Professional Documents
Culture Documents
Department of Psychology
University of Western Ontario, Canada
The factor structure of the Affective and Continuance Commitment Scales (ACS and CCS; Meyer
& Alien, 1984), as well as the causal links between affective and continuance commitment, were
examined. Data were obtained from 2 employee samples on a single occasion and from a sample of
new employees on 3 occasions during their first year of employment. Confirmatory factor analyses
revealed that (a) the ACS and CCS measure different constructs and (b) the CCS can be divided into
2 highly related subscales reflecting costs associated with leaving the organization (lack of alternatives and personal sacrifice). Tests of nonrecursive causal models with cross-sectional data revealed that (a) affective commitment had a negative effect on the alternatives component of continuance commitment in all samples and (b) both components of continuance commitment had a
positive effect on affective commitment for established employees. Analyses of the longitudinal
data revealed only a weak, negative, time-lagged effect of the alternatives component on affective
commitment.
cost of doing otherwise. Meyer and Allen used the terms affective commitment and continuance commitment, respectively, to
characterize these two views of the construct and noted that,
although both views reflect a link between the employee and
the organization that decreases the likelihood of turnover, the
nature of the link is quite different. Employees with a strong
affective commitment remain with the organization because
they want to, whereas those with strong continuance commitment remain because they need to. The antecedents and consequences (other than remaining) of these two forms of commitment are, therefore, also likely to be quite different.
To test these hypotheses, Meyer and Allen (1984; Allen &
Meyer, 1990) developed the Affective Commitment Scale and
the Continuance Commitment Scale (ACS and CCS, respectively). These eight-item self-report measures were found to be
reliable and to yield scores that are relatively independent. Studies in which these scales were used have shown that, as predicted, affective and continuance commitment are differentially related to variables purported to be their antecedents (Allen & Meyer, 1990; Aven, 1988; Withey, 1988) and consequences
(Allen & Smith, 1987; Meyer, Paunonen, Gellatly, Coffin, &
Jackson, 1989). For example, Allen and Meyer (1990) found
that affective commitment was best predicted by work experiences that promote feelings of comfort in the organization (e.g.,
organizational dependability) and personal competence (e.g.,
job challenge). Continuance commitment, on the other hand,
correlated more highly with measures of potential loss (e.g.,
pension benefits) and lack of alternatives. Meyer et al. (1989)
found that affective commitment correlated positively and continuance commitment, negatively, with supervisor ratings of
job performance and promotability.
Factor analyses of the ACS and CCS were performed in two
recent studies. Allen and Meyer (1990, Study 1) found that all
items loaded appropriately on two factors. McGee and Ford
(1987) found that, although all but two of the CCS Hems loaded
710
appropriately when two factors were extracted, when the number of factors was not specified a priori, four orthogonal factors
emerged. The four-factor solution included an affective commitment factor, two continuance commitment factors, and a
fourth factor that was not interpreted. The continuance commitment factors, each denned by three CCS items, were labelled Low Perceived Alternatives and High Personal Sacrifice
to reflect the specific source of the cost associated with leaving.
McGee and Ford (1987) also found that when they created
three-item subscales of the CCS (i.e., CCLoAlt and CCHiSac),
these subscales correlated significantly with one another (r =
.37, p < .001) and correlated significantly, but in opposite directions, with the ACS. The CCLoAlt subscale correlated negatively (r = - .21, p< .001) with the ACS, whereas the CCHiSac
subscale correlated positively if = .34, p < .001). Like Meyer and
Allen (1984), McGee and Ford found that when the full CCS
was correlated with the ACS, the correlation was near zero.
Although the results of these factor-analytic studies support
the hypothesis that affective and continuance commitment are
relatively distinct constructs, they conflict with respect to the
factor structure of the CCS. Whereas Allen and Meyer (1990)
interpreted their findings as evidence that the CCS measures a
unitary construct, McGee and Ford (1987) concluded that (a)
the CCS measures two distinguishable constructs and (b) the
high-personal-sacrifice component best reflects "side-bet commitment" as described by Becker (1960).
There is reason to be cautious in accepting either of these
interpretations. First, the analytic procedures used i n both studies were exploratory rather than confirmatory. Although
McGee and Ford (1987) obtained two different solutions, they
were not explicit about the criteria used in accepting one over
the other. Second, neither factor structure has been shown to
replicate consistently across different employee samples. Finally, examination of the items that define the two continuance
commitment factors obtained by McGee and Ford suggests that
the factors do not reflect low perceived alternatives and high
personal sacrifice as clearly as McGee and Ford implied. For
example, although agreement with the CC:LoAlt item "Right
now, staying with my organization is a matter of necessity as
much as desire" could imply a lack of alternatives, it could just
as easily reflect other perceived costs. Similarly, agreement with
the CCHiSac item "It would be hard for me to leave my organization right now, even if I wanted to" could reflect personal
sacrifice, but it might also indicate difficulty finding a comparable alternative.
In the present study, we examined the factor structure of the
ACS and CCS by applying confirmatory factor analyses to data
obtained from three different samples of employees. With confirmatory factor analysis, it is possible to evaluate the fit of
models generated to reflect the factor structures proposed by
Allen and Meyer (1990) and McGee and Ford (1987). By applying these analyses to data from d ifferent samples, it is also possible to evaluate the generalizability of factor models.
A second purpose of this study was to examine the causal
relation(s) between affective and continuance commitment.
Though McGee and Ford (1987) acknowledged that the procedures they used precluded a causal explanation, they speculated
about the psychological process that might account for the positive correlation between affective commitment and the CCHi-
711
Sac subscale. Following Steers and Porter (1983), they suggested that dissonance-reduction or self-justification processes
may be involved. Steers and Porter proposed that individuals
who feel bound to remain with an organization "typically engage in some form of psychological bolstering in which they
attempt to rationalize or self-justify their situation" (p. 428).
There are at least two reasons for being cautious in accepting
McGee and Ford's (1987) explanation. First, if dissonance-reduction or self-justification processes do operate as a consequence of being bound by the threat of personal sacrifice, as
McGee and Ford suggested, it is not clear why being bound to
an organization by lack of alternatives does not have the same
effect. Second, there are other plausible explanations for the
positive correlation between the ACS and the CCHiSac subscale. It is possible, for example, that the conditions that contribute to the development of affective commitment (e.g, organizational dependability, job challenge) are the very factors that
would be sacrificed if the employee were to leave. Affective
commitment, therefore, might be antecedent to continuance
commitment. The implications of this explanation for organizations' attempts to foster commitment in their employees are
quite different from those following from McGee and Ford's
explanation. Clearly, therefore, there is a need to determine
whether the differential relations between affective commitment and the two components of continuance commitment
observed by McGee and Ford are robust, and, if so, to explain
these relations.
In this study, we examined the nature of the relation between
affective and continuance commitment by applying linear
structural-relations analyses to both cross-sectional and longitudinal data. Although there is disagreement about the utility
of detecting causal effects with cross-sectional data (cf. James,
Mulaik, & Brett, 1982, and Gollob & Reichardt, 1987), we
chose to conduct such analyses with the rationale that thoughts
about one's affective attachment to an organization might have
immediate implications for thoughts about the influence of
costs on the decision to stay and vice versa. Indeed, responding
to questions about affect and cost might produce effects such as
those described above. For example, responding to questions
concerning the personal sacrifice that would result from leaving an organization could influence responses to questions
concerning attachment to the organization and vice versa. This
does not mean, however, that any effects observed are merely
artifactual. To the contrary, we expect that being prompted by a
set of questions to think about one's feelings about the organization and reasons for staying is no different from being
prompted by any other naturally occurring event (e.g, being
offered another job; having a particularly good or bad day at
work). Our objective in testing causal effects with cross-sectional data was to examine how thoughts relevant to one form
of commitment might influence or be influenced by those relevant to another form.
In summary, the need to clarify the distinctions among the
various conceptualizations of commitment is well recognized.
Initial efforts by Meyer and Allen (1984; Allen & Meyer, 1990)
to distinguish and measure commitment as conceptualized by
Porter (Porter et al., 1974; Mowday et aj, 1982) and Becker
(1960) have been helpful in this regard. Recent research using
the ACS and CCS, however, has yielded conflicting results with
712
Method
Subjects and Procedure
Data were collected from three samples of employees. For the first
two samples, commitment was measured on a single occasion; for the
third, commitment was measured for the same employees on three
separate occasions.
Cross-sectional data. In the firet of two cross-sectional studies (Sample 1), data were collected from full-time, nonunionized employees in
three organizations: a retail department store, a hospital, and a university library. In total, 634 questionnaires were delivered to the participating organizations for distribution. Of these, 337 (53.2%) were completed and returned. The majority of respondents were women (80.2%),
the average age was 38 years, and average tenure was 8.86 years. Because of missing data, the sample size was reduced to a maximum of
319 for analyses reported here.
In the second study (Sample 2), data were collected from 292 fulltime employees (79 men, 213 women) in several organizations, including a university (n = 142), a community college (n - 32), a petroleum
company (n = 28), a hospital (n = 24), the administrative office of a
union ( = 10), and a clothing company (n = 8). In addition, 54 questionnaires were completed by part-time university students employed
in a variety of organizations. The average age of respondents was 37
years, and average tenure in the organization was 8 years. Again because of missing data, the sample size for analyses reported here was
reduced to 274.
Longitudinal data. Participants in thisstudy (Sample 3) were university graduates who had recently accepted full-time jobs with a number
of different companies. Students who were involved in the on-campus
recruitment program at the University of Western Ontario from 1983
to 1986 were asked if they would participate in a longitudinal study of
work attitudes. Of those who qualified (by virtue of having accepted
full-time jobs), almost all agreed to participate. Of the 308 individuals
who completed and returned pre-entry questionnaires, 123 (40%) were
women, 279 (90%) were between the ages of 21 and 25 years, all had
undergraduate degrees, and 28 had graduate degrees. Post-entry questionnaires were sent to participants after approximately 1, 6, and 11
months on the job. Questionnaires with usable data were returned by
276, 247, and 210 individuals, respectively. Because of missing data,
the sample size for some analyses was reduced (actual sample sizes are
reported with the results of these analyses).
Measures
Commitment. Commitment was measured with the ACS and CCS,
used in research by Meyer and Allen (1984; Allen & Meyer, 1990) and
McGee and Ford (1987). Responses were made on 7-point scales ranging from strongly disagree (1) to strongly agree(7). (See McGee & Ford,
1987, or Allen & Meyer, 1990, for the ACS and CCS items.)
Predictors of commitment. For analyses undertaken to examine the
causal links between affective and continuance commitment in crosssectional data, it was necessary to include predictors of both forms of
commitment in the causal model. Such predictor measures were available for two of the three samples (Samples 1 and 3). Two measures each
Data Analysis
Both the factor structure of the commitment scales and the structural relations among the resultant factors were examined with LISREL
vi (Joreskog & Sorbom, 1986). In each of the analyses reported here,
maximum likelihood parameter estimates were obtained by applying
LISREL to the appropriate covariance matrix. Although USREL permits
the simultaneous testing of both a measurement model (i.e., a model
specifying the relations of observed variables to their posited underlying constructs) and a structural-relations model (i.e., a model specifying the causal relations among the latent constructs), Anderson and
Gerbing (1988) recommended using a "two-step approach" to avoid
interpretational confounding (cf. Burt, 1976). Following this advice,
we first conducted multiple-group confirmatory factor analyses to ensure an adequate measurement model. We then examined the structural relations among the latent constructs.
Confirmatory factor analysis. Multiple-group confirmatory factor
analyses were performed to assess the factor structure of the CCS
alone, as well as of the CCS in combination with the ACS. In analyses
involving the CCS only, measurement models were developed to correspond to the single-factor structure proposed by Allen and Meyer
(1990) and to correspond to variations of the two-factor structure suggested by McGee and Ford (1987). In the model corresponding most
closely to the structure obtained by McGee and Ford, two orthogonal
factors were specified. (Although McGee & Ford reported a correlation of .37 between the subscale raw scores, they used an orthogonal
[varimax] rotation in their factor analysis.) We also tested an oblique
two-factor model. Moreover, because the findings of Allen and Meyer
and McGee and Ford conflict with regard to the inclusion of two of the
original CCS items, we tested all three models with both an eight- and a
six-item version of the CCS. Assignment of the two extra items in the
eight-item scale to the CQLoAlt and CCiHiSac subscales was based on
the wording of the items. The item "I am not afraid of what might
happen if I quit my job without having another one lined up" was
considered to reflect a concern with the availability of alternatives,
whereas the item "It wouldn't be too costly for me to leave my organization in the near future" was considered to reflect personal sacrifice
(reverse keyed).
The models examined in analyses of the CCS and ACS together
included variations of the two-factor structure reported by Allen and
713
As can be seen from the model in Figure 1, causal paths were specified from the predictor variables to the appropriate commitment variable and in both directions between the commitment variables. Of
particular interest in this study were the parameter estimates for the
paths between the commitment constructs (i.e., ft and 2). If only one
of the two paths is significant, it suggests that the causal effect is unidirectional. If both paths are significant, reciprocal causal effects are
indicated.
The data matrix used as input for these analyses included covariances among the commitment items and the raw predictor scores.
Thus, although latent commitment variables were created in these analyses, the predictor item and scale scores were treated as perfectly reliable measures of the underlying constructs. (The measurement models
for the predictors were not of interest in this study)
To examine long-term effects of continuance commitment on affective commitment, and vice versa, we also tested cross-lagged models
with longitudinal data. (See Rogosa [1980] for a discussion of timelagged regression analyses and Bateman & Strasser [1984] and Meyer
& Allen [1988] for applications using least squares analyses of raw
scores.) Although our analyses involved an evaluation of both the measurement and structural models, for simplicity we present only the
structural model in Figure 2. In this model, affective commitment and
continuance commitment assessed on one occasion are hypothesized
to exert an influence on the same measures obtained subsequently. The
influence of Time 1 measures on Time 3 measures are presumed to be
mediated completely by the Time 2 measures. Of particular interest in
this study were the parameter estimates for the paths reflecting the
time-lagged effects of one commitment construct on the other (i.e., y3,
jt, ft, and ft).
Although the use of latent variables in these structural-relations analyses requires the simultaneous estimation of parameters for both the
measurement and structural models, Anderson and Gerbing (1988)
noted that "interpretational confounding is minimized by prior separate estimation of the measurement model because no constraints are
placed on the structural parameters that relate the estimated constructs to one another" (p. 418). Moreover, the goodness of fit of the
structural model can be tested independent of the goodness of fit of
the measurement model with the relative normed-fit index (RNFI;
described by Mulaik, James, Van Alstine, Lind, & Stilwell, 1989). The
contribution of specific causal paths to the fit of the model can also be
assessed using chi-square difference tests (Bentler & Bonett, 1980).
Results
Confirmatory Factor Analyses
The fit indexes for the null models and the three measurement models examined for the six- and eight-item versions of
the CCS are reported in Table 1. Because there was evidence to
support the most restrictive hypothesis tested (i.e, HM), only
the fit values obtained under these restrictions are presented.
As can be seen, although the one-factor model provided a reasonably good fit, the fit of the oblique two-factor model was
better (chi-square difference = 56.21, p < .01, for the six-item
scale, and 59.49, p < .01, for the eight-item scale). The orthogonal two-factor model provided a poor fit, indicating that the
alternatives and personal-sacrifice components of continuance
commitment are not independent. Indeed, the standardized
phi coefficient reflecting the covariance of the two factors in
the oblique solution was .791 for the six-item models and . 817
for the eight-item models, indicating that the factors are highly
related. Although these phi coefficients are standardized, they
are not correlations. In multiple-group analyses, LISREL "scales
714
Predictor A
Predictor B
1.00
1.00
Predictor C
Predictor D
Figure 1. Nonrecursive model. (Predictors A through D are job challenge, organizational dependability,
loss of pension, and available alternatives, respectively, for Sample 1, and confirmed expectations, job
scope, available alternatives, and wasted time and effort, respectively, for Sample 3. The indicator variables labelled Al through A8 and Cl through C8 refer to the items on the Affective Commitment Scale and
Continuance Commitment Scale, respectively. For analyses involving a component of continuance commitment, only a subset of the indicator variables Cl through C8 was used, and the appropriate component
label [alternatives or personal sacrifice] can be substituted in the model.)
the latent variables so that a weighted average of their eovariance matrix is a correlation matrix, thereby retaining a scale
common to all groups" (Joreskog & Sb'rbom, 1989, p. 266, emphasis in original).
Raw-score correlations between the four-item subscales
ranged from .62 to .73, p < .01, across the three samples, and
the correlations between the three-item subscales ranged from
.54 to .68, p < .01. Finally, because the fit of the oblique two-factor solution for the eight-item models was comparable to that
for the six-item model, there is no need to discard the items
identified by McGee and Ford (1987) as problematic. (Because
the models were not nested, it was not possible to compare the
fit of the six- and eight-item models directly)
Given the results of the preceding analyses, we included all
eight CCS items and allowed the continuance commitment factors to correlate when we evaluated measurement models involving the CCS and ACS items together. Four models were
considered: a one-factor model, an orthogonal two-factor
model corresponding to the structure reported by Allen and
Meyer (1990), an oblique two-factor model, and an oblique
three-factor model corresponding to that obtained by McGee
and Ford (1987), with the modifications described above.
The fit indexes for these four models under three sets of restrictive assumptions (i.e, H fcrm , H A , and H w ) are reported in
Table 2. The one-factor model provided a very poor fit, indicat-
715
Figure 2. Time-lagged model. (Only the structural model is presented here. For analyses involving a
component of continuance commitment, the appropriate component label can be substituted in the
model.)
dimensions underlying the CCS, the structural-relations analyses were performed on the subscales rather than on the full
scale. Separate analyses were performed for the two subscales,
however, to avoid problems of colinearity.
Nonrecursive models within occasion of measurement. Standardized parameter estimates for each of the nonrecursive models are reported in Table 5. In the interest of space, parameter
estimates for the measurement model are not included. For
each of the models tested, the RNFI (Mulaik et al, 1989) exceeded .980, indicating a good fit of the structural model.
When continuance commitment was defined by the four alternatives items, affective commitment had a significant effect
on continuance commitment in all cases. The effect of alternatives-based commitment on affective commitment was significant only for Sample 1. Note that affective commitment exerted
a negative effect on the alternatives component of continuance
commitment, whereas the reciprocal effect, where significant,
was positive.
A somewhat different pattern emerged when continuance
commitment was defined only by the personal-sacrifice items.
Affective commitment had no significant effect on sacrificebased commitment, and the latter had a significant positive
effect on affective commitment only in Sample 1.
Time-lagged analysis of longitudinal data. Standardized pa-
Table 1
Tests oflmariance in Multiple-Group Confirmatory Factor Analyses
of the Six- and Eight-Item Continuance Commitment Scales
GF1
Model
Sample 1
df
Sample 2
Sample 3
NNF1
.516
.968
.865
.979
.597
.921
.898
.951
.861
.576
.920
.514
.965
.881
.972
.549
.925
.893
.949
.888
.648
.932
Six-item scale
Null model
1 factor
2 orthogonal factors
2 oblique factors
1,389.76
165.36
425.27
109.15
57
39
39
38
.556
.923
.883
.951
Eight-item scale
Null model
1 factor
2 orthogonal factors
2 oblique factors
1,812.99
221.87
534.44
162.38
100
76
76
75
.517
.924
.887
.948
Note. GFI = goodness of fit index; NNFI = nonnormed fit index. The NNFI was not applicable to the null
model.
716
Hypothesis/model
x1
df
Sample 1
Sample 2
Sample 3
Null model
Hfm,
1 factor
2 orthogonal factors
2 oblique factors
3 oblique factors
H,
1 factor
2 orthogonal factors
2 oblique factors
3 oblique factors
4,733.42
392
.418
.470
.503
2,335.57
844.64
840.89
696.27
312
312
309
303
.639
.885
.886
.904
.633
.901
.901
.921
.665
.879
.880
.899
.414
.846
.845
.883
2,393.48
927.82
925.47
754.32
342
340
337
329
.638
.876
.877
.903
.628
.889
.889
.911
.662
.865
.865
.889
.458
.844
.842
.883
1 factor
2 orthogonal factors
2 oblique factors
3 oblique factors
2,434.63
971.05
970.61
818.97
344
344
343
341
.643
.867
.868
.892
.629
.888
.888
.908
.649
.860
.859
.878
.45!
.835
.835
.873
NNFI
Note. GFI = goodness of fit index. NNFI = nonnormed fit index. The NNFI was not applicable to the null
model.
mitment constructs over time were all quite high. Only one
ance commitment the fit of the structural model was very good
(RNFI > .980). Parameters reflecting the stability of the corn-
Table 3
Standardized Parameter Estimates for the
Discussion
Item
ACS1
ACS2
ACS3
ACS4
ACS5
ACS6
ACS7
ACS8
CCS1"
CCS3"
CCS4"
CCS8
CCS2b
CCS5b
CCS6b
CCS7
Factor 1
(ACS)
.305
.340
.366
.226
.140
0.764
1.020
0.694
Factor 2
(CCLoAlt)
Factor 3
(CCHiSac)
1.289
1.372
1.310
1.019
clarity in the conceptualization and measurement of organizational commitment. The results of our confirmatory factor analyses of data obtained from several samples provide support
both for the distinction between affective and continuance
commitment and for the utility of the ACS and CCS (Allen &
Meyer, 1990; Meyer & Allen, 1984) in future commitment research.
Evidence was found to support McGee and Ford's (1987) contention that the CCS reflects two dimensions. Confirmatory
factor analyses performed on the CCS alone, and analyses of
the CCS together with the ACS, revealed that a model dividing
the CCS into two oblique factors provided the best fit to the
data. Across samples, however, analyses revealed considerable
covariance between the two continuance commitment factors.
1.172
1.428
1.400
0.962
Note. See McGee & Ford (1987) for items corresponding to these item
numbers. ACS = Affective Commitment Scale; CCS = Continuance
Commitment Scale; CGLoAlt = Continuance Commitment: Low Perceived Alternatives subscale; CCHiSac = Continuance Commitment:
High Personal Sacrifice subscale. Dashes indicate values that were
fixed at zero.
* Items included in McGee & Ford's CCLoAlt subscale. b Items included in McGee & Ford's CCHiSac subscale.
It appears from our results, therefore, that all eight items of the
CCS reflect a common underlying theme, namely, cost associated with leaving an organization. Two sources of this cost,
lack of alternatives and personal sacrifice (e.g., loss of side bets),
seem to be sufficiently well represented by items in the scale
that, when permitted, they define separate but highly related
factors.
McGee and Ford (1987) noted that the CCHiSac (personal
sacrifice) subscale corresponds most closely to Becker's (1960)
definition of commitment and questioned whether commitment based on a lack of alternatives should be considered
717
Table 4
Standardized Parameter Estimates Reflecting Variances and Covariance of the Latent
Commitment Hiriables in the Oblique Three-Factor Solution
Sample 1
Factor
1
2
3
Factor 1 Factor 2
1.343'
-0.189*
0.260'
0.847*
0.828*
Sample 2
Factor 3
1.226*
Factor 1 Factor 2
1.052*
-0.091
0.298*
1.090*
0.919*
Sample 3
Factor 3
1.000*
Factor 1 Factor 2
0.553*
-0.163*
0.084
1.088*
0.714*
Factor 3
0.739*
Note. Factor 1 reflects affective commitment, and Factors 2 and 3 reflect the alternatives and personalsacrifice components of continuance commitment, respectively.
within the "side-bet view" (p. 640). We believe that it should, for
reasons suggested by Becker (1960):
then, justifies the inclusion of items assessing perceived alternatives in a measure of continuance (side-bet) commitment and
explains, in part, why the CCLoAlt and CCHiSac subscales are
so highly correlated.
In confirmatory factor analyses of the ACS and CCS together, we found some evidence consistent with McGee and
Ford's (1987) contention that the two components of continuance commitment are differentially related to affective commitment. The negative association between the alternatives
component and affective commitment was significant in two of
the three samples. In contrast, the link with the personal-sacrifice component was significantly positive in two of three samples. This suggests that the nature of the costs that tie employees to an organization may have implications for their affective
Table 5
Standardized Parameter Estimates for Nonrecursive Models
Sample 3
Sample 1
(#=301)
Parameter CCLoAlt
4,
*3
*5
*6
7i
72
73
74
0,
02
fl
*\
1 month
(N= 142)
6 months
(N = 245)
1 1 months
(N = 203)
.467*
-.066
.180*
-.107
.140*
-.114
.467*
-.066
.180*
-.107
.140*
-.114
.565*
-.147
-.178*
-.080
-.143
-.133
.565*
-.147
-.178*
-.080
-.143
-.133
.665*
.027
-.040
.071
-.201*
-.144
.665*
.027
-.040
.071
-.201*
-.144
.733*
-.037
.021
-.026
-.013
-.029
.733*
-.037
.021
-.026
-.013
-.029
.472*
.454*
-.507*
.251*
.439*
.430*
-.404*
.224*
.434*
.347*
-.442*
.331*
.425*
.346*
-.445*
.203*
.512*
.312*
-.475*
.124
.494*
.291*
-.494*
.067
.300*
.499*
-.337*
.201*
.311*
.504*
-.401*
.194*
-.401*
.189*
-.042
.183*
-.304*
.093
-.080
.175
-.405*
.120
.069
-.008
-.381*
-.006
.178
-.092
.462*
.695*
.406*
.785*
.569*
.615*
.525*
.761*
.522*
.639*
.480*
.742*
.437*
.699*
.446*
.794*
Note. Only parameters relevant to the structural model are included here. See Figure 1 for the variables and
paths to which parameter estimates apply. Components of continuance commitment (CC) were created by
including only relevant items (see Table 3) in the analyses. CCLoAlt = Low Perceived Alternatives subscale; CCHiSac = High Personal Sacrifice subscale.
* p < .05.
718
Table 6
Standardized Parameter Estimates for the Time-Lagged Model
Parameter
CCiLoAlt
CCrHiSac
*,
-.228*
.113
Ti
.746*
.901*
-.016
-.148*
.793*
.832*
-.011
-.106
.849*
.867*
-.061
-.047
.864*
.921*
.045
.014
.370*
.194*
.255*
.215
.379*
.310*
.253*
.152
72
"Is
74
0,
Hi
/},
A
I/-,
*2
\h
*4
particularly advisable for researchers who are interested in examining the specific causes or consequences of alternativesbased or sacrifice-based commitment.
In our examination of the structural relations among the
commitment constructs, the most consistent finding to emerge
from analyses of short-term effects was a significant negative
Note. Sample size for these analyses was 142. Only parameters relevant
to the structural model are included here. See Figure 2 for the variables
and paths to which parameterestimalesapply. Componentsofcontinuance commitment were created by including only relevant items (see
Table 3) in the analyses. CGLoAlt = Continuance Commitment: Low
Perceived Alternatives subscale; COHiSac = Continuance Commitment: High Personal Sacrifice subscale.
* p < .05.
lack of alternatives.
Interestingly, being affectively committed did not have a similar influence on employees' perceptions of the extent to which
zation is less consistent with the belief that it is the only alternative than it is with the belief that leaving would require personal
719
720
live attachment. Interestingly, both sets of effects are explainable from a dissonance-reduction, or self-justification, perspective.
References
Allen, N. I, & Meyer, J. P. (1990). The measurement and antecedentsof
affective, continuance, and normative commitment to the organization. Journal of Occupational Psychology, 63,1-18.
Allen, N. J., & Smith, J. (1987, June). An investigation of "extra-role"
behaviours within organizations. Paper presented at the 48th Annual
Convention of the Canadian Psychological Association, Vancouver,
British Columbia.
Anderson, J. C, &Gerbing, D. W(1988). Structural equation modeling
in practice: A review and recommended two-step approach. Psychological Bulletin, 103. 411-423.
Anderson, J. G. (1978). Causal models in educational research: Nonrecursive models. American Educational Research Journal, 15, 81-97.
Aven, F. F. (1988). A methodological examination of theattitudinaland
behavioral components of organizational commitment. Dissertation
Abstracts International, 40, 3420A. (University Microfilms No.
8902874)
Bateman, T. S., & Strasser, S. (1984). A longitudinal analysis of the
antecedents of organizational commitment. Academy of Management Journal, 27, 95-112.
Becker, H. S. (1960). Notes on the concept of commitment. American
Journal of Sociology, 66, 32-42.
Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness
of fit in the analysis of covariance structures. Psychological Bulletin,
88, 588-606.
Bollen, K. A. (1989). Structural equations with latent variables. New
York: Wiley.
Burt, R. S. (1976). Interpretational confounding of unobserved variables in structural equation models. Sociological Measures and Research, 5, 3-52.
Gollob, H. F, & Reichardt, C. S. (1987). Taking account of time lags in
causal models. Child Development, 58, 80-92.
James, L. R., Mulaik, S. A., & Brett, J. M. (1982). Causal analysis:
Assumptions, models, and data. Beverly Hills: Sage.
Joreskog, K. G., & Sorbom, D. (1986). LISKEL: Analysis of linear structural relationships by maximum likelihood, instrumental variable*,
and least squares methods (4th ed.). Mooresville, IN: Scientific Software.
Joreskog, K.. G., & Sorbom, D. (1989). LISREL 7 User's Guide. Mooresville, IN: Scientific Software.
Marsh, H. W, Balla, J. R., & McDonald, R. (1988). Goodness-of-fit
indexes in confirmatory factor analysis: The effect of sample size.
Psychological Bulletin. 103, 391-410.
Maruyama, G., & McGarvey, B. (1980). Evaluating causal models: An
application of maximum likelihood analysisof structural equations.
Psychological Bulletin, 87, 502-512.