ceeded. 10 (the largest difference was. 13). The X 2 test was nonsignificant, L(37) = -7 .43 (ns), indicating that this trimmed model explained similar amounts of criterion variance as did the full-scale model.

Direct and lndirect Causal Effects of the Path Model

This alternative path model thus provided a reasonable, albeit imper- fect, fit of the data. Adding additional paths would improve reproduction of the correlations. But this procedure was avoided since that would overfit the data and that model was not being subjected to cross-valida- tion. Table 3 shows the path coefficients for this model and Table 1 pre- sents the correlations implied by the model. Figure 1 displays a diagram of the model without causal paths from the residual variables drawn for sake of clarity. Table 4 indicates the direct and summated indirect causal

164 HOM, G R I F F E T H , AND SELLARO

TABLE 4 DmECT AND 1NDIRECT CAUSAL EFFECTS OF EACH VARIABLE IN THE PATH MODEL

Causal effects Total Causal causal

Dependent variable variable Direct Indirect ~ effect

Job satisfaction (x 3)

Thoughts of quitting (x4)

Expected utility of search and quitting (xs)

Evaluation of alternative (x 6)

Comparison of alternative with job (x7)

Intention to quit (x s)

Intention to search (x 9)

Search for alternatives (xl0)

X 1 .50 .00 (0) .50 X 2 --.35 .00 (0) - , 3 5

x 1 - . 2 4 - . 2 4 (1) - , 4 8 x 2 .00 .16 (1) .16 x 3 - . 4 7 .00 (0) - . 4 7

X 1 .00 --.28 (2) --.28 X2 .00 .10 (1) .10 x 3 .00 - . 2 8 (1) - . 2 8 x 4 .59 .00 (0) .59

x 1 .00 - . 0 6 (2) - . 0 6 X 2 ,00 .02 (1) .02 x 3 .00 - . 0 6 (1) - . 0 6 x4 .00 .12 (1) .12 x5 .20 .00 (0) .20

xl .00 - . 2 5 (5) - . 2 5 x 2 .00 .14 (3) .14 x 3 - . 3 1 - . 0 9 (2) - , 4 0 x 4 .00 .19 (2) ,19 x 5 .18 ,14 (1) ,32 x 6 .71 ,00 (0) ,71

x 1 .00 - . 3 3 (7) - . 3 3 x 2 .00 .13 (4) .13 x 3 .00 - . 3 7 (4) - . 3 7 X 4 .59 .04 (2) .63 X 5 .00 .07 (2) .07 X6 .00 .16 (1) .16 X 7 .22 .00 (0) .22

X 1 .00 --.34 (11) --.34 X 2 .00 .12 (6) .12 X 3 .00 --.35 (6) --.35 X 4 .24 ,45 (4) .69 X 5 .30 .03 (2) .33 X 6 .00 ,07 (1) .07 X 7 .00 ,09 (1) .09 X 8 .43 .00 (0) .43

X l .00 --.19 (11) --.19 X 2 .00 ,07 (6) .07 X 3 .00 --.19 (6) --.19 X4 .00 .38 (5) .38

EMPLOYEE TURNOVER

TABLE 4--Continued

165

Dependent variable

Causal effects Total Causal causal

variable Direct Indirect a effect

Turnover (xl0

xs .00 .18 (3) .18 x6 .00 .04 (1) .04 x 7 .00 .05 (1) .05 x 8 .00 .24 (1) .24 x 9 .55 .00 (0) 55

X 1 .00 --.10 (13) --.10 X 2 .00 .04 (7) .04 x 3 .00 - .10 (7) - .10 x 4 .00 .21 (6) .21 x5 .20 .04 (3) .24 X 6 .00 .01 (1) .01 X 7 .00 .01 (1) .01 x 8 .00 .05 (1) .05 x 9 .O0 .13 (1) .13 Xlo .23 .00 (0) .23

Note• x 1 = met expectations, x z = inequity• a Numbers in parentheses indicate the number of different indirect effects of a particular

causal variable on a given dependent variable•

effects o f the causal variables on the dependen t variables. A par t icular causal variable m a y exer t several indirect effects on a given dependen t variable, bu t the fol lowing d i scuss ion is l imited to its meaningful (greater than .05) indirect effects.

Job satisfaction• Me t expec ta t ions (P31 = .50) and perce ived inequi ty (P32 = - . 3 5 ) had direct effects on satisfaction• Dissatisf ied employees had unme t initial j ob expec ta t ions and pe rce ived unfair dis t r ibut ion o f organiza t ional rewards .

Thoughts o f quitting. Cons i s t en t with M o N e y ' s theory, sat isfact ion di- r e c t l y i n f l u e n c e d t h o u g h t s o f qu i t t i ng (P43 = - . 4 7 ) ; d i s s a t i s f a c t i o n a roused these thoughts . C o n t r a r y to M o N e y ' s theory, met expec ta t ions also had a di rect effect (P41 = - . 2 4 ) . This f inding agrees with Horne r , M o N e y , and Meg l ino ' s (1979) study, wh ich found that met expec ta t ions exer ted effects on t u r n o v e r (and its immedia te p recursors ) wh ich were not media ted by sat isfact ion, confl ic t ing with Por te r and Steers ' hypo th - esis (1973). U n m e t expec ta t ions (/943/731 -~ - . 2 4 ) and inequi ty (P43P32 = • 16) also dec rea sed sat isfact ion which in turn led to grea ter thoughts o f quitting.

Expected utility o f search and quitting. As M o N e y (1977) predic ted , only thoughts o f quit t ing had a d i rec t posi t ive impac t on expec ted utility

c~

jvoooofl. Alternatives x?

Expected Comparison of yE,ility of I ~ ~AI tlves l ..... O Search & r75 y with Job

hou his of Intention JExpectations| " P41 ~ ' ~ ~ ~ ~ I u u l t X8 I

I"12~ X l ~ ! ~ ~ P~5~ 8 1 4 tOSearchi°n X_9"lP10 .9 ~-~

FIG. 1. Path analysis of the revised Mobley (1977) turnover model.

E M P L O Y E E T U R N O V E R 167

of search and quitting (P54 = .59). As Mobley hypothesized, met expec- tations enhanced satisfaction, which thereby reduced thoughts of quitting and consequently expected utility of search and quitting (P54P43P31 =

- . 14). Met expectations also diminished thoughts of quitting, which in turn lowered evaluation of search and quitting (P54P41 = - - " 14). Inequity had an indirect effect transmitted through satisfaction and thoughts of q u i t t i n g (P54Pn3P32 = • 10) . Further, thoughts of quitting mediated satis- faction's indirect effect on expected utility (P54P43 = - .28).

Evaluation o f the alternative. Expected utility of search and quitting had a direct effect (/965 = .20); favorable appraisal of performing these withdrawal acts led to greater attraction to the alternative. Satisfaction exerted an indirect effect: it decreased thoughts of quitting, which then decreased expected utility of search and quitting, thus lowering the al- ternative's desirability (P65P54P43 = - - . 0 6 ) . Further, thoughts of quitting increased the evaluation of searching and quitting and accordingly raised the perceived worth of the alternative (P65P54 = • 12).

Comparison of the alternative with the job. Desirability of the alter- native had a direct effect (P76 =-- .71) on comparison of the alternative with the job, but so did satisfaction (/973 --= - - . 3 1) and expected utility of search and quitting (P75 = .18). Therefore, the comparison of an alter- native and the present job is based not only on a review of their varying characteristics, but also on attitudes toward the job and toward seeking alternatives and resignation. Dissatisfied employees who positively eval- uate performing withdrawal actions tend to appraise more favorably the alternative relative to their present job.

Met expectations (P73P31 ----- - - . 16) and inequity (P73P32 = "11) influ- enced comparison of the alternative through their influence on satisfac- tion. Thoughts of quitting had effects on comparison of the alternative transmitted by expected utility of the withdrawal behaviors (,P75P54 = • 11) and by both expected utility and evaluation of the alternative (P76P65P54 = .08). Finally, positively appraising searching and quitting resulted in a higher evaluation of the alternative and a higher comparative judgment of the alternative relative to the present work role (/976065 = . 14).

Intention to quit. Thoughts of quitting (P84 = .59) and comparison of the alternative with the job (P87 = .22) had direct effects. Frequent thoughts of quitting and perceiving some external alternative to be more attractive than the present job resulted in the formation of an intention to quit. The direct effect of thoughts of quitting suggests that this intention can be formed without the individual necessarily first weighing the costs and benefits of searching and of quitting.

Met expectations had effects on quitting intention transmitted through thoughts of quitting (P84P41 = --14) and through both satisfaction and thoughts of quitting (P84P43P31 = - - . 14 ) . Inequity reduced satisfaction,

168 H O M , G R I F F E T H , A N D S E L L A R O

and this in turn enhanced thoughts of quitting and quitting intention (P84Pn3P32 = .10). Dissatisfaction enhanced intention to quit by stimu- lating thoughts of quitting (p8,o43 = - .28) and by increasing the relative attractiveness of the alternative versus the present job (P87P73 = - - . 0 7 ) .

Finally, positive appraisal of the alternative increased its attractiveness relative to the present job, which increased intention to leave (/987/976 = .16).

Intention to search. Intention to quit (P9s = .43), expected utility of search and quitting (P95 = .30), and thoughts of quitting (P94 = .24) di- rectly affected search intention. Thoughts of quitting's direct unmediated effect suggests that some employees decide to look for alternatives without deliberating upon the costs of this search (nor the costs of quit- ting), after they begin thinking of quitting. Although Mobley's model considers the possibility of impulsive quitting, the present result also suggests the possibility of "impulsive" job seeking, not considered in Mobley's model. Consistent with Mobley's theory, an employee favor- ably evaluating searching and quitting has a higher intention to search. This alternative model also proposes that search intention and activity mediate the effect of intention to quit on turnover. Whether or not the intention to quit its carried out depends on the success of this job search.

Met expectations diminished search intention by reducing thoughts of q u i t t i n g (P94P41 = - . 0 6 ) and by reducing both dissatisfaction and thoughts of quitting (P94P43P31 = - - . 0 6 ) . Met expectations also had in- direct effects mediated by thoughts of quitting and quitting intention (P98PsnP41 = - - . 0 6 ) and by satisfaction, thoughts of quitting, and quitting intention (P98P84P43P31 = - - . 0 6 ) . Satisfaction influenced search intention through thoughts of quitting (P94P43 = - - . 11), through thoughts of quitting and attitudes toward search and quitting (.P95P54P43 = -- . 08 ) , and through thoughts of quitting and quitting intention (,O98P84P43 = - - . 12). Expected utility of search and quitting (P95P54 = • 18) and intention to quit (P98PS4 = .25) conveyed indirect effects of thoughts of quitting. Evaluating the alternative favorably increased the relative desirability of the alternative in comparison with the present position and subsequently increased res- ignation and search intentions (P98P87P76 = .07 ) . Finally, when the com- parison favored an alternative over the present job, resignation and search intentions increased as a result (/9981087 = .09 ) .

Search for alternatives. Only search intention (P10,9 = .55) directly affected search behavior. Satisfaction influenced search behavior through thoughts of quitting and search intention (P10,9P94P43 = -- .06) and through thoughts of quitting, quitting and search intentions (Plo,9P98P84P43 = -.07). Thoughts of quitting had effects transmitted via search intention

(P10,9P94 = .13), via expected utility of search and quitting and search intention (/910,9P95P54 = .10 ) , and via intentions to quit and to search

EMPLOYEE TURNOVER 169

(PI0,9P98P84 ~" .14). Positive attitudes toward searching and quitting in- creased search intention and thereby increased search activity (P10,9P95 = . 17). Intention to quit affected search behavior through its effect on search intention (Pl0,gPgs = .24).

Turnover. Search activity (P11,10 = .23) and expected utility of search and quitting (P11,5 = .20) had direct turnover effects. The costs of quitting may be salient during the time when the individual must make a decision to accept or reject an alternative. During this stage in the turnover de- cision process, he or she must decide whether the inducements offered by the alternative could offset the losses incurred by abandoning the present job.

Satisfaction influenced turnover through thoughts of quitting and ex- pected utility of search and quitting (Pll,5P54P43 = -- .06). Thoughts of quitting produced favorable attitudes toward searching and quitting ac- tions and this stimulated turnover (PH,sP54 = • 12). Further, search inten- tion enhanced search activity, leading to greater turnover (Pu,~0Pl0,9 = .13).

Discussion of Path Analytic Results

This alternative model seemed reasonably consistent with the data. It is, however, cautioned that this model was no t cross-validated and ad- ditional models may exist which fit the data equally well. Nevertheless, this model suggests problems in MoNey's (1977) theoretical formulation. MoNey's (1977) model implies that evaluation and comparison of jobs occur after the job search concludes, when the job seeker is reviewing his or her job offers. Research on organizational choice, however, reports that evaluation and comparison of jobs occur before and during the job search process (Lawler, Kuleck, Rhodes, & Sorensen, 1975; Sheridan, Richards, & Slocum, 1975; Vroom, 1966). Job seeking employees may evaluate and compare alternative jobs in order to decide to which em- ployers to submit their applications. While Mobley's model delineates the causal determinants of the initiation (and perhaps magnitude and per- sistence) of search behavior, this model leaves unspecified the determi- nants of the direction of this search activity. The question why some jobs are sought and others not is unanswered. This alternative model suggests that knowledge of specific alternatives prior to an actual search may influence the direction of this search. Additional research is needed on the determinants of an employee's awareness of external alternatives (Baysinger & Mobley, 1982).

Like MoNey's model, this alternative model proposes feedback loops. This model suggests that failure in the search for acceptable alternatives may result in an intention to terminate the search or an intention to remain in the present job (or a decision to defer quitting until a later time when

170 HOM, GRIFFETH, AND SELLARO

a renewed search will succeed). Although not represented in the path diagram (see Fig. 1) since these constructs were not measured, this model also hypothesizes that evaluation and comparison of actual job offers follow the job search, as does Mobley's model. After this search, the employee compares the available alternatives with the present job and decides to make a choice among these jobs. The choice of a particular alternative job then precedes turnover. Should no alternative job offer be sufficiently superior to the present position, the individual initially in- tending to quit may "choose" the present job and not resign.

This model is also more compatible with the possibility that some em- ployees quit without having first searched for alternatives than is Mob- ley's model, although there was no evidence of direct turnover effects for intention to quit and intention to search (i.e., the decision to search may be formed before resignation, but the leaver may decide to carry out this search after leaving). For example, Steers and Mowday (1981) sug- gested that some employees decide to quit even when alternatives are not available. They argued that recent improvements in the social welfare system which economically cushion the unemployed may strengthen this direct turnover effect of intention to quit. Moreover, military personnel may typically leave military duty before finding civilian alternatives.

Further, many employees in certain occupations for which exist a tight labor market may search for alternative jobs after quitting since they can easily obtain alternative jobs. To test this hypothesis, regression analyses were performed separately on two subsamples: nonnurses and nurses (whose services are in high demand because of the nursing shortage (Cun- ningham, 1979; Span, 1981)). Regression of turnover on all predictors for nurses (R = .47, ns) found that quitting intention (ziR 2 = .029, ns) and search intention (ZkR z = .036, ns) seemed to exert stronger direct effects on turnover than did search behavior (fi~R 2 = .025, ns), although no direct effects were significant. Thus, some nurses intending to quit and in- tending to search may quit without having first searched for alternatives. Yet the regression of turnover on all predictors for nonnurses (R = .43, p < .10) found that only search activity (zkR 2 = .041, p ~< .05) had a direct turnover effect, suggesting that for these employees, search activity mediates the effects of search intention (ZkR 2 = .000, ns) and quitting intention (/~R 2 = .005, ns) on turnover. Many quitting nurses therefore may leave without having first acquired alternative jobs. They may seek alternatives after resigning, but make this decision prior to their actual exit from the organization, which may explain the direct turnover effect of search intention. In other words, nurses (and other types of workers in short supply) can "afford" to be temporarily jobless when they quit.

In conclusion, continued research is needed to substantiate the validity of this alternative conceptual model. Although the causal ordering of the

EMPLOYEE TURNOVER 171

theoretical constructs identified by Mobley (1977) has been disputed, these constructs have nonetheless enriched the understanding of the psy- chology of the turnover process. Greater attention is required for further refining the conceptualization and measurement of the basic components of Mobley's (1977) model. Without such attention, neither valid tests of the original 1977 model nor valid tests of the expanded 1979 version are possible.

REFERENCES Adams, J. (1965). Inequity of social exchange. In L. Berkowitz (Ed.), Advances in exper-

imental social psychology. New York: Academic Press. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior.

Englewood Cliffs, NJ: Prentice-Hall. Alwin, D. F., & Hauser, R. M. (1975). The decomposition of effects in path analysis. Amer-

ican Sociological Review, 40, 37-47. Arnold, H. (1982). Moderator variables: A clarification of conceptual, analytic, and psy-

chometric issues. Organizational Behavior and Human Performance, 29, 143-174. Arnold, H., & Feldman, D. (1982). A multivariate analysis of the determinants of job

turnover. Journal of Applied Psychology, 67, 350-360. Asher, H. B. (1976). Causal modeling. Beverly Hills, CA: Sage. Barron, J. M., & Mellow, W. (1979). Search effort in the labor market. The Journal of

Human Resources, 14, 389-404. Baysinger, B. D., & MoNey, W. H. (1982, April). Employee turnover: Individual and or-

ganizational analyses (TR-ONR-6). TX: Texas A&M University. Bannister, B., & Griffeth, R. (1980, November). Extending the Mobley, Homer and Hol-

lingsworth model of turnover model using path analysis. Paper presented at the 12th annual meeting of the American Institute for Decision Sciences, Las Vegas, NV.

Billings, R. S., & Wroten, S. P. (1978). Use of path analysis in industrial/organizational psychology: Criticisms and suggestions. Journal of Applied Psychology, 63, 677-688.

Black, M. (1981). An empirical test of the theory of on-the-job search. The Journal of Human Resources, 16, 129-140.

Bluedorn, A. C. (1982a). A unified model of turnover from organizations. Human Relations, 35, 135-153.

Bluedorn, A. C. (1982b). The theories of turnover: Causes, effects, and meaning. In S. Bacharach (Ed.), Research in the sociology of organizations (Vol. 1). Greenwich, CN: JAI Press.

Bohrnstedt, G., & Felson, R. (1983). Explaining the relations among children's actual and perceived performances and self esteem: A comparison of several causal models. Journal of Personality and Social Psychology, 45, 43-56.

Bowen, D. (1982). Some unintended consequences of intention to quit. Academy of Man- agement Review, 7, 205-211.

Brief, A. (1976). Turnover among hospital nurses: A suggested model. Journal of Nursing Administration, October, 55-58.

Coverdale, S., & Terborg, J. (1980, August). A re-examination of the Mobley, Homer and Hollingsworth model of turnover: A useful replication. Paper presented at the 40th annual meeting of the Academy of Management, Detroit, MI.

Cunningham, L. (1979). Nursing shortage, Yes! American Journal of Nursing, 3, 469-480. Dittrich, J., & Carrell, M. (1979). Organizational equity perceptions, employee job satis-

faction, and departmental absence and turnover rates. Organizational Behavior and Human Performance, 24, 29-40.

172 HOM, GRIFFETH, AND SELLARO

Duncan, O. D. (1975). Introduction to structural equation models. New York: Academic Press.

Dunham, R., & Herman, J. (1975). Development of female faces scale for measuring job satisfaction. Journal of Applied Psychology, 60, 629-31.

Dunham, R. B., Smith, E J., & Blackburn, R. S. (1977). Validation of the index of orga- nization reactions with the JDI, the MSQ, and Faces Scales. Academy of Management Journal, 20, 420-432.

Farrell, D., & Rusbult, C. (1981). Exchange variables as predictors of job satisfaction, job commitment, and turnover: The impact of rewards, costs, alternatives, and invest- ments. Organizational Behavior and Human Performance, 28, 78-95.

Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior. Reading, MA: Addison-Wesley.

Fishbein, M., & Ajzen, I. (1981). On construct validity: A critique of Minieard and Cohen's paper. Journal of Experimental Social Psychology, 17, 340-350.

Fishbein, M., & Jaccard, J. (1973). Theoretical and methodological considerations in the prediction of family planning intentions and behavior. Representative Research in So- cial Psychology, 4, 37-51.

Fralic, M. (1980, May 1). Nursing shortage: Coping today and planning for tomorrow. Hospitals, pp. 65-67.

Gaertner, K. (1981). Work attitudes, non-work situation, and labor force participation of nurses. Unpublished manuscript, 1981 (available from Department of Organizational Behavior, The Weatherhead School of Management, Case Western Reserve University, Cleveland, OH).

Graen, G., & Ginsburgh, S. (1977). Job resignation as a function of role orientation and leader acceptance: A longitudinal investigation or organizational assimilation. Orga- nizational Behavior and Human Performance, 19, 1-17.

Griffeth, R. (1979, September). Attrition after recruit training: Test of a conceptual model. Paper presented at the 87tb annual meeting of the American Psychological Association, New York.

Heise, D. R. (1969). Problems in path analysis and causal inference. In E. E Borgatta & G. W. Bahrnstedt (Eds.), Sociological methodology 1969, San Franciso: Jossey-Bass.

Horn, E W. (1979). A comparative examination of the relative effectiveness of several different approaches in the prediction of withdrawal behavior. Unpublished doctoral dissertation, University of Illinois, Champaign.

Horn, E (1980). Expectancy predictions of reenlistment in the National Guard. Journal of Vocational Behavior, 16, 235-248.

Horn, E, & Hulin, C. (1981). A competitive test of the prediction of reenlistment by several models. Journal of Applied Psychology, 66, 23-39.

Horn, E, Katerberg, R., & Hulin, C. (1979). A comparative examination of three approaches to the prediction of turnover. Journal of Applied Psychology, 64, 280-290.

Homer, S., Mobley, W., & Meglino, B. (1979, September). An experimental evaluation of the effects of a realistic job preview on marine recruit affect, intentions, and behavior (TR-9). Columbia, SC: Center for Management and Organizational Research, Univer- sity of South Carolina.

Ilgen, D., Nebeker, D., & Pritchard, R. (1981). Expectancy theory measures: An empirical comparison in an experimental simulation. Organizational Behavior and Human Per- formance, 28, 189-223.

James, L. R., Mulaik, S. A., & Brett, J. M. (1982). Causal analysis: Assumptions, models and data. Beverly Hills, CA: Sage.

Jermier, J., & Schriesheim, C. (1978). Causal analysis in the organizational sciences and

EMPLOYEE TURNOVER 173

alternative model specification and evaluation. Academy of Management Review, 3, 326-337.

Kenny, D. (1979). Correlation and causality. New York: Wiley. Kerlinger, E, & Pedhazur, E. (1973) Multiple regression in behavioral research. New York:

Holt, Rinehart & Winston. Kim, I., & Kohout, E K. (1975). Special topics in general linear models. In N. H. Nie,

C. H. Hull, J. C. Jenkins, K. Steinbrenner, & D. H. Bent (Eds.), Statistical package for the social science. New York: McGraw-Hill.

Lawler, E., Kuleck, W., Rhodes, J., & Sorensen, J. (1975). Job choice and post decision dissonance. Organizational Behavior and Human Performance, 13, 133-145.

Locke, E. (1976). The nature and causes of job satisfaction. In M. Dunnette (Ed.), Hand- book of industrial and organizational psychology. Chicago: Rand McNally.

March, J., & Simon, H. (1958). Organizations. New York: Wiley. Martin, T. M., Jr. (1979). A contextual model of employee turnover intentions. Academy

of Management Journal, 22, 313-324. Matsui, T., & Ikeda, H. (1976). Effectiveness of self-generated outcomes for improving

predictions in expectancy theory research. Organizational Behavior and Human Per- formance, 17, 289-298.

Michaels, C., & Spector, P. (1982). Causes of employee turnover: A test of the Mobley, Griffeth, Hand, and Meglino model. Journal of Applied Psychology, 67, 53-59.

Miller, H., Katerberg, R., & Hulin, C. (1979). Evaluation of the Mobley, Homer, and Hollingsworth model of employee turnover. Journal of Applied Psychology, 64, 509-517.

Mobley, W. (1977). Intermediate linkages in the relationship between job satisfaction and employee turnover. Journal of Applied Psychology, 62, 237-240.

Mobley, W. (1982) Employee turnover: Causes, consequences and control. Reading, MA: Addison-Wesley.

Mobley, W., Griffeth, R., Hand, H., & Meglino, B. (1979). Review and conceptual analysis of the employee turnover process. Psychological Bulletin, 86, 493-522.

Mobley, W., Hand, H., Baker, R., & Meglino, B. (1979). Conceptual and empirical analysis of military recruit training attrition. Journal of Applied Psychology, 64, 10-18.

Mobley, W., Homer, S., & Hollingsworth, A. (1978). An evaluation of precursors of hospital turnover. Journal of Applied Psychology, 63, 408-414.

Mobley, W., & Meglino, B. (1979, August). Toward further understanding of the employee turnover process. Paper presented at the national meeting of the Academy of Manage- ment, Atlanta, GA.

Mowday, R., Koberg, C., & McArthur, A. (1980, August). The psychology of the with- drawal process: A cross-validation of Mobley's intermediate linkages model of turn- over. Paper presented at the 40th annual meeting of the Academy of Management, De- troit, MI.

Mowday, R., Porter, L., & Steers, R. (1982). Employee-organization linkages. New York: Academic Press.

Muchinsky, P., & Tuttle, M. (1979). Employee turnover: An empirical and methodological assessment. Journal of Vocational Behavior. 14, 43-77.

National League for Nursing. (1979, May/June). Nurse career-pattern study. Hospital Topics, pp. 5-6, 8-9.

Newman, J. (1974). Predicting absenteeism and turnover: A field comparison of Fishbein's model and traditional job attitude measure. Journal o f Applied Psychology, 59, 610-615.

Parker, D., & Dyer, L. (1976). Expectancy theory as a within-person behavioral choice

174 HOM, GRIFFETH, AND SELLARO

model: An empirical test of some conceptual and methodological refinements. Orga- nizational Behavior and Human Performance, 17, 97-117.

Peters, W., & Champoux, J. (1979). The role and analysis of moderator variables in orga- nizational research. In R. Mowday & R. Steers (Eds.), Research in organizations. Santa Monica, CA: Goodyear.

Pfeffer, J., & Lawler, J. (1980). Effects of job alternatives, extrinsic rewards, and behavioral commitment on attitude toward the organization: A field test of the insufficient justi- fication paradigm. Administrative Science Quarterly, 25, 38-56.

Porter, L., & Steers, R. (1973). Organization, work and personal factors in employee turn- over and absenteeism. Psychological Bulletin, 80, 151-176.

Price, J. (1977). The study of turnover. Ames: Iowa State Univ. Press. Price, J., & MueUer, C. (1981). A causal model of turnover for nurses. Academy of Man-

agement Journal, 24, 543-565. Schmitt, N., Coyle, B., White, J., & Rauschenberger, J. (1978). Background, needs, job

perceptions, and job satisfaction: A causal model. Personnel Psychology, 3 I, 889-901. Schneider, J. (1976). The "greener grass" phenomenon: Differential effects of a work con-

text alternative on organizational participation and withdrawal intentions. Organiza- tional Behavior and Human Performance, 16, 308-333.

Schwab, D., Olian-Gottlieb, J., & Heneman, H. (1979). Between-subjects expectancy theory research: A statistical review of studies predicting effect and performance. Psy- chological Bulletin, 86, 139-147.

Sheridan, J., Richards, M., & Slocum, J. (1975). Comparative analysis of expectancy and heuristic models of decision behavior. Journal of Applied Psychology, 60, 361-368.

Span, E (1981, February). Where have all the nurses gone? The New York Times Magazine, pp. 70-71; 78-79; 96; 100-101.

Stahl, M. (1980, November). Anticipated satisfaction, current satisfaction and turnover intentions. Paper presented at the 12th annual meeting of the American Institute for Decision Sciences, Las Vegas, NV.

Stahl, M., & McNichols, C. (1981). Modeling turnover behavior: A comparative analysis of three multivariate turnover models. Paper presented at the 13th annual meeting of the American Institute for Decision Sciences, Boston, MA.

Steers, R., & Mowday, R. (1981). Employee turnover and postdecision accommodation processes. In L. Cummings & B. Staw (Eds.), Research in organizational behavior (Vol. 3). Greenwich, CT: JAI Press.

Vroom, V. (1964). Work and motivation. New York: Wiley. Vroom, V. (1966). Organizational choice: A study of pre and post decision processing.

Organization Behavior and Human Performance, 1, 212-225. Wanous, J., Keon, T., & Latack, J. (1983). Expectancy theory and occupational/organiza-

tional choices: A review and test. Organizational Behavior and Human Performance, 32, 66-86.

Waters, L. K., Roach, D., & Waters, C. W. (1976). Estimate of future tenure, satisfaction, and geographical variables as predictors of termination. Personnel Psychology, 29, 57-60.

Watson, L. (1979, October). Keeping qualified nurses. Supervisor Nurse, pp. 29-34. Weisman, C. (1982). Recruit from within: Hospital nurse retention in the 1980s. Journal of

Nursing Administration, May, 24-31. Youngblood, S., Laughlin, J., Mobley, W., & Meglino, B. (1980, February). A longitudinal

analysis of military recruit attrition: The first 25 months (TR- 11). Columbia, SC: Center for Management and Organizational Research, University of South Carolina.

RECEIVED: October 18, 1982.