Sie sind auf Seite 1von 13

Journal of Applied Psychology Copyright 1986 by the American Psychological Association, Inc.

1986, №1.71. No. 2, 219-231 0021-9010/86/M0.75

Antecedents and Consequences of Satisfaction and Commitment in


Turnover Models: A Reanalysis Using Latent Variable
Structural Equation Methods

Larry J. Williams John T. Hazer


Indiana University School of Business Indiana University-Purdue University at Indianapolis

Recent models of organizational commitment (e.g., Morris & Sherman, 1981; Steers, 1977; Stevens,
Beyer, & Trice, 1978) and turnover (e.g., Mobley, Grifleth, Hand, & Meglino, 1979; Steers & Mowday,
1981) were reviewed. This review indicated that the causal relation between job satisfaction and
organizational commitment had been overlooked and that attempts to identify the antecedents of
these variables had suffered from conceptual and methodological limitations. To examine these two
issues, structural equation methodology was used to reanalyze data from Michaels and Spector (1982)
and Bluedorn (1982). Four causal models were examined. Across both samples, support was obtained
for relations between personal/organizational characteristics and job satisfaction, and between satis-
faction and commitment. Commitment was also an important determinant of turnover intentions.
The implications of the findings and methodology are considered.

Turnover has been a focus of investigation by several disciphnes reanalysis of existing published data. This reanalysis will also
(e.g., psychology, sociology, economics) for a number of years. examine the manner in which other variables influence satisfac-
Many of these efforts have centered around the development of tion and commitment and will be based on a method for incor-
conceptual models of the turnover process and subsequent em- porating measurement concerns into the evaluation of path
pirical validation of the models. Although these models have analysis models.
diverse origins, several have postulated job satisfaction and or-
ganizational commitment to be antecedents of turnover. In fact, Organizational Commitment Models
a recent review (Clegg, 1983) suggested that in turnover research,
most of which has looked at the role of affect on subsequent As noted elsewhere (Mowday, Steers, & Porter, 1979), research
behavior, satisfaction and commitment have been the most fre- on commitment has been based on two perspectives. One view-
quently investigated components of affect. However, because most point has conceptualized commitment in terms of overt indi-
models of turnover have included only one of these two variables cators (e.g., behaviors that exceed role expectations); alternatively,
(Steers & Mowday, 1981), the causal relation between these two the construct has been approached from an attitude perspective,
determinants has generally been overlooked. Furthermore, at- representing the relative strength of involvement and identifi-
tempts to investigate their antecedents have suffered from con- cation of an individual with an organization. This latter per-
ceptual and methodological limitations. Thus, the present article spective has guided the empirical efforts relevant to the present
will briefly review the concept of organizational commitment study, in which commitment is characterized by factors relating
and develop the hypothesis that job satisfaction is a causal an- to belief in the organization's goals and values, willingness to
tecedent. Also, several turnover models and studies will be ex- extend effort for the organization, and desire to remain in the
amined, focusing on the satisfaction-commitment relation and organization (e.g., Morrow, 1983). As an attitude, commitment
the role of other determinants. Next, a review of the biasing is distinguished from job satisfaction in that the former is an
effects of measurement error in path analysis models will be affective response to the whole organization, whereas the latter
presented. This methodological problem has been overlooked by represents an affective response to specific aspects of the job.
researchers working in this area. Finally, an empirical test of the Variables in turnover research that are conceptually similar to
satisfaction-commitment relation will be conducted through a organizational commitment include job attachment and job
commitment. Job attachment (Koch & Steers, 1978) describes
an attitudinal response to a job, characterized by a congruence
between one's real and ideal jobs, an identification with one's
This article is based on the first author's master's thesis, which was chosen occupation, and a reluctance to seek alternative employ-
conducted at Indiana University-Purdue University at Indianapolis.
ment. Alternatively, job commitment (Farrell & Rusbult, 1981;
We thank committee members Jerome Busemeyer and Robert Bringle
Rusbult & Farrell, 1983) refers to the extent to which an employee
and three anonymous reviewers for their helpful comments. We also thank
perceives that he or she is connected to a job and involves feelings
Dr. Bluedorn for making his data available to us and Mary Post for her
assistance in manuscript preparation. of psychological attachment, independent of affect.
Correspondence concerning this article should be addressed to Larry There have been a few studies relevant to establishing the re-
J. Williams, Department of Management, Indiana University School of lation between job satisfaction and organizational commitment.
Business, Bloomington, Indiana 47405. Porter, Steers, Mowday, and Boulian (1974) suggested that sat-

219
220 LARRY J. WILLIAMS AND JOHN T. HATER

isfaction and commitment were related but distinguishable at- cedents of organizational outcomes. Koch and Steers (1978) pre-
titudes. More specifically, they proposed that commitment rep- sented a view of job attachment as an intervening variable be-
resents a global evaluative link between the employee and the tween job satisfaction and overt behavior like turnover. Also, in
organization, with job satisfaction among commitment's specific the tradition of exchange theory (cf. Homans, 1961), Farrell and
components. Porter et al. (1974) further offered that satisfaction Rusbult (1981) described job satisfaction as an intervening vari-
would be associated with aspects of the work environment and able to be a function of rewards and costs associated with the
thus would develop more quickly than commitment, which would job, with job commitment resulting from job satisfaction, in-
require a worker to make a more global assessment of his or her vestments, and alternatives. Rusbult and Farrell (1983) found
relationship to the organization. Whereas the instability and rapid support for their model in a longitudinal investigation with nurses
formation of satisfaction would suggest it as a cause of commit- and accountants and, thus, identified satisfaction as an antecedent
ment, rather than vice versa, Bateman and Strasser (1984) note of commitment. However, as noted earlier, the Rusbult and Far-
that the validity of this perspective has not been established. rell (1983) model is based on a behaviorally oriented commit-
Additional support for the satisfaction-commitment relation ment, whereas the present research takes an attitudinal focus
comes from research examining determinants of commitment. that includes behavioral intentions.
A model developed by Steers (1977) describes the antecedents
and outcomes of commitment. According to this model, three
Turnover Models
main categories of variables influence commitment: Personal
characteristics include variables such as need for achievement, In spite of the models and research just presented, the idea
age, and education. Work experiences, the second category of that job satisfaction and organizational commitment are causally
influences, describe socializing forces that have an impact on related has not been incorporated in recent turnover models,
attachments formed with the organization. This category includes and the roles of these variables as mediators has not been ade-
experiences related to group attitudes toward the organization quately investigated. Studies have either (a) failed to include both
and perceptions of personal investment in and worth to the or- satisfaction and commitment, (b) overlooked their causal rela-
ganization. The third category, job characteristics, includes job tionship, or (c) failed to adequately examine their antecedents.
challenge, opportunities for social interaction, and feedback. The conceptual turnover model that has received the most at-
More important, although Steers did not specifically include job tention in the psychological literature was provided by Mobley
satisfaction as an antecedent, he did propose that it would prob- (1977) and has undergone several revisions and evaluations (e.g.,
ably influence commitment more than would job characteristics Miller, Katerberg, & Hulin, 1979; Mobley, Homer, & Holling-
(Steers, 1977, p. 47). Steers also discussed the role of exchange sworth, 1978). Although this model considers organizational
processes, through which individuals satisfy their needs/desires, commitment as an attitude to be related to satisfaction and at-
use their skills in the work environment, and thus develop at- traction to the present job, no clear causal relation is hypothesized
tachment and commitment to the organization. between satisfaction and commitment.
Similarly, Stevens, Beyer, and Trice (1978) developed a role- A multidisciplinary model of voluntary employee turnover
taking model in their attempt to clarify psychological and struc- has been offered by Muchinsky and Morrow (1980). They also
tural determinants of the managerial commitment process. Their propose three major classes of determinants, including individual
model identified three groups of factors as antecedents of orga- and economic opportunity and work-related factors. This last
nizational commitment: personal attributes (e.g., age, sex, edu- category includes factors such as job characteristics, job satis-
cation), organizational factors (e.g., size, centralization of au- faction, and organizational commitment. However, no causal re-
thority), and role-related factors (e.g., work overload, skill of lations were suggested between variables within a class of deter-
subordinates). These role variables were presumed to represent minants, such as job satisfaction and organizational commitment.
the more dynamic aspects of the job situation that influence Similarly, Steers and Mowday (1981) have attempted to develop
commitment. According to Stevens et al. (1978), the common a comprehensive model of the processes leading to voluntary
element among these diverse antecedents is an exchange process, termination, as well as the subsequent accommodation to the
through which perceived benefits and costs are evaluated and turnover decision. In this model, affective responses (job satis-
attachment and commitment subsequently develops. Although faction, organizational commitment, job involvement) are pro-
job satisfaction was not included in their analyses, they suggested posed to result from three major factors: (a) job expectations;
that it might be an important predictor of commitment (Stevens (b) organizational characteristics and experiences; and (c) job
etal., 1978, p. 391). performance level. Although this model includes more than one
Morris and Sherman (1981) investigated a multivariate pre- job attitude, the relations between the affective responses, par-
dictive model of organizational commitment and focused on the ticularly satisfaction and commitment, are not further described.
role of job level and organizational differences. The antecedents Although research has traditionally and currently focused on
in their model included variables from each of Steer's (1977) comparing the relative contributions of job satisfaction, orga-
categories. It should be noted that although job satisfaction was nizational commitment, and other factors in predicting turnover
not included in these three studies of commitment, the anteced- (e.g., Horn & Hulin, 1981; Horn, Katerberg, & Hulin, 1979;
ents that were supported have also been established as important Mitchel, 1981; Parasuraman, 1982), there has been considerable
determinants of satisfaction (e.g., Mitchell, 1979). interest lately in the use of causal analysis to investigate models
Further support for the job satisfaction-organizational com- of turnover. Three studies included job satisfaction but excluded
mitment relation comes from the previously mentioned research, organizational commitment. Bluedorn (1979) investigated a
which includes job attachment and job commitment as ante- model of turnover in military organizations that included struc-
TURNOVER MODELS 221

tural and environmental variables as antecedents of job satisfac- associated with the work environment and representing personal
tion which, as an intervening variable, was further described as characteristics influence commitment indirectly via satisfaction.
a determinant of turnover intentions. Similar models have been In other words, satisfaction was proposed to be an intervening
evaluated by Martin (1979) and Price and Mueller (1981). Thus, variable between environmental and personal characteristics and
satisfaction played a critical role in these three studies as an organizational commitment. These hypotheses follow from the
intervening variable, but organizational commitment was not conceptual models of commitment previously reviewed and the
included. established relations between environmental and personal char-
More recently, Arnold and Feldman (1982) have proposed a acteristics and satisfaction. Also examined was the role of com-
model of turnover based on path analysis with data collected mitment as an intervening variable between satisfaction and
from accountants. According to this model, age, job satisfaction, turnover intentions. Finally, whereas the preceding discussion
and organizational commitment influence the intention to search and examples provide support for satisfaction leading to com-
for alternatives. Furthermore, intention to search combines with mitment, others have suggested a reverse sequence (Bateman &
tenure and perceived job security to cause turnover. Whereas Strasser, 1984; Weiner & Vardi, 1980). This perspective, as sum-
satisfaction and commitment were both included in this study marized by Bateman and Strasser (1984), suggests that "a cog-
as intervening variables, their causal relation was not considered, nitive outlook such as commitment is rationalized by subsequent
and possible antecedents were not included. attitudes of job satisfaction" (p. 97). Thus, the causal relation
Of more relevance to the present research are the recent studies from commitment to satisfaction was additionally investigated,
by Michaels and Spector (1982) and Bluedorn (1982). Michaels along with the associated hypotheses that commitment mediates
and Spector used path analysis to estimate the strengths of re- the impact of work/personal characteristics on satisfaction and
lations between several variables in the Mobley et al. model that satisfaction intervenes between commitment and turnover
(1979), with data from employees of a community mental health intentions. To provide a strong test of these hypotheses, data
center. After the elimination of nonsignificant variables, their from two previously described turnover path analysis studies were
analyses indicated that the exogenous variables of perceived job reanalyzed. In this process, competing models were examined
characteristics, leadership consideration, and age lead to both through a sequence of model comparisons as described by James,
job satisfaction and organizational commitment, whereas pre- Mulaik, and Brett (1982), while controlling for the biasing effects
employment expectations lead only to job satisfaction. Also, job of measurement error. The goal was to use confirmatory analytic
satisfaction and organizational commitment were demonstrated techniques to provide a strong test of the proposed mediating
to lead to intentions to quit which, in turn, lead to turnover. As hypotheses (James & Brett, 1984).
with the study by Arnold and Feldman, the satisfaction-com-
mitment relation was not examined. Furthermore, the possibility Measurement Error in Path Analysis
that job satisfaction played a mediating role between the effects The biasing effects of method variance and random measure-
of the exogenous variables and organizational commitment was ment error in path analysis models have been noted for industrial/
not considered. organizational psychologists by Billings and Wroten (1978). Both
Finally, Bluedorn (1982) developed a model of turnover that Bagozzi (1980) and Kenny (1979) have provided detailed alge-
was a unification of the Mobley (1977) and Steers (1977) models braic descriptions of how measurement error effects the estimated
previously described. Most relevant for the present research were causal parameters obtained from regression analyses, and James
the findings that equity and age were determinants of both sat- et al. (1982) have given an empirical demonstration. To sum-
isfaction and commitment, instrumental information was marize Kenny (1979), measurement error in an endogenous (de-
causally related to satisfaction, and routinization was an ante- pendent) variable attenuates the standardized path coefficient
cedent of commitment. Furthermore, satisfaction was shown to but not the unstandardized coefficient, whereas the effects of
have an influence on commitment, but not on intent to leave. error in two or more exogenous (independent, predictor) variables
Thus, while the Bluedorn (1982) study is noteworthy for including is difficult to predict. Such error can attenuate estimates of coef-
a satisfaction-commitment relation, it and the Michaels and ficients, make the estimates of zero coefficients nonzero, or yield
Spector (1982) research failed to adequately examine the me- coefficients with the wrong sign.
diating role of these variables with respect to their antecedents. Several solutions have been proposed for the problems that
In other words, the question remains whether personal and work measurement error presents for regression-based analyses. First,
environment characteristics are antecedents of both satisfaction it has been suggested that the correlations be corrected for at-
and commitment, or instead whether satisfaction plays a me- tenuation prior to the use of multiple regression to estimate pa-
diating role between the antecedents and commitment. rameters. However, as noted by Kenny (1979), the major problem
with this approach is that the resulting parameter estimates can-
The Present Study not be tested for significance in the usual way. Kenny (1979) has
also described the use of instrumental variables for the mea-
The present research attempted to improve on the previously surement error problem. This technique requires an exogenous
discussed conceptual limitations of existing path analysis studies variable that is related to the other exogenous variables in the
of turnover. More specifically, it was hypothesized that personal model, but not to the endogenous variable. As such, this tech-
and work characteristics have no direct effect on turnover inten- nique is useful only under very restrictive conditions. A third
tions: Their impact is only through their effect on satisfaction strategy involves the use of latent variable models, most often
and commitment. It was also proposed that satisfaction is a de- with multiple indicators (e.g., Bentler, 1980; Schmitt & Bedeian,
terminant of organizational commitment and that other variables 1982).
222 LARRY J. WILLIAMS AND JOHN T. HAZER

Although the use of latent variable models with multiple in- able or when the decision is made not to use them. This approach
dicators represents an advancement in social science method- would provide an alternative to ignoring the biasing effects of
ology, this technique is not without its limitations. Schmitt and measurement error. Such a method has been developed (James
Bedeian (1982) point out that the mathematics of latent variable et al., 1982; Kenny, 1979) and was used in the present research
analysis with multiple indicators are complex, and obtaining to evaluate the previously described hypotheses regarding the
enough indicators for identification may not always be feasible. relations between job satisfaction and organizational commit-
In other words, there may not be enough information in the ment and their antecedents.
covariances/correlations to allow for unique parameter estimates.
They also mention the distorted estimates of structural param-
eters that can result from the simultaneous solution of the mea- Method
surement and structural models, a problem similar to interpre-
tational confounding as discussed by Bagozzi (1980). Finally, The data used in the present research came from the studies by Michaels
and Specter (1982) and Bluedorn (1982) and will be subsequently referred
there may be cases where well-established scales are available to
to as Sample 1 and Sample 2. The variables used from each sample are
measure a construct, and the decision is made not to break the
defined in Table 1. Note that the exogenous work environment variables
scale into several indicators.
differ in Samples 1 and 2. The selection of three different variables to
Thus, a technique is needed that would allow for the evaluation
describe the work environment in Sample 2 allowed for a more rigorous
of latent variable models when multiple indicators are not avail- test of the mediating role of satisfaction. As previously mentioned, data
from Sample 1 was collected at a community mental health center, whereas
Sample 2 data came from a large insurance company. Further information
on these two samples can be found in the original articles.
Table 1
Variables From Each Sample Used in Analyses

Model Development
Variable Definition

Sample 1 Any causal model can be described graphically and in matrix equation
form. The graphic representation of a general model for Sample 1 is
Pre-employment The extent to which the job is what
presented in Figure 1; the matrix form appears in Table 2. Although
expectations it was expected when it was taken.
three of the exogenous variables are different, a model with the same
Perceived job characteristics A worker's perception of the amount
of variety, autonomy, feedback, causal parameters for Sample 2 appears in Figure 2 and Table 3. These
task identity, and task two models contain all of the causal parameters of theoretical interest.
significance. Thus, these general models served as a basis for the development of the
Leadership consideration The consideration of a supervisor for other models that were examined with each sample. However, it is im-
subordinate's feelings, problems, portant to note that they could not be evaluated due to identification
and input for decisions, as problems. More specifically, /94 and 05 could not be estimated at the
assessed by subordinate. same time. The symbols used to represent the factor loadings (X), mea-
Age An individual's chronological age.
surement error (5, f I, path coefficients (y, /)), and residuals (0 reflect the
Job satisfaction The affective orientation of
standard notation used in the latent variable modeling literature. With
individuals to the work roles they
occupy and characteristics of their only a single measure (indicator) of each construct, it was necessary to
jobs. make assumptions about the values of the measurement parameters. These
Organizational The strength of an individual's parameters were fixed prior to the analyses and are noted with an aster-
commitment identification with and isk(*), whereas the values at which they were set are in parentheses (see
involvement in an organization. Figures 1 and 2). The rationale for selecting these values will now be
Intent to quit Workers' intentions, desires, and considered.
plans to quit their job. As discussed by Kenny (1979), information about the reliabilities of
Turnover The act of actually leaving the
the measured variables can be used when working with standardized
organization.
variables within a path-analytic framework. More specifically, in the
Sample 2 present application, the path from any construct to its measured variable
(X) equals the square root of the reliability of the measured variable,
Equity The extent to which an individual's while the amount of random error variance (6, e) is the quantity one
job inputs and/or outcomes are
minus the reliability. Thus, the coefficient alpha reliabilities reported by
perceived as equivalent to those in
Michaels and Spector (1982) and Bluedorn (1982) were used in this fashion
the individual's reference group.
to select the values at which the factor loadings and error variances were
Routinization The extent to which role
performance in an organization is set. These values appear in the parentheses in Figures 1 and 2.' This
repetitive. approach has been further described by James et al. (1982), and an ap-
Instrumental information The extent to which information plication has been given by Heise and Smith-Lovin (1982).
about role performance is
transmitted to organization
members.
1
Intent to leave The worker's plans for leaving the Michaels and Spector reported only the reliabilities of the measures
organization. of job satisfaction, organizational commitment, and leadership consid-
Age, satisfaction, Same as Sample I. eration. For the present analyses, the reliabilities of the intention questions
commitment, and
and the expectation question were assumed to be .85. The value used for
turnover
perceived job characteristics was .90.
TURNOVER MODELS 223

Jt S-*

1
s

A
'\ *•» ^> 1
~
\ £ °s
\

\
\
<
*"- ' ^S
> ^ e^ |
\ *
\CSJ
N S
u -
«

\ J
•a
«s ~ c ^
FH £
s
\ >-

>-\ J
224 LARRY J. WILLIAMS AND JOHN T. HAZER

Table 2
Matrix Representation for Sample 1

Structural model

JS 0 05 0 0 JS -rl 1-3 T5 77 TPEI r n "I


OC 04 0 0 0 OC T2 t4 T6 y8 PJC {2
+
IQ 01 (32 0 0 IQ ^9 TlO -fll yl2 LC J3
T 0 0 S3 0 T 0 0 0 0 LA J |_f4-
Measurement model
Yl XY1" 0 0 0 ~ JS «!•
Y2 0 XY21 0 0 OC <2«
+
Y3 0 0 XY3" 0 IQ <y
_Y4_ 0 0 0 XY4" T if
~X1~ " XXI' 0 0 0 PE &]•
X2 0 XX2« 0 0 PJC 62'
= +
X3 0 0 XX3' 0 LC sy
X4 0 0 0 XX4' _ A j 54"

Note. JS = job satisfaction; PE = preemployment expectations; OC = organizational commitment; PJC = perceived job characteristics; IQ = intent
to quit; LC = leadership consideration; T = turnover; A = age.
• Fixed parameter.

Model Evaluation ogenous variables describing personal and work environment character-
istics on turnover intentions; (b) satisfaction mediates the impact of the
A correlation matrix based on listwise deletion of missing data was exogenous variables on commitment; (c) commitment similarly serves
obtained from the Sample 1 data (n = 106), and a matrix based on as an intervening variable between satisfaction and turnover intentions.
pairwise deletion of missing data was used from the Sample 2 data (average Model C is identical to Model B, except that it proposes that satisfaction
n = 157)2-3. These correlation matrices were used individually as input and commitment are not related; this model allows for an explicit test
to the computer program LISREL (Version 6, Joreskog & Sorbom, 1985), of the satisfaction-commitment relation. Finally, Model D contains the
which provides maximum likelihood estimates of the parameters that previously described hypotheses that the exogenous variables impact on
are not fixed and a corresponding chi-square statistic and other goodness- commitment, which has an effect on satisfaction, which influences turn-
of-fit measures, which represent the overall adequacy of the model.4 Along over intentions. Although Models B and C are nested under Model A
with the correlation matrix, the information from the estimated reli- and could be evaluated as such, Model D is not nested under the other
abilities from each sample was also entered into the program. Thus, a models and was examined independently.
two-step process was used: The reliabilities were obtained, and then pa-
rameters were estimated based on the reliabilities.
In addition to a Structural Null Model proposing no causal relations Results
between the exogenous and endogenous variables (Bender & Bonett, 1980),
four other models were evaluated with each sample. Three of these models The chi-square values, associated degrees of freedom, and
differed from the general models presented in Figures 1 and 2 in that
probability levels for the models that were evaluated are presented
certain parameters representing relations between constructs were con-
for both samples in Table 5. Also reported in Table 5 are three
strained. Such a set of models can be said to be nested. James et al.
measures of the overall adequacy of each model provided by
(1982) have discussed the need for the development of a sequence of tests
for nested models prior to actually conducting the analyses to avoid "ex- LISREL: root mean square residua] (RMSR), goodness-of-fit index
ploring" the data. Accordingly, each model serves as a basis of comparison (GFI), and the adjusted goodness-of-fit index (AOFI). Also pre-
for those prior to it in the sequence, and as such, must differ by having
additional parameters constrained, or in the present application, fixed
3
to zero. The null hypothesis is that the restricted parameters) equals Although these sample sizes are moderate at best, the use of model
zero, and a chi-square difference test (e.g., Bender & Bonett, 1980) is comparison procedures (rather than individual chi-square values), incre-
used to test the hypothesis. Under these conditions, a comparison of any mental fit indexes (Bentler & Bonett, 1980) and two samples minimizes
two sequential models serves as a test of the importance of the restricted this problem. However, it should be noted that important work is in
parameters). This sequence of model comparisons continues until one progress that examines the distributions of the chi-square statistic and
rejects the null hypothesis, at which point the alternative model with the goodness-of-fit indexes under a variety of conditions (e.g., Anderson &
free parameters) is tentatively "accepted." The fourth model was evaluated Gerbing, 1984; Bearden, Sharma, & Teel, 1982).
3
independently because it was not nested under the first three models. To In a personal communication, Spector indicated that the positive
supplement the chi-square difference test, goodness-of-fit measures were signs of the correlations between satisfaction and intent to leave and
examined. commitment and intent to leave were reported incorrectly. Therefore,
The four models that were evaluated and the parameters that were free we have corrected this error by changing these signs prior to the analyses.
to be estimated in each are presented in Table 4. As can be seen in the * It is also possible to analyze a covariance matrix with LISREL. To
description in Table 4, Model A contains the most free parameters and account for measurement error, the value entered for (X) still equals the
serves as a basis of comparison for Models B and C which are nested square root of the reliability of the measured variable, but the amount
under it. Model B contains the three mediating hypotheses presented of random error variance (5,e) is the quantity: (one minus the reliability)
earlier: (a) satisfaction and commitment mediate the impact of the ex- multiplied by the variance of the measured variable.
TURNOVER MODELS 225

s
•5
226 LARRY J. WILLIAMS AND JOHN T. HAZER

Table 3
Matrix Representation for Sample 2

Structural model
IS 0 05 0 JS 7" 73 75
OC 04 0 0 OC 72 74 76
IL 02 0 IL 79 710
T 0 03 T o 0

Measurement model
1
Yl XYI 0 0 0 JS ~ el"
Y2 0 AY2" 0 0 OC tV
+
Y3 0 0 XY3" 0 IL e3'
Y4 0 0 0 XY4- T t4'

~X1 ~xxi" o o o ~E ~ d\" '


X2 0 XX2- 0 0 R x2"
X3 0 0 XX3" 0 II t3f
X4 0 0 0 XX4"j A 64"

Note. JS = job satisfaction; E = equity; OC = organizational commitment; R = routinization; IL = intent to leave; II = instrumental information;
T = turnover; A = age.
* Fixed parameter.

sented in this table are the values for the two other goodness-of- hypothesized restrictions to zero of 79-712, 72, y4, 76, 78 and
fit measures used. The normed fit index (Bentler & Bonett, 1980) 01. It is also important that these parameters, when allowed to
is calculated as follows: NFI = (x 2 Nu ii - x2Modd)/x2NUu. The for- be free in Model A, were generally nonsignificant. More specif-
mula for the parsimonious fit index (James, Mulaik, & Brett, ically, in Model A only the estimates of f)\ from Sample 1 and
1982) is: PFI = (dfM<xtel/dfNull) X NFI. In general, these five mea- >4 in Sample 2 had critical ratios (estimate/standard error)
sures represent the difference between the correlation matrix greater than 2.0. In other words, only 2 of the 18 parameters
predicted by the model and one actually obtained in the sample. that were restricted across both samples in Model B had signif-
As indicated in Table 5, Model A was shown to have an ad- icant critical ratios in Model A.
equate fit in both samples according to four of the five fit mea- Further support for Model B conies from the fit measures.
sures, which was expected because it contains many parameters Although the NFI drops from .960 and .967 to .873 and .905,
that have been supported in the literature. The relative small respectively, in both samples, these values are still adequate ac-
value for the PFI reflects the large number of parameters in Model cording to the .90 criterion provided by Bentler and Bonett
A. Next, Model B was evaluated, which contained the restriction (1980). Also, the GFI and AGFI values for Model B are not very
to zero of nine paths. In Sample I , the comparison of Model B
with Model A yielded a chi-square difference of 20.69, whereas
in Sample 2 it was 18.64. Although both of these values are Table 5
slightly larger than the .05 critical value of 16.92, the Sample 2 Results ofLJSREL Analyses
result is less than the critical value at the .025 level (19.02), and
the Sample 1 result is less than the critical value at the .01 level Model df x2 RMSR NFI PFI GFI AGFI
(21.67). Thus, based on this test, there was some support for the
Sample 1
A 6 9.60* .040 .960 .262 .979 .874
B 15 30.29 .059 .873 .595 .936 .846
Table 4
C 16 103.57 .225 .566 .412 .854 .671
Model descriptions D 15 55.78 .081 .766 .522 .889 .734
Structural
Model Free parameters Description null 22 238.90 .291 .605 .354
— —
A 7! - 7 l 2 ; 0 1 -04 E — S, C, and I; S — C; Sample 2
S and C — I; I — T
B 7 1,73, 75, 77; 02, E — S; S — C; C — I; A 6 10.12* .038 .967 .264 .984 .906
03,04 I— T B 15 28.76* .055 .905 .617 .958 .898
C 71,73, 75, 77; 02, E — S; C — I; I — T C 16 159.04 .208 .475 .345 .853 .669
03 D 15 54.64 .063 .819 .558 .919 .805
D 72,74, 76, 78; 01, E — C; C — S; S — I; Structural
03,05 I— T null 22 303.21 .263 — — .656 .437
Structural No causal paths No causal paths
null Note. RMSR = root mean square residual; NFI = normed fit index.
PFI = parsimonious fit index; GFI = goodness-of-nt index; AGFI =
Note. E = exogenous variables; S = job satisfaction; C = organizational adjusted goodness-of-nt index.
commitment; I = intent to quit/leave; T = turnover. * p > .05.
TURNOVER MODELS 227

different from Model A, with the largest difference being .043 Model B. Whereas these two models cannot be directly compared
for the CFI in Sample 1. Finally, the strongest evidence in support because they are not nested, the chi-square values and fit mea-
of Model B is provided by the PFI values. As described earlier, sures can be examined for each. Such an examination indicates
this measure summarizes how well a model fits data in terms of Model B to have a lower chi-square value and better values on
the parsimony of the model by taking account of the degrees of each fit measure. It is also important to note that a model iden-
freedom of the model relative to the Null Model. It is clear that tical to Model A except for having 05 rather than /34 would have
Model B is more satisfactory in this respect than Model A, in the same chi-square value as Model A, because in this instance
that the PFI values are considerably higher. More specifically, it is impossible to distinguish between these two models. There-
the PFI increases in Model B from .262 and .264 to .595 and fore, had such a model been compared to Model D, the resulting
.617, respectively, a noticeable improvement. In summary, the chi-square difference would have clearly exceeded the critical
combined evidence from the chi-square difference test, critical value. This fact provides additional evidence in support of Model
ratios, and goodness-of-fit measures supports Model B and the B over Model D.
three mediating hypotheses contained in it. Finally, as suggested by James, Mulaik, and Brett (1982),
The comparison of Model B with Model C was conducted Model B was compared to the Structural Null Model to test the
next to evaluate the importance of 04. The chi-square difference importance of the parameters in Model B. The chi-square dif-
was large and significant in both Sample 1 and 2 (73.28, 130.28, ference was large and significant in both samples (208.61 in
respectively). Thus, the hypothesized restriction of 04 can be Sample 1 and 274.45 in Sample 2). In conclusion, the compar-
rejected. Also, the values for all of the fit measures dropped con- isons of Models A, B, C, and D provide support for the proposed
siderably. Specifically, the drop for the PFI measure (.183 and mediating hypotheses contained in Model B and demonstrate
.272) clearly indicates the importance of 04 to the overall model. the importance of the satisfaction-commitment relation. The
Model D was evaluated as an alternative for comparison to parameter estimates from Model B are provided in Figures 3

(.38)

* CTltlctl utlo < 2.0

Figure 3. Summary of Model B in Sample I.


228 LARRY J. WILLIAMS AND JOHN T. HAZER

and 4. As can be seen in these figures, the values of all the esti- these issues. Although the specific results of the analyses varied
mates were reasonably large, with only the path from routini- slightly across the two samples, in general the model comparisons
zation to satisfaction in Sample 2 having a critical ratio less led to the same conclusions.
than 2.0. With respect to both samples, the results of the comparison
of Model A with Model B provided clear answers to the first
Discussion three questions. First, the analyses indicated that the variables
describing personal characteristics and the work environment
The present research was conducted to address the following have no direct impact on turnover intentions. In view of the
issues: (a) Do personal and organizational characteristics influ- variety of these exogenous variables (e.g., expectations, age, job
ence turnover intentions through their effect on satisfaction and characteristics, instrumental information), this finding substan-
commitment? (b) Are personal and organizational characteristics tiates the importance of satisfaction and commitment as inter-
direct antecedents of both satisfaction and commitment? (c) Do vening variables in models of turnover, these variables represent
both satisfaction and commitment have an impact on turnover the affect that results even from very diverse antecedents.
intentions? (d) Should job satisfaction be considered as a causal The results also provided evidence relevant to the second
factor contributing to organizational commitment? and (e) Is question. With both samples, the comparison strongly supported
there support for a causal relation from commitment to satis- the proposition that personal and organizational characteristics
faction? These questions formed the basis of the mediating hy- influence only satisfaction directly, and influence commitment
potheses that were investigated. The use of two data sets, rep- only indirectly through their impact on satisfaction and its sub-
resenting different samples and different operationalizations of sequent effect on commitment. This conclusion received further
the factors of interest, allowed for a thorough investigation of support from the fact that the paths from the exogenous variables

.U (-48)

* tlltlctl Tttlo < 2.0

Figure 4. Summary of Model B in Sample 2.


TURNOVER MODELS 229

to commitment had critical ratios less than 2.0 (with the excep- longitudinal data result in interpretational difficulties, whereas
tion of -y4 in Sample 2). From a theoretical perspective, this latent variable models are applicable and offer many advantages
suggests that the existing conceptual models of commitment with this type of data. These advantages include preventing some
should be reexamined to determine whether all of the proposed of the problems created by specification errors such as incorrect
determinants might have their effects indirectly via satisfaction. assumptions about measurement error and correlated residuals
Additionally, such an effort might identify causal factors of com- in a set of regression equations.
mitment operating independently of satisfaction. Either way, the Because the data on which the present research was based
result would be a clarification of the processes through which came from existing published research, one implication is that
affective factors influence turnover intentions. the original findings from Michaels and Spector (1982) and
The results also indicate an important methodological limi- Bluedom (1982) need to be reconsidered. With respect to Sample
tation of the path analysis technique: It does not allow for the 1, the results reported by Michaels and Spector clearly were not
simultaneous examination of the effects of exogenous variables supported in the reanalysis. Their failure to consider satisfaction
on two or more endogenous variables. In other words, even in and commitment as causally related was a serious shortcoming,
the previous studies, which included both satisfaction and com- as shown by the substantial increment in fit associated with the
mitment, the use of path analysis precluded addressing the ques- path /34 during the Model B-Model C comparison. They also
tion of the direct effects of determinants on both of these variables reported significant and rather large path coefficients associated
at the same time. On the other hand, working with a system of with the relations between the exogenous variables and com-
equations (as with LISREL) provides an alternative to this problem. mitment. The present analysis demonstrated these causal paths
With respect to the third question, the comparison suggests to be insignificant, both statistically through the chi-square dif-
that commitment had a more important effect on intent to leave ference test and substantively through the goodness-of-fit indexes.
or quit than did satisfaction. However, this should only be con- Finally, although the present research did not directly compare
sidered as a tentative conclusion. Although in Sample 1 the crit- their model with one which made provisions for measurement
ical ratio for the satisfaction-intent path (01) was not significant, error, the preceding discussion suggests that the path coefficients
in Sample 2 it was. Furthermore, such a finding is not surprising they reported and the overall variance accounted for by their
because one of the components of organizational commitment model were influenced by measurement error.
is the desire to remain with the organization. A more desirable The current reanalysis also has implications for the findings
test of the question would be based on excluding the desire-to- reported by Bluedorn (1982), although only four of the exogenous
stay items from the commitment scale. Unfortunately, this was variables from the original data were used. Bluedorn (1982)
not possible in the present case because existing published data demonstrated the importance of the satisfaction-commitment
was used. relation; this was further exemplified by the large increment in
In terms of the fourth question, the comparison of Model B fit associated with it in the present analysis. However, although
with Model C indicated that the direct link from satisfaction to the direct path between age and turnover was not investigated
commitment was supported in both samples. The chi-square currently, the failure during the reanalysis to support a direct
difference between Model B and Model C greatly exceeded the link from age to intent suggests that the age-turnover link might
critical value, and Model C showed a much poorer fit across all be unnecessary. Also, the paths from equity and age to com-
of the goodness-of-fit measures. Furthermore, as shown in Figure mitment were not replicated in the reanalysis; they were dem-
3 and Figure 4, the parameter estimates for 04 were very large onstrated to lead to commitment only indirectly via satisfaction.
in both samples. Clearly, this indicates a strong and important Interestingly, the relation between routinization and satisfaction
relation. did not parallel those of the other exogenous variables: The critical
The evaluation of Model D attempted to address a fifth ques- ratio (LISREL estimate/standard error) did not exceed 2.0, sug-
tion, the issue of the causal impact of commitment on satisfaction. gesting that it was not significant. Also, this may be the result of
As mentioned previously, the evidence favored Model B over the measure of satisfaction used in Sample 2. Finally, as with
Model D. Thus, there was less support for a causal link from the Michaels and Spector results, our discussion of measurement
commitment to satisfaction than for a path in the reverse direc- error suggests that the path coefficients reported by Bluedorn
tion. However, a few comments are appropriate with respect to were biased.
these findings. First, the test of the commitment to satisfaction The findings of the current research present a challenge to the
path was conducted in the context of a model proposing that the previously described models and research investigating com-
exogenous variables influence only commitment directly. As mitment and turnover. The major focus of the commitment lit-
found by Bateman and Strasser (1984), an alternative model erature has been to identify antecedents of commitment from a
would be that the exogenous variables and commitment influence variety of categories (e.g., Morris & Sherman, 1981; Steers, 1977;
satisfaction. Second, although the results of these two models Stevens et al., 1978). These categories have included personal
suggest the possibility of a reciprocal relation between satisfaction characteristics, work experiences, job characteristics, organiza-
and commitment, it was not possible to test for this with the tional factors, and role-related factors. The primary mechanism
present data due to such a model being underidentified. Finally, through which these factors have been assumed to influence
it should be noted that trying to assess directionality with cross- commitment is an exchange process. The present research, in
sectional data is a nearly impossible task, and longitudinal re- conjunction with existing reviews of the antecedents of satisfac-
search such as that by Bateman and Strasser (1984) is necessary tion, which have validated the roles of personal and organizational
to address the question. However, as discussed by Gallini (1983), characteristics (e.g., Mitchell, 1979), supports the proposal that
the specification errors associated with using path analysis with these factors may influence commitment only indirectly. In other
230 LARRY J. WILLIAMS AND JOHN T. HAZER

words, through a process of the evaluation of costs and benefits, Billings, R. S., & Wroten, S. P. (1978). Use of path analysis in industrial/
individual needs and desires are satisfied, and the resulting af- organizational psychology: Criticisms and suggestions. Journal of Ap-
fective state becomes associated with the organization, which has plied Psychology. 63. 677-688.
provided the job and its associated characteristics and environ- Bluedom, A. C. (1979). Structure, environment, and satisfaction: Toward
ment. Commitment results from this association. Such a per- a causal model of turnover from military organizations. Journal of
Military and Political Sociology, 7, 181-207.
spective was the basis of the other previously described research
Bluedom, A. C. (1982). A unified model of turnover from organizations.
by Farrell and Rusbult (1981) and Rusbult and Farrell (1983).
Human Relations, 35, 135-153.
Additionally, an investigation found support for this exchange
Clegg, C. W. (1983). Psychology of employee lateness, absence, and turn-
process in interpersonal relations (Rusbult, 1983). In this lon- over A methodological critique and an empirical study. Journal of
gitudinal study, satisfaction resulted from rewards and led to Applied Psychology, 68, 88-101.
increased commitment in the relation with another. Farrell, D., & Rusbult, C. E. (1981). Exchange variables as predictors of
In conclusion, several conceptual models and path analysis job satisfaction, job commitment, and turnover The impact of rewards,
studies of the processes responsible for turnover were presented, costs, alternatives, and investments. Organizational Behavior and Hu-
and the results of the current research should provide for the man Performance, 27, 79-95.
further development of these models. The support for the sat- Gallini, J. (1983). Misspecifications that can result in path analysis struc-
tures. Applied Psychological Measurement, 7, 125-137.
isfaction-commitment relation and the clarification of the pro-
Heise, D. R., & Smith-Lovin, L. (1981). Impressions of goodness, pow-
cesses through which personal and organizational characteristics
erfulness, and liveliness from discerned social events. Social Psychology
influence satisfaction and subsequently commitment represent
Quarterly, 44, 93-106.
important refinements that should benefit further efforts in this Horn, P. W., & Hulin, C. L. (1981). A competitive test of the prediction
area. Rather than simply categorize determinants of turnover of reenlistment by several models. Journal of Applied Psychology, 66,
(e.g., Muchinsky & Morrow, 1980), researchers can try to detail 23-39.
relations between and within these classes. As an alternative to Horn, P. W, Katerbeig, R., & Hulin, C. L. (1979). Comparative exam-
assuming that satisfaction is the primary affective factor resulting ination of three approaches to the prediction of turnover. Journal of
in turnover (e.g., Mobley et al., 1979; Price, 1977), or that the Applied Psychology, 64, 280-290.
affective responses are unrelated (e.g., Steers & Mowday, 1981), Homans, G. C. (1961). Social behavior: Its elementary forms. New York:
Harcourt, Brace, & World.
attention can be given to detailing the sequence of emotional
James, L. R., & Brett, J. (1984). Mediators, moderators, and tests for
responses in this process. In terms of the empirical turnover
mediation. Journal of Applied Psychology, 69, 307-321.
studies reviewed, the present research suggests that those failing
James, L. R., Mulaik, S. S., & Brett, J. M. (1982). Causal analysis. Beverly
to include both satisfaction and commitment (e.g., Bluedorn, Hills, CA: Sage.
1979; Martin, 1979; Price & Mueller, 1981) should be viewed Jflreskog, K., & Sorbom, D. (1985). LISREL: Analysis of linear structural
cautiously. Ignoring the causal relation between satisfaction and relationships by the method of maximum likelihood. Chicago: National
commitment (e.g., Arnold & Feldman, 1982) was also demon- Educational Resources, Inc.
strated to represent a serious limitation. Finally, all of the path- Kenny, D. A. (1979). Correlation and causality. New York: Wiley.
analysis studies presented ignored measurement error: The Koch, J. L., & Steers, R. M. (1978). Job attachment, satisfaction, and
reanalyses that were conducted presently demonstrate a technique turnover among public sector employees. Journal of Vocational Be-
havior, 12, 119-128.
for accounting for measurement error that should be useful to
Martin, T. N. (1979). A contextual model of employee turnover intentions.
researchers in this area.
Academy of Management Journal, 22, 313-324.
Michaels, C. E., & Specter, P. E. (1982). Causes of employee turnover
A test of the Mobley, Griffeth, Hand, and Meglino model. Journal of
References Applied Psychology. 67, 53-59.
Miller, H. E., Katerberg, R., & Hulin, C. L. (1979). Evaluation of the
Anderson, J. C, & Gerbing, D. W. (1984). The effect of sampling error Mobley, Homer, and HoUingsworth model of employee turnover. Jour-
on convergence, improper solutions, and goodness-of-fit indices for nal of Applied Psychology, 64, 509-517.
maximum likelihood confirmatory factor analysis. Psychometrika, 49, Mitchel, J. O. (1981). The effect of intentions, tenure, personal, and or-
155-173. ganizational variables on managerial turnover Academy of Management
Arnold, H., & Feldman, D. C. (1982). A multivariate analysis of the Journal, 24, 742-751.
determinants of job turnover. Journal of Applied Psychology, 67, 350- Mitchell, T. R. (1979). Organizational behavior. In M. R. Rosenzweig &
360. L. W. Porter (Eds.), Annual review of psychology, (Vol. 30, pp. 243-
Bagozzi, R. R. (1980). Causal models in marketing. New York: Wiley. 281). Palo Alto, CA: Annual Reviews.
Bateman, T. S.. & Strassei; S. (1984). A longitudinal analysis of the an- Mobley, W. H. (1977). Intermediate linkages in the relationship between
tecedents of organizational commitment Academy of Management job satisfaction and employee turnover. Journal of Applied Psychology,
Journal, 27, 95-112. 62. 237-240.
Bearden, W. Q, Sharma, S., & Teel, J. E. (1982). Sample size effects on Mobley, W. H., Griffeth, R. W., Hand, H. H., & Meglino, B. M. (1979).
chi-square and other statistics used in evaluating causal models. Journal Review and conceptual analysis of the employee turnover process. Psy-
of Marketing Research 19, 425-430. chological Bulletin, 86, 493-522.
Bentler, P. M. (1980). Multivariate analysis with latent variables: Causal Mobley, W. H., Homer, S. O., & HoUingsworth, A. T. (1978). An evaluation
modeling. In M. R. Rosenzweig & L. W. Porter (Eds.), Annual review of the precursors of hospital employee turnover. Journal of Applied
of psychology (Vol. 31, pp. 419-456). Palo Alto, CA: Annual Reviews. Psychology, 63, 408-414.
Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness Morris, J. H., & Sherman, J. D. (1981). Generalizability of an organi-
of fit in the analysis of covariance structures. Psychological Bulletin, zational commitment model. Academy of Management Journal, 24,
88, 588-606. 512-526.
TURNOVER MODELS 231

Morrow, P. C. (1983). Concept redundancy in organizational research: of variations in rewards, costs, alternatives, and investments. Journal
The case of work commitment. Academy of Management Review, 8, of Applied Psychology, 68, 429^*38.
486-500. Schmitt, N., & Bedeian, A. G. (1982). A comparison of USREL and two-
Mowday, R. T., Steers, R. M., & Porter, L. W. (1979). The measurement stage least squares analysis of a hypothesized life-job satisfaction re-
of organizational commitment. Journal of Vocational Behavior, 14, ciprocal relationship. Journal of Applied Psychology, 67, 806-817.
224-247. Steers, R. M. (1977). Antecedents and outcomes of organizational com-
Muchinsky, P. M., & Morrow, P. C. (1980). A multidisciplinary model mitment. Administrative Science Quarterly, 22, 46-56.
of voluntary turnover. Journal of Vocational Behavior, 17, 263-290. Steers, R. M., & Mowday, R. T. (1981). Employee turnover and post-
Parasuraman, S. (1982). Predicting turnover intentions and turnover be- decision accommodation processes. Research in Organizational Be-
havior. A multivariate analysis. Journal of Vocational Behavior, 21, havior, Volume 3. JAI Press, Inc.
111-121. Stevens, J. M., Beyer, J. M., & Trice, H. M. (1978). Assessing personal,
Porter, L. W., Steers, R. M., Mowday, R. T., & Boulian, P. V. (1974). role, and organizational predictors of managerial commitment. Acad-
Organizational commitment, job satisfaction, and turnover among emy of Management Journal, 21, 380-396.
psychiatric technicians. Journal of Applied Psychology, 59, 603-609. Weiner, Y., & Vardi, Y. (1980). Relationships between job, organization,
Price, J. L., & Mueller, C. W. (1981). A causal model of turnover for and career commitments and work outcomes—An integrative ap-
nurses. Academy of Management Journal, 24, 543-565. proach. Organizational Behavior and Human Performance, 26, 81-
Rusbult, C. E. (1983). A longitudinal test of the investment model: The 96.
development (and deterioration) of satisfaction and commitment in
heterosexual involvements. Journal of Personality and Social Psy-
chology, 45, 101-117.
Rusbult, C. E., & Farrell, D. (1983). A longitudinal test of the investment Received April 30, 1985
model: The impact on job satisfaction, job commitment, and turnover Revision received August 2, 1985

Das könnte Ihnen auch gefallen