relationships among five forms of commitment: an empirical assessment

24
Relationships among five forms of commitment: an empirical assessment AARON COHEN * Department of Political Science, University of Haifa, Mount Carmel, Haifa 31905, Israel Summary This paper examined whether Morrow’s (1993) model of five universal forms of work commitment (aective organizational commitment, continuance organizational commit- ment, career commitment, job involvement and work ethic endorsement) has empirical support. The paper explored the discriminant validity among the five scales of these commitment forms. It also tested Morrow’s model regarding the interrelationships among them in comparison with an alternative conceptualization proposed by Randall and Cote (1991). Two hundred and thirty-eight nursing sta from two hospitals in western Canada were surveyed. Results of confirmatory factor analysis (LISREL VIII) and correlation analysis showed an acceptable discriminant validity among the five commitment foci. The results of path analysis (LISREL VIII) showed a poor fit with the data of Morrow’s model and a better fit with the alternative model of Randall and Cote. Revised models based on the modification indices from both models were tested, and supported the Randall and Cote approach. The paper concludes with implications regarding the continuing assessment of Morrow’s conceptualization to establish an acceptable definition and measurement of universal forms of work commitment. Copyright # 1999 John Wiley & Sons, Ltd. Introduction The subject of work commitment is of increasing concern among researchers and practitioners. Recent literature has tended to focus on the broader concept of work commitment that includes specific commitment objects such as the organization, work group, occupation, union, and one’s job (Blau, Paul and St. John, 1993; Randall and Cote, 1991). The importance of such research has been emphasized by Randall and Cote (1991) and Mueller, Wallace and Price (1992) who argued that we are still much in need of conceptual and empirical work in sorting out how forms of commitment are related and how they relate to work behavior. In fact, work commitment forms have been shown to predict important outcomes such as turnover, turnover intentions, performance, job satisfaction, prosocial organizational behavior, absenteeism, and tardiness * Correspondence to: Aaron Cohen, Department of Political Science, University of Haifa, Mount Carmel, Haifa 31905, Israel. Tel: 972 4 8240019. Fax: 972 4 8257785. E-mail: [email protected]. CCC 0894–3796/99/030285–24$17.50 Received 4 December 1995 Copyright # 1999 John Wiley & Sons, Ltd. Accepted 10 February 1997 Journal of Organizational Behavior J. Organiz. Behav. 20, 285–308 (1999)

Upload: aaron-cohen

Post on 06-Jun-2016

213 views

Category:

Documents


1 download

TRANSCRIPT

Relationships among ®ve forms ofcommitment: an empirical assessment

AARON COHEN*

Department of Political Science, University of Haifa, Mount Carmel, Haifa 31905, Israel

Summary This paper examined whether Morrow's (1993) model of ®ve universal forms of workcommitment (a�ective organizational commitment, continuance organizational commit-ment, career commitment, job involvement and work ethic endorsement) has empiricalsupport. The paper explored the discriminant validity among the ®ve scales of thesecommitment forms. It also tested Morrow's model regarding the interrelationshipsamong them in comparison with an alternative conceptualization proposed by Randalland Cote (1991). Two hundred and thirty-eight nursing sta� from two hospitals inwestern Canada were surveyed. Results of con®rmatory factor analysis (LISREL VIII)and correlation analysis showed an acceptable discriminant validity among the ®vecommitment foci. The results of path analysis (LISRELVIII) showed a poor ®t with thedata of Morrow's model and a better ®t with the alternative model of Randall and Cote.Revised models based on the modi®cation indices from both models were tested, andsupported the Randall and Cote approach. The paper concludes with implicationsregarding the continuing assessment of Morrow's conceptualization to establishan acceptable de®nition and measurement of universal forms of work commitment.Copyright # 1999 John Wiley & Sons, Ltd.

Introduction

The subject of work commitment is of increasing concern among researchers and practitioners.Recent literature has tended to focus on the broader concept of work commitment that includesspeci®c commitment objects such as the organization, work group, occupation, union, and one'sjob (Blau, Paul and St. John, 1993; Randall and Cote, 1991). The importance of such researchhas been emphasized by Randall and Cote (1991) and Mueller, Wallace and Price (1992) whoargued that we are still much in need of conceptual and empirical work in sorting out how formsof commitment are related and how they relate to work behavior. In fact, work commitmentforms have been shown to predict important outcomes such as turnover, turnover intentions,performance, job satisfaction, prosocial organizational behavior, absenteeism, and tardiness

* Correspondence to: Aaron Cohen, Department of Political Science, University of Haifa, Mount Carmel, Haifa 31905,Israel. Tel: 972 4 8240019. Fax: 972 4 8257785. E-mail: [email protected].

CCC 0894±3796/99/030285±24$17.50 Received 4 December 1995Copyright # 1999 John Wiley & Sons, Ltd. Accepted 10 February 1997

Journal of Organizational BehaviorJ. Organiz. Behav. 20, 285±308 (1999)

(Becker, 1992; Blau, 1986; Wiener and Vardi, 1980). These papers showed that a multivariateapproach to work commitment could predict outcomes better than each commitment separately.

In a recent publication Morrow (1993) proposed two ways to advance work commitmentresearch. First she suggested that among all foci of work commitment ®ve are basic in the sensethat they are relevant to as many employees as possible, and she termed them universal forms ofwork commitment. These are a�ective commitment to the organization, continuance commitmentto the organization, work ethic endorsement, career commitment, and job involvement. Morrow(1993) argued that these ®ve forms could be viewed as dimensions of work commitment in thesame way that the ®ve dimensions of job satisfaction within the Job Descriptive Index are viewed(Smith, Kendall and Hulin, 1969). The goal, Morrow suggested, should be a similar workcommitment scale that would make possible formulation of commitment pro®les, testing theirrelationships with organizational outcomes (Becker and Billings, 1993), thus allowing managersto pinpoint what forms of work commitment are less than optimal (Morrow, 1993). Second,Morrow proposed a conceptual model regarding how these forms are related to each other. Sheargued that a major question impeding understanding of work commitment is whether each formis independent or whether some are antecedents and consequences of others. This question hasimportant implications for the nature of the relationship among commitment foci and workoutcomes. Morrow (1993) concluded that these ®ve forms appear quite promising but need moreempirical substantiation. The goal of this paper is to validate Morrow's model of work commit-ment empirically. Because the main problem in work commitment research is concept redund-ancy and overlap (Morrow, 1983, 1993), this research will ®rst test the discriminant validity of the®ve commitment forms. Second, the interrelationships among these forms as suggested byMorrow will be tested against an alternative model proposed by Randall and Cote (1991). Our®ndings will indicate whether there is empirical support for Morrow's proposition and will thusdirect research to the commitment forms that should receive more attention, as well as to theirinterrelationships.

Universal forms of work commitment

An important condition for establishing the ®ve commitment forms as universal is that they mustbe distinguishable. In view of the ®nding that some forms of commitment are somewhatredundant and insu�ciently distinguishable to warrant continued separation (Morrow andGoetz, 1988; Morrow and McElroy, 1986; Morrow, Eastman and McElroy, 1991), Morrow(1993) emphasized the importance of establishing whether respondents' discriminant abilities aresu�ciently sensitive to allow them to report multiple work commitment attitudes accuratelywithin a single data collection. Therefore, an empirical test showing that these ®ve forms assessdistinct attitudinal constructs becomes vital. Thus, the ®rst goal of this research is to test thediscriminant validity of these ®ve commitments. First, a con®rmatory factor analysis isperformed. Furthermore, discriminant validity will be examined by testing the relationshipbetween the work commitment foci and a number of variables selected on the basis of priorempirical research and conceptual arguments. Assuming that the ®ve measures of commitmentassess distinct constructs, we would expect them to relate di�erently to measures of demographicand other job-related variables. As Brooke, Russell and Price (1988) pointed out, this procedurerepresents a more rigorous test of discriminant validity. The variables to be tested in theirrelationships to work commitment represent demographic characteristics (gender, children, age,education), work experience (tenure, years in occupation, job satisfaction and job tension), workoutcome (perceived performance) and nonwork domains (life satisfaction).

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

286 A. COHEN

Protestant work ethic (PWE) determinants are felt to be primarily a function of personalityand secondarily a function of culture (Morrow, 1983). The personality link is based on theobservation that ethical endorsement covaries with stable personality and demographic traits.The expectation here is that PWE will be related to demographic variables such as gender (male),education (negative), and age (positive). It is not expected to be related to work experiencevariables or to work outcomes (Morrow, 1983). Furnham (1990) argued for a positive spilloverbetween PWE and nonwork, which leads to the expectation of a positive relationship betweenPWE and life satisfaction. There is consensus that job involvement is a function of personality/individual di�erence and the work situation (Morrow, 1993). Thus demographic and workexperience variables are expected to relate to job involvement. Positive relationships are expectedwith age, tenure, years in occupation, education, having children, and gender (male). There is noevidence for a strong relationship between job involvement and performance (Morrow, 1993) orbetween job involvement and nonwork domains. Hence it is not expected to relate to perceivedperformance or to life satisfaction. As for career commitment, individual di�erences and situa-tional characteristics were suggested as its primary determinants (Blau, 1985). Therefore, for thedemographic variables the same pattern of relationship is expected for career commitment as forjob involvement. Career commitment is also expected to have a positive relationship with jobsatisfaction. No relationship was found between career commitment and performance (Meyer,Allen and Smith, 1993) or between career commitment and nonwork domains (Wallace, 1995),and therefore no relationship will be expected here with perceived performance and lifesatisfaction.

A�ective organizational commitment was found to be related to a wide variety of correlates.The literature (Mowday, Porter and Steers, 1982; Morrow, 1993) suggested that a�ectivecommitment is related to both demographic characteristics and work experience. A�ectivecommitment was also found to be positively related to performance (Meyre, Paunonen, Gellatly,Go�n and Jackson, 1989). The literature also indicated positive spillover between variablesrepresenting nonwork domains and a�ective commitment (Kirchmeyer, 1992). Thus, all thecorrelates presented here are expected to relate to a�ective commitment positively. However,because the exchange approach is the main theory explaining the development of a�ectivecommitment (Mowday et al., 1982) it is expected that the e�ect of variables that represent workexperience (job satisfaction, job tension), and are therefore an important component of theexchange process, will demonstrate a stronger relationship with a�ective commitment than theother correlates. Continuance commitment, which re¯ects the recognition of costs associatedwith leaving the organization, should be related to anything that increases perceived costs. Director indirect investments in the organization, side bets, represent such costs best, and wereoperationalized mainly by variables like age, education, and tenure (Becker, 1960). Therefore, theabove demographic variables and tenure are expected to demonstrate the strongest relationshipwith continuance commitment. Meyer et al. (1989) found a negative relationship betweencontinuance commitment and performance based on the expectation that people who feel`stacked' in an organization will not exert too much e�ort. Such a relationship is expected here.

Research has shown that the variable continuance organizational commitment has twodimensions: `personal sacri®ces' and `high alternatives' (Hackett, Bycio and Hausdorf, 1994;McGee and Ford, 1987; Somers, 1993). There is little research on antecedents of the twodimensions of continuance commitment. The theoretical rationale behind the two constructssuggests that `personal sacri®ces' will be related to variables that represent side bets, namelyinvestments that might be lost if one leaves the organization. Demographic variables such as ageand tenure are considered good indicators of such side bets (Becker, 1960). For example, olderand veteran employees will hesitate to leave the organization so as not to lose pension plans or

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 287

other accumulated bene®ts. Conceptually, the `low alternatives' dimension represents a constructsimilar to withdrawal cognitions, and thus is expected to relate to situational variables a�ectingstay/leave factors like job satisfaction, perceived performance, or job tension.

Commitment models

Morrow's modelAnother recommendation of Morrow (1983) was for conceptual reevaluation of work commit-ment. Little research has tried to explore the relationships among the forms of work commit-ment. A notable exception is Randall and Cote (1991) who examined the interrelationships ofwork commitment forms, proposing and examining a preliminary theoretical model. Morrow(1993) argued that the nature of interrelationships among forms of work commitment needs to beresolved as soon as possible since evidence that one form of work commitment can moderaterelationships involving other forms is beginning to accumulate (Hunt and Morgan, 1994; Witt,1993). Morrow (1993) suggested that one way to think about the interrelationships among the®ve commitment forms is to locate each within a series of ®ve concentric circles, with work ethicinnermost, followed by career commitment, continuance organizational commitment, a�ectiveorganizational commitment, and on the outermost circle, job involvement. The concentric circlemodel facilitates thinking about the composition of a given employee's work commitments.Morrow argued that no assumptions can be made regarding the size of each circle or even the sizeof the entire model. A high level of one form of work commitment does not imply that lesscommitment must exist elsewhere: one does not have a ®xed amount of loyalty to distributeamong objects.

According to Morrow, commitment forms are positioned in this manner to re¯ect the idea thatthe inner circles are more dispositional, cultural, and cohort-based in nature and thereforerelatively stable over time. The outer circles are thought to be more situationally determined andtherefore more subject to change and in¯uence. Morrow argued that the inner circles impact theouter circles. `Work ethic endorsement, as an illustration, would be expected to exhibit a strongerrelationship with career commitment than a�ective organizational commitment, controllingfor other factors (Morrow, 1993, p. 163). While Morrow mentioned speci®cally that a givencommitment focus will be most related to the next circle outside it, the above quotation allows thepossibility that it will also be related to the circle two stages away, but not as strongly as it isrelated to the adjacent one. Morrow's model is shown in Figure 1a. In Morrow's model, jobinvolvement is the endogenous variable and the two forms of organizational commitment are thevariables which mediate the relationships between the PWE and career commitment, theexogenous variables, and job involvement, the endogenous variable.

Morrow's model concentrates on the positioning of and the conceptual di�erences betweencommitment forms in the inner and outer circles. It does not explain why and how each is relatedto the other. This paper will elaborate on these issues from previous conceptual and empiricalresearch. The logic behind Morrow's model seems similar to that in models proposed by Yoon,Baker and Ko (1994), and Gregerson (1993). Yoon et al. (1994) based their arguments onLawler's (1992) theory of attachment attributed mainly to nested subgroups. This theory orprinciple of proximal rules explicates why `actors develop stronger a�ective ties to subgroupswithin a social system rather than to the social system, to local communities rather than to states,to work organization, and so forth' (Lawler, 1992, p. 334). According to this theory themechanism that leads to stronger commitment to proximal targets is stronger interpersonalattachment among their members. Yoon et al. (1994) de®ned personal attachment as the degree

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

288 A. COHEN

of a�ective personal relationships an individual has with other members in her or his immediatework unit. Interpersonal attachment produces a stronger commitment to subgroups than to thelarger group, because the credit for positive e�ect from interpersonal bonds is likely to beattributed to the proximal subgroups, while the blame for negative a�ects is likely to be attributedto the large group. According to the above logic one can develop stronger attachment to the job,which is a proximal target in one's immediate work unit, than to one's career, which is a muchmore distant target. Gregerson (1993) argued in this regard that proximal variables exert the mostsigni®cant in¯uence on employees' actions because proximity provides more opportunities forexchange relationships.

PWE is a dispositional variable characterized by a belief in the importance of hard work(Mirels and Garrett, 1971) and is assumed to be a relatively ®xed attribute over an individual's

Figure 1. (a) Morrow's model of interrelationships among commitment forms. (b) Randall and Cote'smodel of interrelationships among commitment forms

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 289

life course (Morrow, 1983). As a general work value orientation learned early in the socializationprocesses (Shamir, 1986), it is the most stable commitment form because it is less related tocharacteristics of the work setting. It is also the least proximal variable among the ®ve commit-ment foci because it is not related speci®cally to any concrete characteristic in the workplace.Thus PWE very logically represents the exogenous variable in Morrow's model. According toMorrow, PWE should be related to career commitment. The logic of proximity supports such arelationship. Career commitment represents a more immediate and proximal construct thanPWE. While PWE re¯ects general values about work, career commitment represents a morespeci®c, narrow focus, one's own career or vocation (Blau, 1985). How can PWE be related tocareer commitment? Furnham (1990) applies Holland's (1973) theory according to whichdi�erent types of people have di�erent interests, competencies, and dispositions. They tend tosurround themselves with people and situations congruent with their interests, capabilities, andoutlook. People tend to search for environments that will let them exercise their skills and abilitiesand express their personality. Thus people with high PWE will search for and work in careers that®t their personality and will be more attached to them. The higher their level of PWE the moreattached they will be to the vocation they choose. On the other hand, a low PWE simply does not®t into an occupational setting that values high PWE. The poor ®t should manifest itself,therefore, in low attachment to the occupation.

The literature provides conceptual as well as empirical support for a relationship betweenoccupational commitment and organizational commitment. Witt's (1993) explanation is basedon Schneider's (1983) attraction±selection±attrition (ASA) framework, which states that peopleselect themselves into and out of organizations. In other words, people seek the organizationthat ®ts them. In the ASA framework, individuals highly committed to their occupationmay have carefully selected an organization as an appropriate workplace. Hence they may behighly committed to the organization. Another explanation for this relationship was providedby Vandenberg and Scarpello (1994) who argued that organizational commitment dependsin part on a perceived match or congruence between a person's own values and those espousedby the organization. Given that occupational values and expectations characterize a personalvalue system, the commitment of occupational members to the organization depends in parton realizing their occupational values and expectations within that employment setting.Occupationally committed individuals tend to seek employment (i.e. self-select) in settingsthat encourage them to behave according to the occupational value system. In this case there isa greater chance that individuals' values will ®t those of the organization. A low psycho-logical attachment to an occupation simply does not ®t in an employment setting which valuesthat occupation. The poor ®t should manifest itself, therefore, in low attachment to theorganization.

None of the foregoing distinguishes between continuance and a�ective commitment. Morrow'smodel includes two forms of organizational commitment proposed by Meyer and Allen (1984).The ®rst dimension was termed a�ective commitment and was de®ned as `positive feelings ofidenti®cation with, attachment to, and involvement in, the work organization' (Meyer and Allen,1984, p. 375). The second was termed continuance commitment and was de®ned as `the extent towhich employees feel committed to their organizations by virtue of the costs that they feel areassociated with leaving (e.g. investments or lack of attractive alternatives)' (Meyer and Allen,1984, p. 375). McGee and Ford (1987) in their factor analysis found that the continuancecommitment scale is a two-dimensional construct in itself. One subdimension representsthe personal sacri®ces that would result from leaving the organization and was termed`personal sacri®ces'. The other, `low alternatives', represents the role of available employmentalternatives in the decision to remain in one's organization. Research ®ndings have supported the

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

290 A. COHEN

two-dimensional structure of the continuance commitment construct (Hacket et al. 1994;Somers, 1993).

According to Morrow, career commitment should be related to continuance commitment ®rst.There is conceptual support for presenting continuance commitment before a�ective commit-ment. Witt (1993) states that among some employees, a specialized occupation may lead tolimited alternative opportunities. Because continuance commitment develops partly as a result ofsuch `low alternatives', occupational commitment will be related to continuance commitmentmore strongly than to a�ective commitment. The relationship between continuance and a�ectivecommitment was explained by McGee and Ford (1987) and Meyer, Allen and Gellatly (1990).They stated that the accumulated investments that bind an individual to an organization can lead,through self-justi®cation or dissonance reduction, to an a�ective attachment to the organization.That is, the binding properties of high personal investments are thought to be translated into agreater degree of a�ective attachment to the organization to blunt the disagreeable reality thatone might be stuck as a result of the high costs associated with exit. Partial support for thisrelationship was provided by Meyer et al. (1990) who found that respondents who indicated thatit would be costly to leave the organization reported feeling more a�ective attachment to it.Somers' (1993) ®ndings of a positive relationship between the sacri®ce component of continuancecommitment and a�ective commitment also supported the above relationship. Interestingly,Somers did not ®nd a signi®cant relationship between a�ective commitment and the `lowalternatives' component of continuance commitment.

The outermost circle in Morrow's model is job involvement. Job involvement is viewed byKanungo (1982) as the cognitive or belief state of psychological identi®cation with one's job. Ofall forms of commitment the job is the closest, most tangible and most concrete focus. How doesa�ective commitment a�ect job involvement? Two explanations can be o�ered. First, the ASA(attraction±selection±attrition) framework described earlier argues that because the occupation-ally committed employees may have selected their organization carefully and thus may be moreconcerned with the long term, initial work assignments and the fairness of current work assign-ments may be less salient for them. Therefore, a positive relationship can be expected. Second,Witt (1993) relates three forms of commitmentÐoccupational, organizational, and job involve-mentÐand explains the interrelationships in the same order as in Morrow's model. Accordingly,a specialized occupation may lead to limited opportunities for employment elsewhere, so thatleaving the organization may not be a viable alternative. Hence employees with few alternativeemployment opportunities (high continuance commitment) will develop a�ective attachment tothe organization (high a�ective commitment) and positive attitudes toward any job assignmentthey receive there (high job involvement).

The Randall and Cote modelMorrow's model should be compared with an earlier one proposed and tested by Randall andCote (1991), who tested slightly di�erent forms of commitment. These are PWE, career commit-ment, organizational commitment, work group attachment, and job involvement. They includedonly one form of organizational commitment, the a�ective one, and instead of the continuanceorganizational commitment in Morrow's model theirs had the work group attachment form.Four forms of commitmentÐa�ective organizational commitment, PWE, career commitment,and job involvementÐwere tested in both models. Randall and Cote's approach di�ers fromMorrow's. It postulates that job involvement in¯uences both organizational commitment andcareer salience directly and strongly. It a�ects organizational commitment because situationalfactors have been identi®ed as potentially the most important set of antecedents of organizational

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 291

commitment. Of these, job involvement in particular exerts a powerful in¯uence on commitmentto the organization. Job involvement was also found to predict career salience because it fostersjob challenge, which in turn leads to career identi®cation. Job involvement itself was stronglya�ected by PWE which has a key role in in¯uencing an employee's a�ective responses in theworkplace, and was a�ected to a lesser extent by work group attachment (Randall and Cote,1991). The main di�erence between the Randall and Cote and the Morrow models is the role ofjob involvement, the endogenous variable according to Morrow, and an important mediatoraccording to Randall and Cote. Hence their approach is applied here and tested againstMorrow's model. The model based on their approach is presented in Figure 1b.

While Randall and Cote do not provide su�cient justi®cation for what they termed the pivotalrole of job involvement, this paper will attempt to clarify and justify their argument. Clearly, theproximity argument that in¯uenced Morrow's model is not utilized by Randall and Cote. Insteadthey apply a strong situational approach. Accordingly, experiences in the work setting asrepresented in the level of job involvement will determine a�ective reactions towards otherconstituents in the workplace. As in Morrow, PWE is the exogenous variable. But in Randall andCote's model it a�ects job involvement instead of career commitment, as in Morrow. Schnake(1991) explained that people with a strong work ethic tend to have contempt for idleness and self-indulgence. Employees with a high work ethic are likely to make sure that they put in a fair day'se�ort, and believe that work is its own reward. Persons with a strong work ethic may bemotivated to apply more e�ort, to continue to do so even when bored or fatigued, and to acceptresponsibility for their work. They may feel a moral obligation to perform the task to the best oftheir abilities. Such persons may also feel guilty when they believe they are not working as hard asthey should. Thus they are more likely to be job involved than persons with lower levels of PWE.Shamir (1986) argued that it is therefore reasonable to expect that positive attitudes to work ingeneral will lead to positive attitudes to the speci®c job. People who value the work role highlywill demonstrate it in their attitude toward the job. According to Shamir it would seem thatcommitment to the job, which is based on the individual meaning of the job, has a strongerin¯uence on organizational commitment or occupational commitment. This argument ®ts theRandall and Cote model.

How does job involvement mediate the relationship among PWE, occupational commitmentand the two forms of organizational commitment? Witt (1993) suggested that employees givenunsatisfactory work assignments may develop unfavorable and durable attitudes to the organ-ization, arguing that the reactions to one's work assignment(s) may be salient in the commitmentdecision. His results were consistent with previous ®ndings showing that early work experiencescontribute to later commitment (Pierce and Dunham, 1987). Witt concluded that one's workexperiences and attitudes are an important factor in later job attitudes. The foregoing supportsthe notion of job involvement as a mediator in work commitment interrelationships. Jobinvolvement is strongly a�ected and can be perceived as a re¯ection of work experiences. Themore positive these experiences, the higher the job involvement. Higher job involvement will leadto positive attitudes toward organization and career.

By inference, Randall and Cote seem to conceptualize job involvement as a mediator basedon the social exchange theory that given certain conditions people seek to reciprocate thosewho bene®t them. Employees who are involved in their job have positive work experiences whichare attributed to the organization or their career. To the extent that positive experiences areattributed to the e�orts of organizational o�cials, these are reciprocated with increased a�ectiveorganizational commitment to the persons who caused them. Finally, one highly involvedin a job will also attribute it to the occupation and will reciprocate with high occupationalcommitment.

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

292 A. COHEN

The di�erence between the Morrow and the Randall and Cote models provides an importantreason to compare them. While there are probably other potential models these two are justi®edconceptually and present the best alternatives to be tested and compared. Another issue thisresearch addresses is the continuance commitment form. As mentioned, earlier research clearlyshowed that continuance commitment has two dimensions, `personal sacri®ces' and `lowalternatives'. Morrow (1993) mentioned but did not consider it in developing the work commit-ment model. Thus the need to include the full scale of continuance commitment should be tested.Possibly the relationships among the commitment foci di�er depending on the kind ofcontinuance commitment scale the model includes. Therefore Morrow's and Randall and Cote'smodels will be tested not only with the full scale of continuance commitment but also with eachof its dimensions.

Method

Participants

Nurses from two hospitals in Western Canada, one of medium size and one small, were surveyed.Questionnaires were distributed to all 300 nurses in the medium-size hospital and all 200 nurses inthe small one, a total of 500 questionnaires. In the ®rst hospital 157 nurses completed usablequestionnaires (52 per cent response rate) and in the second, 81 (40 per cent response rate). Atotal of 238 usable questionnaires was returned, a response rate of 47 per cent. Of the nurses95 per cent were females. The mean age of the respondents was 38.6 years and the mean tenure inthe organization and occupation was 10.4 and 15.1 years respectively. Of the nurses 55 per centhad college education and 15.1 per cent had completed university studies. Among the nurses28.3 per cent of them were registered nurse assistants, 60.4 per cent registered nurses, 7 per centnursing unit managers and 4.3 per cent in-service educators or administrators.

Measures

Commitment fociTwo measures of organizational commitment were collected: the two 8-item scales of a�ectiveand continuance commitment of Meyer and Allen (1984). The two dimensions of the con-tinuance commitment scale, `high sacri®ces' and `low alternatives', were also analyzed. Careercommitment was measured by the 8-item measure developed by Blau (1985). Job involvement(10 items) was measured by the measure developed by Kanungo (1982). PWE was measured by10 items from the 19-item scale developed byMirels and Garrett (1971). Items negatively phrasedor seeming to overlap, mainly those touching on work commitment and job involvement, wereexcluded. The idea was to concentrate on items which, as stated by Shamir (1986), measure theimportance of work in a more abstract and remote way than does the job involvement scale, itemscontaining reference to society and to moral judgement. Items that focus on potentialsatisfactions with the work role similar to job involvement (e.g. `I feel uneasy when there is littlework for me to do') were omitted. All work commitment constructs were measured on a 7-pointscale (1� strongly disagree, to 7� strongly agree). The scales applied in this research werementioned by Morrow (1993) as the most commonly used and the most reliable and valid workcommitment scales. Each was also mentioned as having a strong discriminant validity in its

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 293

relationship to other work commitment forms. Thus the selection of the scales ®ts well withMorrow's conceptualization.

Commitment correlatesAge, tenure and years in the occupation were measured as ratio scales, gender and children asdichotomous variables, and education on a scale from 1 to 7 (1� elementary school, 7� higheruniversity degree). Perceived performance was measured by a 3-item measure developed by Halland Hall (1976). The scale ranged from 1 (never) to 7 (almost all of the time). The variable job-induced tension was measured by seven items based on the measure developed by House andRizzo (1972), one of several scales from the anxiety±stress questionnaire developed by theauthors to measure tensions and pressures growing out of job requirements, including possibleoutcomes in terms of feelings or physical symptoms. The scale ranged from 1 (strongly disagree)to 7 (strongly agree). Job satisfaction was measured by a 5-item measure based on one developedby Schriesheim and Tsui (1980). The respondents were asked to indicate on a 7-point scale howsatis®ed they were with the following aspects of their work: pay, promotion, supervision, co-workers, and job. The variable life satisfaction was measured by ®ve items based onVrendenburgh and Sheridan (1979). The respondents were asked to indicate on a scale from 1(very dissatis®ed) to 7 (very satis®ed) how satis®ed they were with ®ve aspects of their life:residence, nonwork activities, family life, friendship, and health.

Data analysis

Con®rmatory factor analysisThis analysis was performed following the procedure outlined by Brooke et al. (1988) andMathieu and Farr (1991). Three indicators were established for each multi-item measure by ®rst®tting a single-factor solution to each set of items and then averaging the items with highest andlowest loadings to form the ®rst indicator, averaging the items with the next highest and lowestloadings to form the second indicator and so on until all items were assigned to one of the threeindicators for each variable. This was necessary to reduce the number of parameters and toreduce the scale items to three parallel indicators of each construct, as in developing parallel testforms (Nunnally, 1978). The extent to which the three indicators tapped the more generalunderlying constructs was then assessed by ®tting con®rmatory factor analysis models. Acovariance matrix of these indicators using listwise deletion of missing values formed the inputfor the LISREL analysis.

LISREL VIII (Joreskog and Sorbom, 1993) analysis was performed to test the discriminantvalidity involving a comparison of the relative ®t of 6-, 5-, 4-, 3-, and 1-factor measurementmodels. The 5-factor model placed the three indicators of continuance organizational commit-ment, a�ective organizational commitment, career commitment, job involvement, and PWE onseparate latent factors. The 10 4-factor models were established by forcing six indicators of twoconstructs onto a single factor and placing the nine indicators of the remaining three constructson three factors. The 10 3-factor models were established by forcing nine indicators of threeconstructs onto a single factor and placing the six indicators of the remaining two constructs ontwo factors. The 1-factor model forced all 27 indicators onto a single latent factor. Because ofthe bi-dimensionality of the continuance commitment form, three additional analyses wereperformed. First, a 6-factor model was tested with the two dimensions of continuance commit-ment placed on separate latent factors. Second, two 5-factor solutions were tested, where each of

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

294 A. COHEN

the dimensions of continuance commitment represented continuance commitment while theother dimension was omitted. In these two analyses continuance commitment and its dimensionswere not analyzed by using the 3-subscale indicator, but instead each of the dimensions wasanalyzed at the item level.

Correlation analysisConvergence or divergence of the various work commitment forms was further examined usingsimple bivariate correlations. This procedure examined the correlations between 10 demographicand work-related variables and the ®ve commitment forms. This analysis tested the hypothesisthat the correlations were the same between the demographic and job-related variables and the®ve commitment forms.

Path analysisThe two models regarding the interrelationships among the ®ve commitment forms as presentedin Figures 1a and b were assessed by path analysis using LISREL VIII (Joreskog and Sorbom,1993). The models were evaluated by using the two-stage approach to structural equationmodeling suggested by Anderson and Gerbing (1988). This approach entails comparing thegoodness-of-®t indices of a sequence of nested models. Because the models compared are nestedmodels, a chi-square di�erence test (Bollen, 1989) was applied to compare the models. A signi®-cant chi-square will indicate that the constraints imposed on the restricted models reduce their ®tin comparison with the saturated or the hypothesized models. It should be noted that becauseMorrow's and Randall and Cote's models are not nested, because they are not identical models(Bollen, 1989), they are not compared with each other but each one of them is compared with themeasurement/saturated model. The LISREL program calculates a `modi®cation index' for every®xed parameter in a model. The modi®cation index re¯ects the minimum reduction in thechi-square statistics if the parameter is changed from ®xed to free. The two models tested herewere revised on the basis of these modi®cation indexes. The revised models are compared with themeasurement model as well as with their original model. Breckler (1990) argued that the modi®edmodel should be assessed by using di�erent data. Otherwise, they should be cautiouslyinterpreted.

Model evaluationThe ®t of the models was assessed by means of eight indices. The chi-square test is the most basicand is essential for the nested model comparison. Because the chi-square test is sensitive tosample size, the ratio of the model chi-square to degrees of freedom was used as another ®t index.The following ®t indices are reported as less sensitive to sample size di�erences and to the numberof indicators per latent variable increase (Medsker, Williams and Holahan, 1994): the relative ®tindex (RFI), the comparative ®t index (CFI), RNI, NFI, the non-normed ®t index (NNFI orTLI), and the RMSEA. The ®t of a given model to the data is an important criterion of thequality of the model but it does not necessarily imply that this model is the correct causal model(Saris and Stronkhorst, 1984). The path coe�cients, their signi®cance, and their magnitudeprovide an important criterion for model evaluation, termed the plausibility criterion. Theplausibility of a model refers to a judgement made about the theoretical argument underlying thespeci®ed model (Saris and Stronkhorst, 1984). In other words there has to be some balancebetween the ®t indices and the theoretical predictions regarding the relationships among research

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 295

variables. A correlation matrix among the 15 indicators of the ®ve commitment forms, threeindicators for each form, using listwise deletion of missing values, formed the input for the pathanalysis.

Results

Correlations

Table 1 shows the descriptive statistics, reliabilities and the intercorrelations among researchvariables. Results show acceptable reliabilities of the measures of this study with somewhat lowreliability (a� 0.69) of the continuance organizational commitment scale. Following previousresearch (Hackett et al., 1994; McGee and Ford, 1987; Somers, 1993) the continuance commit-ment scale was divided into two: `high sacri®ces' and `low alternatives'. However, the reliabilitiesof the 4-item scales were very low (0.54 and 0.60 respectively). Reliability analysis shows thatomitting two items, one from each scale, will improve reliability. Thus for each of the twosubscales a 3-item scale was formed with a reliability of 0.60 for `high sacri®ces' and 0.65 for `lowalternatives'. These reliabilities are still low, but higher than those found by Somers (1993), alsoin a sample of nurses (0.57 and 0.59 respectively). Three-item scales for the two dimensions ofcontinuance commitment were applied in most research (Dunham, Grube and Castaneda, 1994;Hackett et al., 1994; McGee and Ford, 1987; Somers, 1993). The correlations of the two subscaleswith work commitment constructs support their being distinguished. For example, `highsacri®ces' has positive signi®cant relationships with a�ective organizational commitment(r� 0.193, p5 0.01), career commitment (r� 0.130, p5 0.05) and job involvement (r� 0.159,p5 0.01), while `high sacri®ces' has negative relationships with a�ective organizational commit-ment (r�ÿ0.201, p5 0.001) and with career commitment (r�ÿ0.272, p5 0.001). The di�er-ential relationships of the two subscales of continuance commitment with work commitmentconstructs explains the nonsigni®cant relationships of the full scale with the same constructs.

The correlations among the work commitment scales show that the a�ective commitmentdimension has the highest correlation with other commitments such as job involvement(r� 0.527) and career commitment (r� 0.483). The PWE has a nonsigni®cant relationship withthe two forms of organizational commitment and signi®cant but weak correlations with career

Table 1. Descriptive statistics, reliabilities and intercorrelations among research variables (reliabilities inparentheses)

Mean S.D. 1 2 3 4 5 6 7

1. Continuance organizational 40.08 7.41 (0.69)commitment

2. Personal sacri®ces 15.08 3.79 0.822 (0.60)3. Low alternatives 15.51 3.46 0.750 0.370* (0.65)4. A�ective organizational 32.69 7.87 0.047 0.193{ ÿ0.201* (0.79)

commitment5. Career commitment 36.99 9.31 ÿ0.109 0.130{ ÿ0.272* 0.483* (0.87)6. Job involvement 32.84 8.25 0.050 0.159{ ÿ0.079 0.527* 0.568* (0.79)7. Protestant Work Ethic 35.05 8.79 0.062 0.003 0.084 0.070 0.112{ 0.272* (0.76)

* P5 0.001; {P5 0.01; {P5 0.05.Note: N� 222.

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

296 A. COHEN

Table 2. Con®rmatory factor analysis results

Model/description df w2 w2/df RNI RFI NFI TLI CFI RMSEA

1. Null model (with continuance commitment analyzed 105 1252.90*

as a full scale with the three-indicators approach)2. One factor 90 579.91* 6.44 0.57 0.46 0.54 0.50 0.57 0.163. Five factors 80 141.00* 1.76 0.95 0.85 0.89 0.93 0.95 0.0594. Null model (with continuance commitment analyzed 153 1521.53*

at the item level)5. Five factors 125 261.71* 2.09 0.90 0.79 0.83 0.88 0.90 0.0706. Six factors with continuance commitment as a 120 181.74* 1.51 0.95 0.85 0.88 0.94 0.95 0.048

two-dimensional construct7. Five factors with `personal sacri®ces' as representing 80 138.16* 1.73 0.95 0.86 0.89 0.93 0.95 0.057

continuance commitment8. Five factors with `low alternatives' as representing 80 125.42* 1.56 0.96 0.90 0.90 0.95 0.96 0.051

continuance commitment

P5 0.001.

Copyrig

ht#

1999JohnWiley

&Sons,Ltd.

J.Organiz.

Behav.

20,285±308(1999)

RELATIO

NSHIPSAMONG

FIV

EFORMSOFCOMMIT

MENT

297

commitment (r� 0.112) and job involvement (r� 0.272). The correlations indicate a patternwhere strong intercorrelations exist among a�ective organizational commitment, job involve-ment, and career commitment, and weak relationships between PWE and continuance organ-izational commitment and to the other three forms.

Con®rmatory factor analysis

Table 2 shows the results of con®rmatory factor analysis. Several alternative models were tested:a 5-factor model which represents Morrow's proposed model, a 1-factor model, 10 4-factormodels, 10 3-factor models and three models based on dividing continuance commitment intotwo dimensions. Table 2 does not present the 10 4-factor models and the 10 3-factor models.However, the ®ndings which are available clearly show that the 5-factor model ®ts the data betterthan the 1-factor model, any 4-factor models, and any 3-factor models. The w2/df ratio is 1.76, theTLI is 0.93, and CFI is 0.95. These ®t indexes are much better than those in the 1-factor model orany of the 10 4-factor models, and show an acceptable discriminant validity of the ®ve forms ofwork commitment. Also, a chi-square di�erence test which compares the 5-factor model with the10 4-factor models and the 10 3-factor models revealed signi®cant di�erences between the5-factor models and any of the other 20. This ®nding shows a better ®t with the data of the5-factor model than with any other.

An earlier question asked whether a 6-factor model with a two-dimensional continuancecommitment would provide a better ®t. The lower portion of Table 2 shows the results wherecontinuance commitment and its dimensions were analyzed at the item level. The ®ndings of thechi-square test showed a signi®cant di�erence between the 5- and the 6-factor models (w2� 79.97;df� 5; p5 0.001) which indicates a better ®t of the 6-factor (model 6) than the 5-factor model(model 5). When each of the two dimensions of continuance commitment was analyzedseparately as representing this construct (models 7 and 8), results of the chi-square test showed asigni®cance di�erence between them and the 5-factor model (model 5). This ®nding supports abetter ®t of models that use any of the dimensions of continuance commitment than models thatcombine them into one scale. The chi-square test revealed no signi®cant di�erence between the6-factor model (model 6) and the models with the dimensions of continuance commitment(models 7 and 8). This showed that the ®t of the two models with the data is not better than the6-factor model. Finally, the ®t indices showed that the 5-factor model with `low alternatives' asrepresenting continuance commitment has a somewhat better ®t with the data than the modelwith `personal sacri®ces'.

Correlation analysis

Results of the correlation analysis are presented in Table 3. A review of the correlations providessome evidence of di�erences in relations for the ®ve commitment constructs. Contrary toexpectations, data showed that PWE was not related to any of the demographic variables. Asexpected, PWE was not related to work experience variables or to work outcomes, with oneexceptionÐthe signi®cant relationship with job tension (r� 0.147; p5 0.05). Nor did the datasupport the expectation for a positive relationship between PWE and life satisfaction. Jobinvolvement was expected to be related to demographic and work experience variables. Itssigni®cant relationships with gender, tenure, and job satisfaction provide some support here. Asexpected, job involvement was not related to perceived performance or to life satisfaction. Career

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

298 A. COHEN

Table 3. Relations (Pearson's r) between correlates and work commitment forms

Correlate A�ective Continuance Low Personal Job Career Protestantorganizational organizational alternatives sacri®ces involvement commitment work ethiccommitment commitment

Gender* (mean� 0.03; S.D.� 0.16) 0.110 0.017a ÿ0.016a 0.068 0.195b 0.090 0.033Children{ (mean� 0.71; S.D.� 0.45) 0.258a 0.178 0.074b 0.208 0.115b 0.144 0.064bEducation (mean� 3.43; S.D.� 1.50) ÿ0.085 ÿ0.217a ÿ0.132b ÿ0.195c 0.017b 0.022b ÿ0.108Age (mean� 38.60; S.D.� 9.06) 0.223a 0.054 0.086 0.006 0.106c 0.034b ÿ0.070bdTenure (mean� 10.41; S.D.� 7.35) 0.243a 0.181c 0.188eg 0.122 0.140e ÿ0.023df ÿ0.011bdYears in occupation 0.158a 0.061 0.151c ÿ0.019df 0.057 ÿ0.099b ÿ0.100b(mean� 15.05; S.D.� 9.14)

Perceived performance 0.143a ÿ0.098bc ÿ0.063ih ÿ0.072hi 0.105bd 0.159de ÿ0.044bf(mean� 15.73; S.D.� 2.73; l� 0.91)

Job satisfaction 0.392a ÿ0.089bc ÿ0.276ag 0.066hi 0.160bd 0.308de 0.005bf(mean� 25.06; S.D.� 5.17; l� 0.70)

Life satisfaction 0.068 ÿ0.071a ÿ0.134g ÿ0.019 ÿ0.066c 0.131bde ÿ0.087f(mean� 29.22; S.D.� 4.29; l� 0.70)

Job tension ÿ0.077a 0.176bc 0.254j 0.069ik 0.109be ÿ0.144adfg 0.147bh(mean� 28.14; S.D.� 8.51; l� 0.84)

N ranges from 195 to 235 due to missing values. Interpretation of the subscripts based on pair comparison is only as follows: a di�ers signi®cantly from b; c di�erssigni®cantly from d; e di�ers signi®cantly from f; g di�ers signi®cantly from h; i di�ers from j; k di�ers from l. Correlations that do not share a common subscript withineach row di�er signi®cantly (p5 0.05) from one another. Correlations that do not have any subscript within each row do not di�er signi®cantly from all the othercorrelations (p5 0.05). Correlations 40.13 di�er signi®cantly from zero.* 0� female, 1�male; { 0� no children, 1� having children.

Copyrig

ht#

1999JohnWiley

&Sons,Ltd.

J.Organiz.

Behav.

20,285±308(1999)

RELATIO

NSHIPSAMONG

FIV

EFORMSOFCOMMIT

MENT

299

commitment was expected to be related to individual di�erences and situational characteristics.Its relationships with children, job satisfaction, and job tension supported this expectation.However, stronger relationships with other demographic variables were expected. Its relationshipwith perceived performance (r� 0.159; p5 0.05) and life satisfaction (r� 0.131; p5 0.05) werenot expected. A�ective organizational commitment was related to most variables tested here, asexpected. The expectation of a strong relationship with situational variables was supported by therelatively strong correlation with job satisfaction (r� 0.392, p5 0.05). Its nonsigni®cantrelationships with life satisfaction and job tension were not expected.

Continuance commitment was expected to be related to demographic variables, and signi®cantrelationships with children, education, and tenure supported this. The nonsigni®cant relationshipof this variable with the situational variables was expected except for the one signi®cantrelationship with job tension (r� 0.176; p5 0.05). When the two dimensions of continuanceorganization commitment were examined separately, ®ndings revealed signi®cant di�erencesbetween the two in their relationships with many of the antecedents examined here. As expected,the `personal sacri®ces' dimension was related to demographic variables such as children andeducation. Its nonsigni®cant relationships with the situational variables were also expected.However, higher correlations were expected between this variable and age and tenure. The `lowalternatives' dimension had strong relationships with situational variables such as job tension(r� 0.276; p5 0.05) and job satisfaction (r� 0.276; p5 0.05) as expected. Its signi®cantcorrelations with the variables tenure and years in the occupation, however, were not expected.The di�erences between the two dimensions in their relationships with some of the correlates arequite strong. In fact, the data show signi®cant di�erences among the commitment forms in manyof the correlations as seen in Table 3.

Path analysis

Table 4 shows the ®t indices for the models tested. Each of the models was tested four times: withthe full continuance commitment scale using the three-indicator approach, with the same fullscale using the item approach, with the `personal sacri®ces' dimension, and with the `lowalternatives' dimension. Table 4 shows that in the four analyses Morrow's model did not ®t thedata well. This was demonstrated in the ®t indices, which are lower in comparison with themeasurement model and with all the other models, and mainly in the results of the chi-squaretest. As the table shows, this test consistently yielded a high and signi®cant chi-square value in thecomparison betweenMorrow's model and the saturated/measurement model, demonstrating thatthe constraints posed by the paths representing Morrow's model worsen the ®t with the data. Inaddition, Table 5, which presents the structural coe�cients for the model, shows that only threepaths in the models are signi®cant: from PWE to career commitment, from career commitment toa�ective organizational commitment, and from a�ective organizational commitment to jobinvolvement. Applying each of the two dimensions of continuance commitment instead of thefull scale yielded one more signi®cant path, PWE to continuance organizational commitment.

The results in Table 4 show that Randall and Cote's (1991) model ®ts the data better thanMorrow's. This is demonstrated in a lower chi-square di�erence in this model than in thesaturated one in all analyses except for that which used `low alternatives' as representingcontinuance commitment. Also, the ®t indexes of the Randall and Cote model are better thanthose of the Morrow model. Three of the four path coe�cients in Randall and Cote's model aresigni®cant. The only nonsigni®cant path is the one from job involvement to continuanceorganizational commitment. This path becomes signi®cant when the `low alternatives' dimension

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

300 A. COHEN

Table 4. Overall ®t indexes for the work commitment models

Model/description df w2 w2/df Model nw2 RNI RFI NFI TLI CFI RMSEAcomparison

With full scale of continuance commitment1. Measurement model 80 141.00* 1.76 0.95 0.85 0.89 0.93 0.95 0.0592. Morrow's model 83 170.84* 2.06 1 vs. 2 29.84* 0.92 0.83 0.86 0.90 0.92 0.0693. Morrow's revised model 85 160.25* 1.89 1 vs. 3 19.25{ 0.93 0.84 0.87 0.92 0.93 0.0634. Randall and Cote's model 86 162.86* 1.89 1 vs. 4 21.25{ 0.93 0.84 0.87 0.92 0.93 0.0645. Randall and Cote's revised model 84 148.44* 1.77 1 vs. 5 7.44 0.94 0.85 0.88 0.93 0.94 0.059

With full scale of continuance commitment (item level)1. Measurement model 125 261.71* 2.09 0.90 0.79 0.83 0.88 0.90 0.0702. Morrow's model 128 291.57* 2.28 1 vs. 2 29.85* 0.88 0.77 0.81 0.86 0.88 0.0763. Morrow's revised model 130 273.86* 2.11 1 vs. 3 12.15{ 0.89 0.79 0.82 0.88 0.89 0.0714. Randall and Cote's model 131 276.47* 2.11 1 vs. 4 14.76{ 0.89 0.79 0.82 0.88 0.89 0.0715. Randall and Cote's revised model 129 269.39* 2.09 1 vs. 5 7.68 0.90 0.79 0.82 0.88 0.90 0.070

With `personal sacri®ces' as representing continuance commitment1. Measurement model 80 138.16* 1.73 0.95 0.86 0.89 0.93 0.95 0.0572. Morrow's model 83 168.59* 2.03 1 vs. 2 30.43* 0.93 0.83 0.87 0.91 0.93 0.0683. Morrow's revised model 85 151.57* 1.78 1 vs. 3 13.41{ 0.94 0.85 0.88 0.93 0.94 0.0604. Randall and Cote's model 86 154.11* 1.79 1 vs. 4 15.95{ 0.94 0.85 0.88 0.93 0.94 0.0605. Randall and Cote's revised model 84 148.52* 1.77 1 vs. 5 10.36{ 0.95 0.85 0.88 0.93 0.95 0.059

With `low alternatives' as representing continuance commitment1. Measurement model 80 125.42* 1.56 0.96 0.90 0.90 0.95 0.96 0.0512. Morrow's model 83 156.04* 1.88 1 vs. 2 30.62* 0.94 0.84 0.88 0.92 0.94 0.0633. Morrow's revised model 85 153.13* 1.80 1 vs. 3 27.71* 0.94 0.85 0.88 0.93 0.94 0.0604. Randall and Cote's model 86 155.65* 1.81 1 vs. 4 30.23* 0.94 0.85 0.88 0.93 0.94 0.0615. Randall and Cote's revised model 84 137.22* 1.63 1 vs. 5 11.8{ 0.95 0.86 0.89 0.94 0.95 0.054

p5 0.001; { p5 0.01; { p5 0.05.

Copyrig

ht#

1999JohnWiley

&Sons,Ltd.

J.Organiz.

Behav.

20,285±308(1999)

RELATIO

NSHIPSAMONG

FIV

EFORMSOFCOMMIT

MENT

301

Table 5. Structural coe�cients for research models

Morrow's Morrow's Morrow's Randall Randall and Randall andmodel* model with model with and Cote's Cote's model Cote's model

`personal `low model* with `personal with `lowsacri®ces' alternatives' sacri®ces' alternatives'

Parameters for proposed modelsPath coe�cientsProtestant work ethic! Career commitment 0.09 (0.09) 0.09 0.09Protestant work ethic! Job involvement 0.28{ (0.28{) 0.28{ 0.28{Protestant work ethic! Continuance 0.15 (0.09) 0.00 0.21{

organizational commitmentCareer commitment! Continuance ÿ0.23{ (ÿ0.09) 0.10 ÿ0.36{

organizational commitmentCareer commitment! A�ective 0.62{ (0.60*) 0.58{ 0.58{

organizational commitmentContinuance organizational commitment 0.14 (0.10) 0.17{ ÿ0.03! A�ective organizational commitment

Continuance organizational commitment 0.00 (ÿ0.04) ÿ0.07 0.04! Job involvement

A�ective organizational commitment 0.73* (0.73*) 0.74{ 0.74{! Job involvement

Job involvement! Continuance 0.00 (0.01) 0.13 ÿ0.19{organizational commitment

Job involvement! A�ective 0.71{ (0.71{) 0.71{ 0.71{organizational commitment

Job involvement! Career commitment 0.61{ (0.61{) 0.62{ 0.62{

Parameters for revised modelsPath coe�cientsProtestant work ethic! Job involvement 0.26{ (0.26{) 0.26{ 0.25{ 0.30{ (0.30{) 0.30{ 0.30{Career commitment! Job involvement 0.60{ (0.60{) 0.60{ 0.61{Career commitment! Continuance ÿ0.35{ (ÿ0.13) 0.05 ÿ0.41{

organizational commitmentCareer commitment! A�ective 0.23{ (0.23{) 0.23{ 0.24{

organizational commitmentJob involvement! Career commitment 0.58{ (0.58{) 0.58{ 0.58{Job involvement! Continuance ÿ0.01 (0.01) 0.13 ÿ0.19{ 0.21 (0.09) 0.09 0.10

organizational commitmentJob involvement! A�ective organizational 0.70{ (0.70){ 0.71{ 0.71{ 0.54{ (0.53{) 0.53{ 0.52{

commitment

* The numbers in parentheses are the path coe�cients based on analyzing continuance commitment at the item level.{ p5 0.05.

Copyrig

ht#

1999JohnWiley

&Sons,Ltd.

J.Organiz.

Behav.

20,285±308(1999)

302

A.COHEN

represents continuance commitment. A noteworthy ®nding is that the structural coe�cientsin Table 5 show that each dimension of continuance commitment relates di�erently to therelevant commitment foci. For example, career commitment has a signi®cant negative relation-ship with the full scale of continuance commitment (ÿ0.23), a positive and nonsigni®cantrelationship with the `personal sacri®ces' dimension (0.10), and a high negative relationship withthe `low alternative' dimension (ÿ0.36). The same pattern can be found in all the paths thatinvolve continuance commitment.

However, while the ®ndings show a better ®t of Randall and Cote's model thanMorrow's, theyindicate that neither model has a very good ®t with the data. Therefore, the two models wererevised based on the modi®cation indices. In Morrow's model the modi®cation indices showedthat the model would be signi®cantly improved if the following paths were included: from jobinvolvement to a�ective organizational commitment; from job involvement to career commit-ment; from career commitment to job involvement; and from PWE to job involvement. Themodi®cation indices in Randall and Cote's model showed a reduction in the size of chi-square iftwo paths were included: a path from career commitment to a�ective organizational commitmentand a path from career commitment to continuance organizational commitment. Modi®cationindices also suggested paths from a�ective and continuance commitment to career commitment,which were not tested because paths are not allowed to go backward. Based on these indices themodels were revised as presented in Figure 2. Table 4 shows that the revised models ®t the datasigni®cantly better than the original ones. Morrow's revised model shows in general a better ®twith the data than the original one. However, in three out of the three analyses, the chi-squarevalue that compared this model with the saturated one is still signi®cant yet much lower than inthe original model. It should be noted that when `low alternatives' was used as representingcontinuance commitment Morrow's revised model revealed almost no improvement in its ®t incomparison with the original model. A good indication of the improvement in Morrow's revisedmodel is provided by the path coe�cients presented in Table 5. Except for the job involvement tocontinuance commitment path, all the other paths are signi®cant in Morrow's revised model.Moreover, even this problematic path is signi®cant when `low alternatives' represents continu-ance commitment in Morrow's revised model.

The revised Randall and Cote's model ®ts the data very well, as demonstrated in a non-signi®cant chi-square when compared with the measurement/saturated model. This was found inboth. All the paths in the revised model are signi®cant and support the notion of the two modelswhere the full scale of continuance commitment was analyzed. Even in the two other analyses,where each of the continuance commitment dimensions was analyzed separately, the chi-squarewas very low, although signi®cant, at the 0.05 level. The ®t indices of the revised Randall andCote's model were similar to those of the measurement model and better than Morrow's originaland revised models. Morrow's revised model was much di�erent from the original one andtherefore was not compared with it as a nested model. However, the revised Randall and Cote'smodel was the same as the original but with the addition of two paths. As a nested unconstrainedmodel it was compared with the original nonrevised model. The signi®cant chi-square in threeout of the four analyses (except for the model using `personal sacri®ces' as representing con-tinuance commitment) showed that the constraints imposed on the nonrevised model worsen its®t and therefore support the usefulness of the revised Randall and Cote model. Most of the pathsin the revised model are signi®cant except for two paths that include continuance commitment:the one from job involvement to continuance commitment, and in two cases the one from careercommitment to continuance commitment. It should be noted that in the revised models, as in thenonrevised ones, the structural coe�cients are related in general di�erently to each dimension ofcontinuance commitment, as seen in Table 5.

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 303

Discussion

The goal of this paper was to test Morrow's model of ®ve components of work commitment. The®ndings support Morrow's ®rst contention that each of the ®ve commitment forms is anindependent construct. However, because the ®t indices are not perfect, more work is needed toimprove the discriminant validity among the scales. One way to improve it is to decide whichof the dimensions of continuance commitment should be included as part of the work commit-ment foci. Nevertheless, no empirical support was found for Morrow's model regarding the

Figure 2. (a) Morrow's revised model. (b) Randall and Cote's revised model

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

304 A. COHEN

interrelationships among the commitment foci. The main reason for rejecting Morrow's model isthe role of job involvement. Morrow suggests that job involvement is more situation-a�ectedthan any other commitment form and is therefore the endogenous variable. The data stronglyrule out this argument by their poor ®t with Morrow's model regardless of the continuancecommitment dimensions applied, in the modi®cation indices all showing that the model wouldimprove its ®t if job involvement were not an endogenous variable, and in the substantialimprovement e�ected in the revised model when job involvement was placed as a mediator.Possibly job involvement is not a situational dependent variable, as Morrow argued. Blau andBoal (1989) argued from other research that job involvement is a more stable work attitude thanorganizational commitment in the sense that job involvement may be more di�cult to change.They argued that several behavioral scientists (e.g. Lodhal, 1964; Siegal, 1969) note thatindividual di�erences in job involvement can be traced back to orientations toward work early inthe person's socialization process such as early school experiences.

The ®ndings show that Randall and Cote's model ®ts the data much better than Morrow'smodel. The best ®t with the data was found in the revised model of Randall and Cote. The®ndings support Randall and Cote's argument that job involvement seems to be a key mediatingvariable in the interrelationships among work commitment constructs. It seems therefore that theproximity explanation regarding the commitment foci interrelationships does not have empiricalsupport. Instead, the notion of exchange was supported as the main rationale for the mediatingrole of job involvement. That is, job involvement is related to commitment foci that representcultural and socialization e�ects such as PWE. However, its relationship with the other commit-ment foci, in particular the two organizational commitment forms, is a�ected by the kind andtype of exchange relationship developed in the work setting. Employees who are highly involvedin their job have more positive work experiences, attributed to organizational o�cials or theircareer decision, and will reciprocate with high commitment to these foci.

The ®ndings also show that the models operate di�erently in terms of their ®t and thestructural coe�cients, depending on whether the full continuance commitment scale or each ofits dimensions is used. The question of which of the two dimensions represents continuancecommitment better is important because a decision is required regarding which dimension ofcontinuance organizational commitment should be included as a universal form. The results heredo not facilitate such a decision. The discriminant analysis ®ndings showed a better ®t with the`low sacri®ces' dimension. The structural models showed that the Randall and Cote model hasthe best ®t with the `personal sacri®ces' dimension of continuance commitment. The revisedmodels showed that the revised Morrow model ®ts best with the `personal sacri®ces' dimension,while the revised Randall and Cote model ®ts best with the full scale of continuance commitment.Clearly then, future research must resolve this issue. But conceptually, `personal sacri®ces' ®tsbetter into the universal forms of work commitment because the `low alternatives' dimensionseems to represent cognitions of withdrawal from the workplace, which is more an outcomevariable than a commitment focus. The correlations that show that `low alternatives' is related tovariables that are turnover correlates such as tenure, years in the occupation, and job satisfaction(Cotton and Tuttle, 1986) partially support this contention. The `personal sacri®ces' dimensionseems to capture Becker's (1960) idea of potential costs more than the `low alternatives' one asre¯ected in `It would not be too costly for me to leave my organization in the near future'. Theinclusion of the `personal sacri®ces' dimension and not `low alternatives' can lessen the di�cultythat all the other four commitment foci represent a�ective or cognitive components of attitudes,while the continuance commitment dimension has a behavioral intent or cognitive emphasis.Future research should examine other forms of commitment from a multidimensional perspect-ive because there is evidence for multidimensionality of other forms of work commitment such as

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 305

PWE (Furnham, 1990), occupational commitment (Meyer et al., 1993) and even job involvement(Paullay, Alliger and Stone-Romero, 1994).

The ®ndings therefore seem to substantiate the notion that in work commitment modelsorganizational commitment is an endogenous variable and PWE is an exogenous one, while jobinvolvement mediates this relationship. On the assumption that these conclusions will bereplicated in future research, the role of career commitment is an issue that needs to be resolved.The ®ndings suggest two options. First, career commitment is an endogenous variable in theinterrelationships together with the two forms of organizational commitment, as suggested byRandall and Cote (1991) and presented in model 1a. Second, career commitment moderates therelationship between job involvement and organizational commitment, as the revised model(Figure 2b) shows. Future research and theoretical work should resolve this issue. Such researchis also needed to validate the revised models, which should be tested with new data before any®rm conclusions about them can be made. Several other steps are desirable in future research toestablish the usefulness of these ®ve commitment forms. First, the ®ndings here should bereplicated in other samples and work settings. Consistent results across di�erent settings areimportant in this regard to support the discriminant validity of these forms and the nature oftheir interrelationships. Second, this research applied measures extensively used in commitmentresearch. But measures di�erent from the constructs applied here should also be tested.

Several limitations of this research should be mentioned. First, because the study is based onself-report data it carries the possibility of source bias or general method variance. Second,because of the use of cross-sectional, correlational data, the data must be interpreted cautiouslyin terms of the causal inferences concerning the various hypothesized relationships. But becausethere are hardly any data on the interrelationships among forms of commitment, these presentedhere may be found interesting and valuable. Research regarding the interrelationships amongcommitment foci is in its exploratory stage and the data here, together with the few previousstudies and more future ®ndings, could assist in developing appropriate hypotheses in long-itudinal designs regarding these interrelationships. Finally, there is a need to test the predictivepower of these forms. The main justi®cation for work commitment research is the assumptionthat outcomes such as turnover, organizational citizenship behavior, and absenteeism are betterexplained as a function of multiple commitments than of one (Mueller et al., 1992). For example,if we accept Randall and Cote's model then the two organizational commitment foci as thedependent variables should have a stronger relationship with work outcomes than the othercommitment foci. An important contribution of this study is that the interrelations among thecommitment forms explored here should provide the basis for hypotheses regarding possiblee�ects of di�erent commitment forms on work outcomes.

References

Anderson, J. C. and Gerbing, D. W. (1988). `Structural equation modeling in practice: A review andrecommended two-step approach', Psychological Bulletin, 103, 411±423.

Becker, H. S. (1960). `Notes on the concept of commitment', American Journal of Sociology, 66, 32±40.Becker, T. E. (1992). `Foci and bases of commitment: Are they distinctions worth making?' Academy ofManagement Journal, 35, 232±244.

Becker, T. E. and Billings, R. S. (1993). `Pro®les of commitment: An empirical test', Journal ofOrganizational Behavior, 14, 177±190.

Blau, G. J. (1985). `The measurement and prediction of career commitment', Journal of OccupationalPsychology, 58, 277±288.

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

306 A. COHEN

Blau, G. J. (1986). `Job involvement and organizational commitment as interactive predictors of tardinessand absenteeism', Journal of Management, 12, 577±584.

Blau, G. J. and Boal, K. B. (1989). `Using job involvement and organizational commitment interactively topredict turnover', Journal of Management, 15, 115±127.

Blau, G., Paul, A. and St. John, N. (1993). `On developing a general index of work commitment', Journal ofVocational Behavior, 42, 298±314.

Bollen, K. A. (1989). Structural Equations With Latent Variables, Wiley, New York.Breckler, S. J. (1990). `Applications of covariance structure modeling in psychology: Cause for concern',Psychological Bulletin, 107, 260±273.

Brooke, P. P., Russell, D. W. and Price, J. L. (1988). `Discriminant validation of measures ofjob satisfaction, job involvement and organizational commitment', Journal of Applied Psychology, 73,139±145.

Cotton, J. L. and Tuttle, J. M. (1986). `Employee turnover: A meta-analysis and review with implicationsfor research', Academy of Management Review, 11, 55±70.

Dunham, R. B., Grube, J. A. and Casteneda, M. B. (1994). `Organizational commitment: The utility of anintegrative de®nition', Journal of Applied Psychology, 79, 370±380.

Gregerson, H. B. (1993). `Multiple commitments at work and extra-role behavior during three stages oforganizational tenure', Journal of Business Research, 26, 31±47.

Furnham, A. (1990). The Protestant Work Ethic, Routledge, New York.Hackett, D. R., Bycio, P. and Hausdorf, P. (1994). `Further assessment of Meyer's and Allen's (1991) three-component model of organizational commitment', Journal of Applied Psychology, 79, 15±23.

Hall, D. T. and Hall, F. S. (1976). `The relationship between goals, performance, self-image, and involve-ment under di�erent organization climates', Journal of Vocational Behavior, 9, 267±278.

Holland, J. L. (1973). Making Vocational Choices, Prentice-Hall, Englewood Cli�s, NJ.House, R. J. and Rizzo, J. R. (1972). `Role con¯ict and ambiguity as critical variables in a model oforganizational behavior', Organizational Behavior and Human Performance, 7, 467±505.

Hunt, S. D. and Morgan, R. M. (1994). `Commitment: One of many commitments or key mediatingconstruct?' Academy of Management Journal, 37 6, 1568±1587.

Joreskog, K. G. and Sorbom, D. (1993). Structural Equation Modeling With SIMPLIS CommandLanguage, Scienti®c Software, Hillsdale, NJ.

Kanungo, R. N. (1982). `Measurement of job and work involvement', Journal of Applied Psychology, 67,341±349.

Kirchmeyer, C. (1992). `Nonwork participation and work attitudes: A test of scarcity vs. expansion modelsof personal resources', Human Relations, 45, 775±795.

Lawler, E. J. (1992). `A�ective attachment to nested groups: A choice process theory', AmericanSociological Review, 57, 327±339.

Lodhal, T. (1964). `Patterns of job attitudes in two assembly technologies', Administrative ScienceQuarterly, 8, 482±519.

Mathieu, J. E. and Farr, J. L. (1991). `Further evidence of the discriminant validity of measures oforganizational commitment, job involvement, and job satisfaction', Journal of Applied Psychology, 76,127±133.

McGee, G. W. and Ford, R. C. (1987). `Two (or more) dimensions of organizational commitment:Reexamination of the a�ective and continuance commitment scales', Journal of Applied Psychology, 72,638±642.

Medsker, G. J., Williams, L. J. and Holahan, P. J. (1994). `A review of current practices for evaluatingcausal models in organizational behavior and human resources management research', Journal ofManagement, 20, 239±264.

Meyer, P. J. and Allen, J. N. (1984). `Testing the side bet theory of organizational commitment:Some methodological considerations', Journal of Applied Psychology, 69, 372±378.

Meyer, J. P., Allen, N. J. and Gellatly, I. R. (1990). `A�ective and continuance commitment to theorganization: Evaluation of measures and analysis of concurrent and time-lagged relations', Journal ofApplied Psychology, 75, 710±720.

Meyer, J. P., Allen, N. J. and Smith, C. A. (1993). `Commitments to organizations and occupations:Extension and test of a three-component conceptualization', Journal of Applied Psychology, 78, 538±551.

Meyer, J. P., Paunonen, S. V., Gellatly, I. R., Go�n, R. D. and Jackson, D. N. (1989). `Organizationalcommitment and job performance: It's the nature of the commitment that counts', Journal of AppliedPsychology, 74, 152±156.

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

RELATIONSHIPS AMONG FIVE FORMS OF COMMITMENT 307

Mirels, H. L. and Garrett, J. B. (1971). `The Protestant work ethic as a personality variable', Journal ofConsulting and Clinical Psychology, 36, 40±44.

Morrow, P. C. (1983). `Concept redundancy in organizational research: The case of work commitment',Academy of Management Review, 8, 486±500.

Morrow, P. C. (1993). The Theory and Measurement of Work Commitment, JAI Press Inc., Greenwich, CT.Morrow, P. C. and Goetz, J. F. (1988). `Professionalism as a form of work commitment', Journal ofVocational Behavior, 32, 92±111.

Morrow, P. C. and McElroy, J. C. (1986). `On assessing measures of work commitment', Journal ofVocational Behavior, 28, 214±228.

Morrow, P. C., Eastman, K. and McElroy, J. C. (1991). `Concept redundancy and rater naivety inorganizational research', Journal of Applied Social Psychology, 21, 219±232.

Mowday, R. T., Porter, L. M. and Steers, R. M. (1982). Employee±Organizational Linkage, AcademicPress, New York.

Mueller, C. W., Wallace, J. E. and Price, J. L. (1992). `Employee commitment: Resolving some issues',Work and Occupation, 19, 211±236.

Nunnally, J. (1978). Psychometric Theory, 2nd edn, McGraw-Hill, New York.Paullay, I. M., Alliger, G. M. and Stone-Romero, E. F. (1994). `Construct validation of two instrumentsdesigned to measure job involvement and work centrality', Journal of Applied Psychology, 79, 224±228.

Pierce, J. L. and Dunham, R. B. (1987). `Organizational commitment: Pre-employment propensity andinitial work experiences', Journal of Management, 13, 163±178.

Randall, D. M. and Cote, J. A. (1991). `Interrelationships of work commitment constructs', Work andOccupation, 18, 194±211.

Saris, W. and Stronkhorst, H. (1984). Causal Modelling in Non Experimental Research: An Introduction tothe Lisrel Approach, Sociometric Research Foundation, Amsterdam.

Schnake, M. (1991). `Organizational citizenship: A review, proposed model, and research agenda', HumanRelations, 44 7, 735±759.

Schneider, B. (1983). `An interactionist perspective on organizational e�ectiveness'. In: Cameroon, K. S.and Wheeton, D. S. (Eds) Organziational E�ectiveness: A Comparison of Multiple Models, AcademicPress, New York, pp. 27±54.

Schriesheim, C. and Tsui, A. S. (1980). `Development and validation of a short satisfaction instrument foruse in survey feedback interventions'. Paper presented at the Western Academy of Management Meeting.

Shamir, B. (1986). `Protestant work ethic, work involvement and the psychological impact of unemploy-ment', Journal of Occupational Behavior, 7, 25±38.

Siegal, L. (1969). Industrial Psychology, 2nd edn, Irwin, Homewood, IL.Smith, P. C., Kendall, L. M. and Hulin, C. L. (1969). The Measurement of Satisfaction in Work andRetirement, Rand-McNally, Chicago.

Somers, M. J. (1993). `A test of the relationship between a�ective and continuance commitment usingnon-recursive models', Journal of Occupational and Organizational Psychology, 66, 185±192.

Vrendenburgh, D. J. and Sheridan, J. E. (1979). `Individual and occupational determinants of lifesatisfaction and alienation', Human Relations, 32, 1023±1038.

Vandenberg, R. G. and Scarpello, V. (1994). `A longitudinal assessment of the determinant relationshipbetween employee commitments to the occupation and the organization', Journal of OrganizationalBehavior, 15, 535±547.

Wallace, J. E. (1995). `Organizational and professional commitment in professional and nonprofessionalorganizations', Administrative Science Quarterly, 40, 228±255.

Wiener, Y. and Vardi, Y. (1980). `Relationships between job, organization and work outcomes:An integrative approach', Organizational Behavior and Human Performance, 26, 81±96.

Witt, L. A. (1993). `Reactions to work assignments as predictors of organizational commitment:The moderating e�ect of occupational identi®cation', Journal of Business Research, 26, 17±30.

Yoon, J., Baker, M. R. and Ko, J. W. (1994). `Interpersonal attachment and organizational commitment:Subgroup hypothesis revisited', Human Relations, 47, 329±351.

Copyright # 1999 John Wiley & Sons, Ltd. J. Organiz. Behav. 20, 285±308 (1999)

308 A. COHEN