discriminant validation of measures of job satisfaction, job involvement, and organizational...

7
Journal of Applied Psychology 1988, Vol. 73, No. 2,139-145 Copyright 1988 by the American Psychological Association, Inc. 0021-9010/88/$00.75 Discriminant Validation of Measures of Job Satisfaction, Job Involvement, and Organizational Commitment Paul P. Brooke, Jr., and Daniel W. Russell College of Medicine University of Iowa James L. Price Department of Sociology University of Iowa Discriminant validity of measures of job satisfaction, job involvement, and organizational commit- ment was empirically evaluated using data collected from a sample of 577 full-time employees of a 327-bed Veterans Administration Medical Center, The LISREL vi computer program was used to conduct a confirmatory factor analysis of items from measures of these three concepts and to evalu- ate relations between other job-related variables and the three attitudinal measures. Results of these analyses indicated that the measures of job satisfaction, job involvement, and organizational com- mitment assess empirically distinct concepts. Despite the plausibility of commonly accepted conceptual distinctions between the attitudes of job satisfaction, job in- volvement, and organizational commitment (Kanungo, 1982; Locke, 1976; Mowday, Porter, & Steers, 1982), surprisingly little rigorous empirical evidence of their discriminant validity has been reported (Blau, 1985;Morrow, 1983). In keeping with sep- arate research traditions regarding each construct, studies have reported correlations in the range of .50 between job satisfac- tion and job involvement, job satisfaction and organizational commitment, or job involvement and organizational commit- ment (Cheloha & Farr, 1980; Gechman & Wiener, 1975; Hall & Schneider, 1972; Mowday et al., 1982; Mowday, Steers, & Porter, 1979; Rabinowitz & Hall, 1977; Weissenberg & Gruen- feld, 1968; Wood, 1974). Research has also indicated similar patterns of relations between measures of these three attitudinal variables and other job-related variables (e.g., absenteeism and turnover; see reviews by Brooke, 1986a, 1986b). However, the three variables have rarely been included in the same study or measured simultaneously within a single sample (Kanungo, 1982; Morrow, 1983; Mowday et al., 1979; Rabinowitz & Hall, 1977; Steers, 1977). In the absence of rigorous empirical evi- dence of their discriminant validity, the substantial associations that have consistently been found between measures of the three constructs and the similarities in their relations with other vari- ables raise the possibility that job satisfaction, job involvement, and organizational commitment may not be empirically dis- tinct. The increasing emphasis on multivariate models of organiza- Paul Brooke is a Lieutenant Colonel in the U.S. Army Medical Ser- vice Corps. This article is based on his doctoral dissertation in Hospital and Health Administration at the University of Iowa, which was per- formed under the guidance of James Price and Daniel Russell. The research was conducted under the auspices of the U.S. Army Long-Term Civilian Training Program. Opinions expressed herein do not reflect an official position of the Department of Defense. Correspondence concerning this article should be addressed to Dan- iel W. Russell, Graduate Program in Hospital and Health Administra- tion, College of Medicine, University of Iowa, Iowa City, Iowa 52242. tional behavior in such areas as employee absenteeism (Brooke, 1986a; Steers & Rhodes, 1978, 1984) and turnover (Bluedorn, 1982; Mobley, 1982; Price & Mueller, 1981, 1986a) make it likely that researchers will incorporate measures of job satisfac- tion, job involvement, and organizational commitment in sin- gle studies that address comprehensive models of organiza- tional behavior. It therefore becomes important that an empiri- cal basis be demonstrated for the assumption that these measures assess distinct attitudinal constructs. Distinctions Between the Constructs The attitudes of job satisfaction (Brayfield & Rothe, 1951; Locke, 1976), job involvement (Kanungo, 1982; Lawler& Hall, 1970; Lodahl & Kejner, 1965; Rabinowitz & Hall, 1977), and organizational commitment (Mowday etal., 1979; Steers, 1977) have generally been considered to represent distinct constructs (Blau, 1985; Hammer, Landau, & Stern, 1981; Kanungo, 1982; Locke, 1976; Mowday etal., 1982;Siegel&Ruh, 1973; Wiener & Vardi, 1980). As a positive emotional state reflecting an affective response to the job situation (Locke, 1976), job satis- faction traditionally has been distinguished from job involve- ment, which is defined as a cognitive belief state reflecting the degree of psychological identification with one's job (Kanungo, 1982; Lawler & Hall, 1970; Locke, 1976; Rabinowitz & Hall, 1977). Although both constructs refer to the specific job, dis- tinctions between the emotional state of liking one's job (job satisfaction) and the cognitive belief state of psychological iden- tification with one's job (job involvement) have been advanced for some time (Locke, 1976; Kanungo, 1982). Organizational commitment has been defined as the relative strength of an individual's identification with and involvement in a particular organization, which is characterized by belief in and acceptance of organizational goals and values, willingness to exert effort on behalf of the organization, and a desire to maintain membership in the organization (Mowday et al., 1982, p. 27). Mowday et al. (1982) argued that, with its focus on the organization as a whole rather than the specific job and emphasis on congruence between individual and organizational 139

Upload: james-l

Post on 08-Feb-2017

216 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Discriminant validation of measures of job satisfaction, job involvement, and organizational commitment

Journal of Applied Psychology1988, Vol. 73, No. 2,139-145

Copyright 1988 by the American Psychological Association, Inc.0021-9010/88/$00.75

Discriminant Validation of Measures of Job Satisfaction, JobInvolvement, and Organizational Commitment

Paul P. Brooke, Jr., and Daniel W. RussellCollege of MedicineUniversity of Iowa

James L. PriceDepartment of Sociology

University of Iowa

Discriminant validity of measures of job satisfaction, job involvement, and organizational commit-ment was empirically evaluated using data collected from a sample of 577 full-time employees of a327-bed Veterans Administration Medical Center, The LISREL vi computer program was used toconduct a confirmatory factor analysis of items from measures of these three concepts and to evalu-ate relations between other job-related variables and the three attitudinal measures. Results of theseanalyses indicated that the measures of job satisfaction, job involvement, and organizational com-

mitment assess empirically distinct concepts.

Despite the plausibility of commonly accepted conceptual

distinctions between the attitudes of job satisfaction, job in-

volvement, and organizational commitment (Kanungo, 1982;

Locke, 1976; Mowday, Porter, & Steers, 1982), surprisingly little

rigorous empirical evidence of their discriminant validity has

been reported (Blau, 1985;Morrow, 1983). In keeping with sep-

arate research traditions regarding each construct, studies have

reported correlations in the range of .50 between job satisfac-

tion and job involvement, job satisfaction and organizational

commitment, or job involvement and organizational commit-

ment (Cheloha & Farr, 1980; Gechman & Wiener, 1975; Hall

& Schneider, 1972; Mowday et al., 1982; Mowday, Steers, &

Porter, 1979; Rabinowitz & Hall, 1977; Weissenberg & Gruen-

feld, 1968; Wood, 1974). Research has also indicated similar

patterns of relations between measures of these three attitudinal

variables and other job-related variables (e.g., absenteeism and

turnover; see reviews by Brooke, 1986a, 1986b). However, the

three variables have rarely been included in the same study or

measured simultaneously within a single sample (Kanungo,

1982; Morrow, 1983; Mowday et al., 1979; Rabinowitz & Hall,

1977; Steers, 1977). In the absence of rigorous empirical evi-

dence of their discriminant validity, the substantial associations

that have consistently been found between measures of the three

constructs and the similarities in their relations with other vari-

ables raise the possibility that job satisfaction, job involvement,

and organizational commitment may not be empirically dis-

tinct.

The increasing emphasis on multivariate models of organiza-

Paul Brooke is a Lieutenant Colonel in the U.S. Army Medical Ser-vice Corps. This article is based on his doctoral dissertation in Hospitaland Health Administration at the University of Iowa, which was per-formed under the guidance of James Price and Daniel Russell.

The research was conducted under the auspices of the U.S. ArmyLong-Term Civilian Training Program. Opinions expressed herein do

not reflect an official position of the Department of Defense.Correspondence concerning this article should be addressed to Dan-

iel W. Russell, Graduate Program in Hospital and Health Administra-tion, College of Medicine, University of Iowa, Iowa City, Iowa 52242.

tional behavior in such areas as employee absenteeism (Brooke,

1986a; Steers & Rhodes, 1978, 1984) and turnover (Bluedorn,

1982; Mobley, 1982; Price & Mueller, 1981, 1986a) make it

likely that researchers will incorporate measures of job satisfac-

tion, job involvement, and organizational commitment in sin-

gle studies that address comprehensive models of organiza-

tional behavior. It therefore becomes important that an empiri-

cal basis be demonstrated for the assumption that these

measures assess distinct attitudinal constructs.

Distinctions Between the Constructs

The attitudes of job satisfaction (Brayfield & Rothe, 1951;

Locke, 1976), job involvement (Kanungo, 1982; Lawler& Hall,

1970; Lodahl & Kejner, 1965; Rabinowitz & Hall, 1977), and

organizational commitment (Mowday etal., 1979; Steers, 1977)

have generally been considered to represent distinct constructs

(Blau, 1985; Hammer, Landau, & Stern, 1981; Kanungo, 1982;

Locke, 1976; Mowday etal., 1982;Siegel&Ruh, 1973; Wiener

& Vardi, 1980). As a positive emotional state reflecting an

affective response to the job situation (Locke, 1976), job satis-

faction traditionally has been distinguished from job involve-

ment, which is defined as a cognitive belief state reflecting the

degree of psychological identification with one's job (Kanungo,

1982; Lawler & Hall, 1970; Locke, 1976; Rabinowitz & Hall,

1977). Although both constructs refer to the specific job, dis-

tinctions between the emotional state of liking one's job (job

satisfaction) and the cognitive belief state of psychological iden-

tification with one's job (job involvement) have been advanced

for some time (Locke, 1976; Kanungo, 1982).

Organizational commitment has been defined as the relative

strength of an individual's identification with and involvement

in a particular organization, which is characterized by belief in

and acceptance of organizational goals and values, willingness

to exert effort on behalf of the organization, and a desire to

maintain membership in the organization (Mowday et al.,

1982, p. 27). Mowday et al. (1982) argued that, with its focus

on the organization as a whole rather than the specific job and

emphasis on congruence between individual and organizational

139

Page 2: Discriminant validation of measures of job satisfaction, job involvement, and organizational commitment

140 P. BROOKE, JR., D. RUSSELL, AND 3. PRICE

goals, the attitude of attachment or loyalty to the employing

organization represented by organizational commitment is

conceptually distinct in its focus and time frame from the job-

specific attitudes of job satisfaction and job involvement. Sim-

ilar arguments for distinguishing organizational commitment

from the other two constructs on the basis of differences in their

referent objects have also been advanced (Kanungo, 1982;

Locke, 1976; Price & Mueller, 1986b; Steers, 1977).

Moderate zero-order correlations in the range of .30 to .56

between job satisfaction (particularly overall satisfaction or sat-

isfaction with the work itself), job involvement, and organiza-

tional commitment have been observed repeatedly in studies

that have investigated relations between these constructs (Chel-

oha&Farr, 1980; Gechman & Wiener, 1975; Hall & Schneider,

1972; Mowday et al., 1979, 1982; Rabinowitz & Hall, 1977;

Weissenberg & Gruenfeld, 1968; Wood, 1974). In one of the

rare studies that included measures (abbreviated) of all three

constructs within a single sample, Hammer et al. (1981) re-

ported correlations of .47 between job satisfaction and job in-

volvement, .51 between job satisfaction and organizational

commitment, and .37 between job involvement and organiza-

tional commitment. In the absence of a well-developed theory

of causal relations among the three constructs (Bateman &

Strasser, 1984; Kanungo, 1982; Locke, 1976; Mowday et al.,

1982; Steers, 1977), consistent correlations of this magnitude

between different measures of the constructs have given rise to

concern over the potential for serious concept redundancy be-

tween job satisfaction and job involvement (Cheloha & Farr,

1980; Lawler & Hall, 1970; Saal, 1978) and between job in-

volvement and organizational commitment (Morrow, 1983).

Because the presence of even acceptable levels of random mea-

surement error (i.e., reliabilities of .80) considerably attenuates

the extent to which correlations reflect actual relations between

true scores on measures (Carmines & Teller, 1979), it seems

reasonable to assert that the moderate intercorrelations that

have been observed in previous studies represent lower bound

estimates of the actual relations among these constructs.

Although they both refer to the specific job and have generally

been considered to share common antecedents (Rabinowitz &

Hall, 1977), previous empirical evidence in support of retain-

ing a distinction between job satisfaction and job involvement

appears to be fairly persuasive. Lawler and Hall (1970) reported

factor analytic and correlational evidence that supported their

conceptual distinction between job satisfaction and job involve-

ment. Items pertaining to these two constructs and a third index

of intrinsic motivation loaded on three separate factors with

clearly interpretable structures. In addition, the three variables

differentially correlated with selected job characteristics. Al-

though they have not reported the use of factor analysis to assess

discriminant validity, several subsequent correlational studies

have concluded that job satisfaction and job involvement can

be differentiated (Blau, 1985; Gechman & Wiener, 1975; Gorn

& Kanungo, 1980; Wood, 1974).

The case for discriminant validity between organizational

commitment and the other two attitudinal constructs is not as

well developed. The arguments made by the proponents of the

Organizational Commitment Questionnaire (Mowday et al.,

1982, pp. 27-28) contain semantic ambiguities that suggest two

areas of concern. First, organizational commitment is described

as an affective response. Such a description raises a question

as to whether organizational commitment is different from job

satisfaction, which is also described as an affective response.

Second, Mowday et al. (1982) incorporate the "willingness to

exert effort" as one of the dimensions of organizational com-

mitment. As a commonly accepted definition of motivation

(Price, 1972; Price & Mueller, 1986b), the willingness to exert

effort traditionally has been viewed as a closely related outcome

of job involvement (Kanungo, 1982; Lawler & Hall, 1970; Ra-

binowitz & Hall, 1977).

In their validation of the Organizational Commitment Ques-

tionnaire, Mowday et al. (1979, 1982) clearly sought to deal

with the issue of discriminant validity among measures of job

satisfaction, job involvement, and organizational commitment.

However, none of their studies collected data regarding the three

variables within a single sample. Thus, the authors were forced

to rely on the analysis of correlations within and between sam-

ples to support their argument that the three constructs are em-

pirically distinct.

Her extensive review and comparison of the major forms of

"work commitment" used in organizational research (which in-

cluded job involvement and organizational commitment) led

Morrow (1983) to three conclusions that are relevant to the

present discussion. First, all of the measures analyzed were

marked by some concept redundancy. Second, the issue of

whether respondents can empirically discriminate among what

researchers see as logically distinct conceptualizations about

their values, jobs, and organizations has not yet been resolved.

Third, at a minimum, empirical comparisons of different forms

of "work commitment" within a single sample are in order

(Morrow, 1983, p. 497).

The present investigation was designed to evaluate the dis-

criminant validity of measures of job satisfaction, job involve-

ment, and organizational commitment within a single sample.

Two sets of analyses were conducted to address this issue. First,

a confirmatory factor analysis was conducted that tested

whether measures of these constructs were assessing distinct at-

titudinal dimensions. Second, relations between these three at-

titudinal variables and a number of other job-related variables

were examined. If measures of these three attitudinal variables

are assessing distinct constructs, then we would expect to find

that the three measures are differentially related to other job-

related variables.

Method

Sample

Data for the present analysis were collected as part of a larger studyof employee absenteeism among a sample of full-time employees of a327-bed Veterans Administration Medical Center located in the upperMidwest (Brooke, 1986b). In addition to blue-collar workers, clericalemployees, and registered nurses, the sample included managers andnon-nursing clinical staff. Questionnaires were distributed to all full-time employees (excepting physicians and dentists) through the organi-zational distribution system and returned by mail to Paul Brooke at the

University of Iowa.A total of 577 usable responses to the survey yielded a response rate of

74.5%. This response rate compares favorably with other mail surveys(Dillman, 1978). The sample consisted of 197 men (34.1%) and 380

Page 3: Discriminant validation of measures of job satisfaction, job involvement, and organizational commitment

DISCRIMINANT VALIDATION OF ATTITUDE MEASURES 141

women (65.9%), of whom 346 (60.0%) were administrators and clinicalstaff, 151 (26.2%) were registered nurses, and 80 (13.8%) were nonpro-fessional employees. Sample representativeness appeared to be ade-quate on the basis of comparisons of sample and population distribu-

tions by sex, pay category, and organizational subunit.

Measures

Multiple-item survey measures were used to operationalize the threeattitudinal variables of interest and other variables (i.e., routinization,

centralization, distributive justice, role stress, work involvement, and

kinship responsibility) that have generally been considered to be relatedto job satisfaction, job involvement, and organizational commitment(Kanungo, 1982; Locke, 1976;Mowdayetal., 1982; Rabinowitz& Hall,1977). In addition to the survey measures, pay (annual salary obtainedfrom an organizational roster) was also included in the analysis.

Job satisfaction was operationalized using a 6-item index that Priceand Mueller (1981,1986b) adapted from Brayfield and Rothe (1951) tomeasure the extent of global satisfaction with the job. These items havedemonstrated satisfactory levels of reliability and validity in other re-

search (Price & Mueller, 1981, 1986a; Sorensen, 1985; Wakefleld,1982). Job involvement was assessed using the 10-item index developedby Kanungo (1982) to measure the degree to which the individual iden-tifies with his or her present job. Kanungo (1982) reported evidencesupporting the reliability and validity of this measure. Organizationalcommitment was operationalized using the 9-item version of the Organ-izational Commitment Questionnaire developed by Porter, Steers,Mowday, and Boulian (1974) to measure the extent to which the individ-

ual identifies with organizational goals, is willing to exert effort on be-half of the organization, and intends to remain a member of the organi-zation (Mowday etal., 1982). The reliability and validity of the 15-itemversion of this index have been demonstrated (Ferris & Aranya, 1983;

Mowday et al., 1979). Acceptable results on the basis of the 9-item ver-sion have also been reported (Angle & Perry, 1981; Price & Mueller,1981,1986a; Sorensen, 1985; Wakefield, 1982).

Routinization was assessed using a 6-item index developed by Priceand Mueller (1981) to measure the degree to which tasks are repetitive.This index has consistently demonstrated satisfactory levels of reliabil-ity and validity in other research (Price & Mueller, 1981, 1986a; Sor-ensen, 1985; Wakefield, 1982). Centralization was operationalized using

a 4-item index developed by Price and Mueller (1981) to measure thedegree to which power is concentrated in the organization. These itemshave demonstrated satisfactory levels of reliability and validity in otherstudies (Price & Mueller, 1981, 1986a; Wakefield, 1982). Distributivejustice was assessed using a 6-item index adopted from Price andMueller (1981,1986a). This index measures the degree of fairness withwhich organizational rewards are related to performance inputs by or-ganizational members. Sorensen (1985) reported satisfactory levels of

reliability and validity for this index in a previous study.Role stress was operationally denned using a combined index of role

ambiguity (three items) and role conflict (three items) adopted fromKahn, Wolfe, Quinn, and Snoek (1964) and Rizzo, House, and Lirtz-

man (1970). A confirmatory factor analysis of the six items using theLISREL VI program supported the existence of two factors that were cor-related at .88, suggesting that the measures were not assessing distinctconstructs. Hem loadings for a one-factor model ranged from .60 to.75. On the basis of this empirical evidence and a reexamination of theconflict items, it was concluded that all six items were measuring a singleconstruct rather than the two distinct constructs that had been in-

tended.Work involvement was measured using a 6-item index developed by

Kanungo (1982) to measure the extent to which the work role occupiesa position of centrality in the individual's life. Work involvement refersto a personal code of ethics regarding work in general (normative beliefs)

Table 1Descriptive Statistics

Variable

Pay"Job satisfactionJob involvementOrganizational commitmentRoutinizationCentralizationDistributive j usticeRole stressWork involvementKinship responsibility

M

9.9020.7127.2928.2914.6010.3317.7515.0917.642.34

SD

0.324.676.265.783.653.296.124.234.261.53

8 This variable represents the natural log of pay, which was used in theanalysis.

and is distinguished from job involvement, which refers to cognitivebeliefs regarding a specific job (Kanungo, 1982, p. 116). A confirmatoryfactor analysis of the items measuring work involvement and job in-

volvement replicated the findings of Kanungo (1982) concerning thepresence of two correlated but distinct factors.

Kinship responsibility refers to family obligations arising from kin-ship groups within the local community. Following the approach of

Price and Mueller (1981,1986a), this variable was operationalized as acomposite index based on the sum of marital status (0 = not married,1 = married), presence of children under 7 years of age (0 = none, 1 =one child, 2 = two or more children), presence of children from 7 to 17years of age (0 = none, 1 = one child, 2 = two or more children), presenceof adult relatives other than spouse (0 = none, 1 = one ormorerelatives),

respondent's relatives residing within 25 miles (0 = none, 1 = one or

more), and spouse's relatives residing within 25 miles (0 = none, I =one or more). The reliability and validity of this measure appear to beadequate on the basis of previous research (Curry, Wakefield, Price,Mueller, & McCloskey, 1985; Price & Mueller, 1981, 1986a). Descrip-tive statistics for these measures on the basis of the study sample are

provided in Table 1.

Analytic Methods

The analysis used the maximum likelihood methods of LISREL vi (Jor-eskog & Sorbom, 1984) to assess the discriminant validity of the threeattitudinal variables. This computer program is designed to conductstructural equation analysis using latent variables. As such, the LISRELprogram represents a combination of factor and causal modeling analy-ses. In testing causal models, the program uses measured or manifestvariables as indicators of latent variables or factors. Relations amongthe latent variables are evaluated to test the hypothesized causal model.By using latent variables or factors in testing a causal model, the effectsof random measurement error, which tend to attenuate measures of

association, are removed from the analysis. As a result, the LISREL pro-gram provides less biased estimates of structural coefficients or corre-lations than do procedures that are based on measured or manifest vari-

ables.As its second major advantage, the LISREL program provides an over-

all chi-square test of the extent to which the hypothesized model (oralternative models) are able to account for relations among the mea-sured variables. This chi-square statistic is computed on the basis ofthe differences between the sample covariance matrix (i.e., the relationsamong the measured variables that are observed in the data) and a "re-

produced" covariance matrix (i.e., relations among the measured vari-ables that are predicted on the basis of the parameters of the model).One problem with the chi-square goodness-of-fit statistic is that it is

Page 4: Discriminant validation of measures of job satisfaction, job involvement, and organizational commitment

142 P. BROOKE, JR., D. RUSSELL, AND J. PRICE

sensitive to sample size and violations of the assumption of multivariatenormality (Bentler, 1983; Joreskog & Sorbom, 1984). As a result, trivialdifferences between the sample and reproduced covariance matrices arelikely to produce a statistically significant chi-square in a model with asmany variables and a sample size as large as those of the present study(Bentler, 1980; Bentler & Bonett, 1980; Cutrona, Russell, & Jones,1984).

Because of this limitation of the chi-square test, the evaluation ofthe goodness of model fit was based on considerations other than thestatistical significance of the chi-square. One statistic that was com-puted was the ratio of the chi-square to degrees of freedom (Carmines& Mclver, 1981; Hoetler, 1983). A ratio of less than 2.0 is generallyconsidered indicative of an excellent model fit. Second, the normed fitindex (A), developed by Bentler and Bonett (1980), was computed bycomparing the chi-square for the model to the chi-square for a nullmodel that hypothesized complete independence among the measuredvariables. One advantage of A is that it is not affected by sample size orthe number of degrees of freedom for the model.

Finally, we also evaluated differences in the chi-square statistics foralternative models. The latter approach had particular relevance fortesting discriminant validity, because it is a means of evaluating the ex-tent to which alternative models provide an improved fit to the data.Bentler (1980) has suggested that, if one can specify an alternative modelthat is a subset of an initial model, the difference in chi-square valuesbetween the two models is distributed as a chi-square with degrees offreedom equal to the difference in degrees of freedom for the models(Bentler, 1980, pp. 428-429). As discussed later, the analysis of differ-ences in chi-squares between alternative models formed a statistical ba-sis for testing hypotheses related to the discriminant validity of the threeattitudinal measures.

Two procedures were used to evaluate discriminant validity. The firstinvolved a confirmatory factor analysis of the items pertaining to theindices of job satisfaction, job involvement, and organizational commit-ment. This analysis compared the fit of a three-factor model (where thefactors represented each of the attitudinal constructs) that permittedcorrelations between the factors to the fit of an alternative model thathypothesized a single factor as underlying the attitudinal variables.

The second procedure investigated relations between other job-re-lated variables (i.e., routinization, centralization, distributive justice,role stress, work involvement, kinship responsibility, and pay) and thethree attitudinal variables of interest (i.e., job satisfaction, job involve-ment, and organizational commitment). This analysis tested the hy-pothesis that the correlations were the same between these job-relatedvariables and each of the attitudinal variables. Assuming that the mea-sures are assessing distinct constructs, we would expect to find that mea-sures of these constructs relate differently to measures of other job-re-lated variables. This second procedure represents a more rigorous testof discriminant validity for the three attitudinal measures. Becauseeach measure uses a different method of assessment (e.g., differing re-sponse formats), distinct but correlated factors could be found from theconfirmatory factor analysis that simply reflected different methods ofassessment. In order to establish discriminant validity, therefore, it isalso necessary to demonstrate different patterns of association withmeasures of other job-related variables for each attitudinal variable.

The technique for testing the hypothesis that the relations are equalbetween each of the job-related variables and the three attitudinal vari-ables involved comparing the chi-square statistics for two models. Inthe first model, the correlations between the job-related variables andjob satisfaction, job involvement, and organizational commitment werefreely estimated using the LISREL program. In the second model, thecorrelations between the job-related variables and the three attitudinalvariables were specified as being equal (e.g., routinization had the samerelation with job satisfaction, job involvement, and commitment). Anonsignificant difference in the goodness of fit between these two

Table 2

Confirmatory Factor Analysis Results

Factor loadings

Indicator SAT JI COM

SAT1SAT2SATSJI1JI2JI3COM1COM2COM3

.825

.808

.894.794.829.800

.860

.829

.876

Note. SAT = Job Satisfaction; JI = Job Involvement; COM = Organiza-tional Commitment. All factor loadings were statistically significant,p<.001.

models would constitute a failure to reject the null hypothesis. Inabilityto reject the hypothesis of equal correlations would raise serious con-cerns regarding the discriminant validity of these measures.

For these LISREL analyses, the individual items from the measureswere combined to form three indicators per construct. This was doneto reduce the ratio between the number of subjects and the number ofparameters to be estimated by the model. Bentler (1985) has recom-mended that a ratio of at least five subjects per parameter be maintainedin order to derive reliable estimates of the parameters. These ratiosranged from 5.95 to 31.38 for the models that were tested.

In order to develop three indicators for each construct, a one-factormodel was fit to the items from each measure using the LISREL program.Three indicators for each variable were then formed on the basis of thefactor analysis results. Those items with the highest and lowest loadingswere combined first, items with the next highest and lowest loadingswere combined next, and so on, until all items pertaining to a givenconstruct had been assigned to one of the three indicators. Scores foreach indicator were then computed as the mean of the scores on thepertinent items. Pay and kinship responsibility were treated as manifestvariables that were assumed to be measured without error. A covariancematrix of these indicators using listwise deletion of missing valuesformed the input for the LISREL analyses.

Results

Confirmatory Factor Analyses

The first model that was tested hypothesized that three corre-

lated factors underlie the measures of job satisfaction, job in-

volvement, and organizational commitment, with the factors

representing each of the attitudinal constructs. Estimates of the

factor loadings for the indicators of each variable are presented

in Table 2. This model yielded a x2(24, N = 565) = 58.25, p <

.001, for a chi-square/degrees of freedom ratio of 2.43 and a

normed fit index (A) of .981. On the basis of criterion that A be

greater than or equal to .90 described by Bentler and Bonett

(1980), this model appeared to provide an excellent fit to the

data. Correlations between factors were as follows: .591, be-

tween job satisfaction and job involvement; .552, between job

satisfaction and organizational commitment; and .547, between

job involvement and organizational commitment.

A second model was fit to these data that hypothesized a sin-

Page 5: Discriminant validation of measures of job satisfaction, job involvement, and organizational commitment

DISCRIMINANT VALIDATION OF ATTITUDE MEASURES 143

Table 3

LISREL Estimates of Relations Between Job-Related Variables

and Job Satisfaction, Job Involvement, and

Organizational Commitment

Job Job OrganizationalJob-related variable satisfaction involvement commitment

PayRoutinizationCentralizationDistributive justiceRole stressWork involvementKinship responsibility

.247**-.554*'-.560"

.393**-.386**

.339**

.050

.265*'-.425**-.385**

.364"-.101*

.728"-.047

.107*-.244**-.391"

.462"-.308**

.431"

.128*

*p<.05.**p<.001.

gle factor underlying the measures. This model specified that

the instruments represented alternative measures of the same

construct. If such a model fit the data as well as the three-factor

model, then we would conclude that the measures lacked dis-

criminant validity. This model provided a very poor fit to the

data, x2(27, N = 565) = 1050.01, p < .001, for a chi-square/

degrees of freedom ratio of 38.89 and a A of .655. A comparison

of the fit of these two models indicated that the three-factor

model provided a significantly better fit to the data, %2(3, N =

565) = 991.76, p < .001. Thus, we can reject the hypothesis that

a single factor underlies these measures.

Correlations With Other Job-Related Variables

Table 3 presents the LISREL estimates of correlations between

the seven job-related variables (i.e., routinization, centraliza-

tion, distributive justice, role stress, work involvement, kinship

responsibility, and pay) and the attitudes of job satisfaction, job

involvement, and organizational commitment. The model

summarized in Table 3 yielded a x2(256, N = 565) = 594.29,

p < .001, for a chi-square/degrees of freedom ratio of 2.32 and

a A of .936.

A review of the correlations presented in Table 3 provides

some evidence of differences in relations for the three attitudi-

nal constructs. Routinization had a strong negative relation

with job satisfaction and job involvement, whereas the negative

association with organizational commitment was much smaller

in magnitude. Role stress had strong negative correlations with

job satisfaction and organizational commitment, whereas the

association with job involvement was very weak. Work involve-

ment had positive correlations with all three attitudinal vari-

ables, but was much stronger in its relation with job involve-

ment. Kinship responsibility was significantly related to organi-

zational commitment, but not to the other two attitudinal

constructs. Individuals who reported greater kinship responsi-

bility also indicated greater commitment to the organization.

Finally, pay was moderately positively related to job satisfaction

and job involvement, whereas the association with organiza-

tional commitment was less strong.

A statistical test of the equality of relations between the job-

related variables and the attitudinal variables was conducted by

specifying a model in which the correlation of each job-related

variable with job involvement and organizational commitment

was constrained to equal the correlation of the job-related vari-

able with job satisfaction. Thus, for example, the correlation of

routinization with job involvement and organizational com-

mitment was specified as being equal to the correlation between

routinization and job satisfaction. This model also fit the data

well, yielding a *2(270, N = 565) = 824.61, p < .001, for a chi-

square/degrees of freedom ratio of 3.05 and a A of .911. A com-

parison of the fit between this model representing a hypothesis

of no differences between the attitudinal variables and the

model presented in Table 3 yielded a x2( 14, N = 565) = 230.32,

p < .001. The magnitude and statistical significance of this

difference in chi-squares was evidence that constraining the cor-

relations between the job-related variables and the attitudinal

variables to be equal resulted in a significantly poorer represen-

tation of the relations present in the data. Therefore, these re-

sults indicated that the measures of job satisfaction, job in-

volvement, and organizational commitment were assessing em-

pirically distinct constructs that were related differently to the

other job-related variables that were included in the analysis.

Discussion

The results of these analyses provide evidence that respon-

dents are able to distinguish between the extent to which they

like their job (satisfaction), the degree to which they are ab-

sorbed in or preoccupied with their job (involvement), and the

degree of attachment or loyalty they feel toward their employ-

ing organization (commitment). Confirmatory factor analysis

of the items from the three attitudinal measures provided evi-

dence of three empirically distinct constructs that were moder-

ately intercorrelated. These results are consistent with pre-

viously cited analyses of associations between the three vari-

ables. They also replicate the favorable findings of Mowday et

al. (1979, 1982) regarding the Organizational Commitment

Questionnaire, and of Kanungo (1982) concerning the index of

job involvement.

The ability to reject statistically a null hypothesis of equal

relations between the three attitudinal variables and seven job-

related variables also provides empirical support for the dis-

criminant validity of the measures of job satisfaction, job in-

volvement, and organizational commitment. These results are

further supported by the consistency of observed differences in

the correlations with theoretical arguments for conceptual dis-

tinctions between these job-related attitudes. For example, al-

though for routinization there were significant negative re-

lations with all three attitudinal variables, this variable was

strongly related to job satisfaction and job involvement and

only weakly related to organizational commitment. These find-

ings are consistent with the hypothesized job-specificity of job

satisfaction and job involvement and, therefore, support dis-

tinctions between organizational commitment and these other

attitudinal measures. The strong negative relation between role

stress and job satisfaction in conjunction with the weak relation

between role stress and job involvement is evidence that these

two job-specific attitudes are empirically distinct. There were

positive relations for work involvement with all three attitudi-

nal variables, but this variable was much more strongly related

to job involvement than to job satisfaction or organizational

Page 6: Discriminant validation of measures of job satisfaction, job involvement, and organizational commitment

144 P. BROOKE, JR., D. RUSSELL, AND J. PRICE

commitment. This is consistent with the long tradition that has

considered work values resulting from prior socialization into

"middle class work norms" to be a major determinant of job

involvement (Kanungo, 1982). The positive relation between

kinship responsibility and organizational commitment along

with the nonsignificant relation between kinship responsibility

and job satisfaction and job involvement are consistent with

the argument that attachment to the employing organization is

influenced by considerations of kinship ties to the local commu-

nity (Mowdayetal., 1982).

Note that the present results may overstate both the strength

of relations among the three job attitude variables and their re-

lations with the other job-related variables that were included

in the study. Because all of these data involved the same basic

method of assessment (i.e., self-reports of employees), shared

method variance may have inflated the relations among the la-

tent variables. Such a method artifact would not, of course,

affect our conclusions regarding discriminant validity of the at-

titudinal measures. However, use of multiple methods of assess-

ment (e.g., behavioral observations, ratings by observers) would

be necessary to develop more accurate estimates of the relations

among the latent variables. The latent variable methods used

in the LISREL program are uniquely suited to analyzing such

multimethod data.

Three suggestions for future research derive from the present

findings. First, there is a need to replicate these findings. The

use of a single federal hospital as the research site is a potential

limitation on the generalizability of these results. Attempts

should be made to replicate these analyses in other types of hos-

pitals and among organizations in other industries. Second, fu-

ture research efforts aimed at assessing relations between other

job-related variables and the three attitudes should include at

least two additional variables—work group integration and op-

portunities for alternative employment—whose absence from

the present analysis constitutes a limitation. Third, the present

analysis specifically avoided statements or hypotheses regarding

possible causal relations among the attitudinal constructs. On

the basis of the empirical evidence of discriminant validity

demonstrated in this study, future research should build on the

conceptual distinctions between these three attitudinal vari-

ables, which have long been suggested by the literature, and in-

vestigate the possibility of causal interrelations as an explana-

tion for the substantial associations that have repeatedly been

found to exist between job satisfaction, job involvement, and

organizational commitment.

References

Angle, H. L., & Perry, J. L. (1981). An empirical assessment of organi-zational commitment and organizational effectiveness. Administra-

tive Science Quarterly, 26, 1-14.Bateman, T. S., & Strasser, S. (1984). A longitudinal analysis of the ante-

cedents of organizational commitment. Academy of Management

Journal, 27, 95-112.Bentler, P. M. (1980). Multivariate analysis with latent variables: Causal

modeling. Annual Review of Psychology, 31, 419-456.Bentler, P. M. (1983). Some contributions to efficient statistics in struc-

tural models: Specification and estimation of moment structures.

Psychometrika, 48, 493-517.Bentler, P. M. (1985). Theory and implementation ofEQS:A structural

equations program. Los Angeles, CA: BMDP Statistical Software.

Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness

of fit in the analysis of covariance structures. Psychological Bulletin,88, 588-606.

Blau, G. J. (1985). A multiple study investigation of the dimensionaltyof job involvement. Journal of Vocational Behavior, 27, 9-36.

Bluedorn, A. C. (1982). The theories of turnover: Causes, effects, andmeaning. In S. B. Bacharach (Ed.), Research in the sociology of orga-

nizations (pp. 75-128). Greenwich, CT: JAI Press.

Brayfield, A. H., & Rothe, H. F. (1951). An index of job satisfaction.Journal of Applied Psychology, 35, 307-311.

Brooke, P. P., Jr. (1986a). Beyond the Steers and Rhodes model of em-ployee attendance. Academy of Management Review, 11, 345-361.

Brooke, P. P., Jr. (1986b). A causal model of employee absenteeism.

Unpublished doctoral dissertation, University of Iowa.

Carmines, E. G., & Mclver, J. P. (1981). Analyzing models with unob-served variables: Analysis of covariance structures. In G. W. Born-stedt & E. F. Borgatta (Eds.), Social measurement: Current issues (pp.65-116). Beverly Hills, CAi Sage.

Carmines, E. G., & Zeller, R. A. (1979). Reliability and validity assess-ment. Beverly Hills, CA: Sage.

Cheloha, R. S., & Farr, J. L. (1980). Absenteeism, job involvement andjob satisfaction in an organizational setting. Journal of Applied Psy-

chology, 65,467-473.

Curry, J. P., Wakefield, D. S., Price, J. P., Mueller, C. W., & McCloskey,J. C. (1985). Determinants of turnover among nursing departmentpersonnel. Research in Nursing and Health, 8,397-411.

Cutrona, C. E., Russell, D., & Jones, R. D. (1984). Cross-situationalconsistency in causal attributions: Does "attributional style" exist?Journal of Personality and Social Psychology, 47,1043-1058.

Dillman, D. A. (1978). Mail and telephone surveys: The total design

method. New York: Wiley.Ferris, K. R., & Aranya, N. (1983). A comparison of two organizational

commitment scales. Personnel Psychology, 36, 87-98.

Gechman, A. S., & Wiener, Y. (1975). Job involvement and satisfactionas related to mental health and personal time devoted to work. Jour-

nal of Applied Psychology, 60, 521-523.

Corn, G. J., & Kanungo, R. N. (1980). Job involvement and motiva-tion: Are intrinsically motivated managers more job involved? Or-ganizational Behavior and Human Performance, 26, 265-277.

Hall, D. T., & Schneider, B. (1972). Correlates of organizational identi-fication as a function of career pattern and organizational type. Ad-

ministrative Science Quarterly, 17, 340-350.

Hammer, T. H., Landau, J. C., & Stem, R. N. (1981). Absenteeismwhen workers have a voice: The case of employee ownership. Journalof Applied Psychology, 66, 561-573.

Hoetler, J. W. (1983). The analysis of covariance structures: Goodness-of-fit indices. Sociological Methods and Research, 11, 325-344.

Joreskog, K. G., & Sorbom, D. (1984). LISREL vi user's guide. Moores-

ville, IN: Scientific Software.Kahn, R. L., Wolfe, D. M., Quinn, R. P., &Snoek, J. D. (1964). Organi-

zational stress: Studies in role conflict and ambiguity. New \fork: Wi-

ley.Kanungo, R. N. (1982). Work alienation. New York: Praeger.

Lawler, E. E., & Hall, D. T. (1970). Relationship of job characteristicsto job involvement, satisfaction, and intrinsic motivation. Journal of

Applied Psychology, 54, 305-312.

Locke, E. A. (1976). The nature and causes of job satisfaction. InM. D. Dunnette (Ed.), Handbook of industrial and organizational

psychology (pp. 1293-1349). Chicago: Rand McNally.

Lodahl, T. M., & Kejner, M. (1965). The definition and measurementof job involvement. Journal of Applied Psychology, 49, 24-33.

Mobley, W. H. (1982). Employee turnover: Causes, consequences and

control. Reading, MA: Addison-Wesley.Morrow, P. C. (1983). Concept redundancy in organizational research:

Page 7: Discriminant validation of measures of job satisfaction, job involvement, and organizational commitment

DISCRIMINANT VALIDATION OF ATTITUDE MEASURES 145

The case of work commitment. Academy of Management Review, 8,

486-500.

Mowday, R. T., Porter, L. W., & Steers, R. M. (1982). Employee-organi-

zation linkages: The psychology of commitment, absenteeism and

turnover. New York: Academic Press.

Mowday, R. T., Steers, R. M., & Porter, L. W. (1979). The measurementof organizational commitment. Journal of Vocational Behavior, 14,

224-247.

Porter, L. W., Steers, R. M., Mowday, R. T., & Boulian, P. V. (1974).

Organizational commitment, job satisfaction, and turnover amongpsychiatric technicians. Journal of Applied Psychology, 59,603-609.

Price, J. L. (1972). Handbook of organizational measurement. Lexing-ton, MA: Heath.

Price, J. L., & Mueller, C. W. (1981). Professional turnover: The case ofnurses. New York: Spectrum.

Price, J. L., & Mueller, C. W. (1986a). Absenteeism and turnover among

hospital employees. Greenwich, CT: JAI Press.

Price, J. L., & Mueller, C. W. (1986b). Handbook of organizational

measurement. Marshfield, MA: Pittman.

Rabinowitz, S., & Hall, D. T. (1977). Organizational research on job

involvement. Psychological Bulletin, 84, 265-288.

Rizzo, J. R., House, R. J., & Lirtzman, S. I. (1970). Role conflict andambiguity in complex organizations. Administrative Science Quar-

terly, 15, 150-163.

Saal, F. E, (1978). Job involvement: A multivariate approach. Journal

of Applied Psychology, 63, 53-61.

Siegel, A. L., & Ruh, R. A. (1973). Job involvement, participation in

decision making, personal background and job behavior. Organize

tionat Behavior and'Human Performance, 9, 318-327.Sorensen, W. fl. (1985). A causal model of organizational commitment.

Unpublished doctoral dissertation, University of Iowa.

Steers, R. M. (1977). Antecedents and outcomes of organizational com-mitment. Administrative Science Quarterly, 22,46-56.

Steers, R. M., & Rhodes, S. R. (1978). Major influences on employeeattendance: Aprocess model. Journal of Applied Psychology, 63, 391-407.

Steers, R. M., & Rhodes, S. R. (1984). Knowledge and speculationabout absenteeism. In P. S. Goodman & R. S. Atkin (Eds.), Absentee-

ism (pp. 229-275). San Francisco: Jossey-Bass.Wakefield, D. S. (1982). Organizational commitment of full-time and

part-time registered nurses: A contingency approach. Unpublisheddoctoral dissertation, University oflowa.

Weissenberg, P., & Gruenfeld, L. W. (1968). Relationship between jobsatisfaction and job involvement. Journal of Applied Psychology, 52,469-473.

Wiener, Y., & Vardi, Y. (1980). Relationship between job, organizationaland career commitments and work outcomes: An integrative ap-proach. Organizational Behavior and Human Performance, 26, 81 -96.

Wood, D. A. (1974). Effect of worker orientation differences on job atti-

tude correlates. Journal of Applied Psychology, 59, 54-60.

Received April 23, 1986

Revision received June 23, 1987

Accepted July 9, 1987 •

Call for Nominations for Behavioral Neuroscience

The Publications and Communications Board has opened nominations for the editorship of

Behavioral Neuroscience for the years 1990-1995. RichardF. Thompson is the incumbent edi-

tor. Candidates must be members of APA and should be available to start receiving manuscripts

in early 1989 to prepare for issues published in 1990. Please note that the P&C Board encour-

ages more participation by women and ethnic minority men and women in the publication

process and would particularly welcome such nominees. Submit nominations no later than

August 1, 1988 to

Martha Storandt

Department of Psychology

Washington University

St. Louis, Missouri 63130

Other members of the search committee are Byron Campbell, Mortimer Mishkin, Mark Rosen-

zweig, and Shepard Siegel.