Cohesiveness and organizational citizenship behavior: a multilevel analysis using work groups and individuals – A Special Issue: Focus on Hierarchical Linear Modeling
Roland E. Kidwell, Jr.
Despite their substantial importance, the effects of group, organizational, and environmental contexts on employees’ attitudes and behaviors have been given insufficient attention in the management literature (Griffin & Hofmann, 1997; House, Rousseau, & Thomas-Hunt, 1995; Mowday & Sutton, 1993). Recently however, researchers have successfully used multilevel modeling techniques to empirically examine how variables at the organizational and work group levels might impact individual-level variables and their interrelationships. The effect of organization-level goal congruence on individual attitudes and individual goal congruence (Vancouver, Millsap & Peters, 1994), group cohesion on individual intention to remain in a group (Griffin & Hofmann, 1997), and the relation between individual satisfaction and organizational performance (Ostroff, 1992) are among some of the topics examined.
The thesis of the present paper is that adopting a multilevel perspective would facilitate a better understanding of the role that contextual factors play regarding organizational citizenship behavior (OCB). OCB has been defined as individual helping behaviors and gestures that are organizationally beneficial, but are not formally required (Organ, 1990). In discussing the related concept of prosocial behavior, Brief and Motowidlo (1986) intimated that behaviors like these could be influenced by contextual factors such as group cohesiveness and reciprocity norms. Organ (1990) reviewed potential organizational-level effects, in the guise of market and clan cultures (cf. Kerr & Slocum, 1987), on employee helping behaviors. However, despite the potential for multilevel effects on OCB, previous research has focused exclusively on either the individual (e.g., Smith, Organ & Near, 1983) or group (e.g., George & Bettenhausen, 1990) levels of analysis. One purpose of this study was to determine if multilevel effects on OCB would be manifested in terms of theoretically relevant individual- and group-level variables. Specifically, we examined the relationship of work group cohesiveness with OCB and its effect on the relations of job satisfaction and organizational commitment with OCB.
A second purpose was to demonstrate the use of hierarchical linear modeling (HLM), a methodology introduced recently into organizational research areas, as a procedure for conducting multilevel analyses. The HLM approach to multilevel analysis has a significant history in education research, and has recently received attention in the management literature (cf. Griffin & Hofmann, 1997; Hofmann, Jacobs, & Baratta, 1993; Vancouver et al., 1994). HLM provides for a more robust examination of models for data having two or more levels (Bryk & Raudenbush, 1989, 1992). It is a particularly appropriate analytical strategy to employ in the present study because the focus is on the potential relationships of both individual- (e.g., satisfaction, commitment) and group- (e.g., cohesiveness) level variables with individuals’ OCB.
OCB and Individual-Level Effects
OCB involves discretionary behavior that helps co-workers, supervisors, and the organization. Assisting newcomers to the organization, not abusing the rights of co-workers, not taking extra breaks, attending elective company meetings, and enduring minor impositions that occur when working with others are examples of OCB that help in coping with various organizational uncertainties. A key element to OCB is voluntarily aiding others with job-related problems. Multidimensional delineations have identified OCB facets such as conscientiousness, sportsmanship, civic virtue, courtesy, and altruism (Podsakoff, MacKenzie, Moorman, & Fetter, 1990). Other researchers have divided OCB into two types: behavior that is directed mainly at individuals in the organization (OCBI), and behavior that is concerned more with helping the organization as a whole (OCBO) (Williams & Anderson, 1991). Courtesy and altruism are viewed as mainly benefitting coworkers, whereas conscientiousness, sportsmanship and civic virtue are directed at the organization (cf. Van Dyne, Cummings & Parks, 1995; Williams & Anderson, 1991).
Global job satisfaction and affective commitment are among the affective work reactions that have been offered most often as antecedents to affiliative/promotive extra-role behavior (Van Dyne et al., 1995). Studies have found individuals’ job satisfaction and organizational commitment to be associated with several OCB facets (Bateman & Organ, 1983; O’Reilly & Chatman, 1986; Organ, 1990; Puffer, 1987; Smith, Organ & Near, 1983). For example, Smith, et al. (1983) found a causal linkage between job satisfaction and the OCB dimension of altruism. Other studies have found evidence of significant correlations between satisfaction and OCB components (e.g., Puffer, 1987). When defined as a psychological identification with the organization and its values, organizational commitment has also displayed links with OCB (O’Reilly & Chatman, 1986; Organ, 1990). More recently, Morrison (1994) found positive relationships between affective commitment and several OCB dimensions, though these were mediated by job breadth.
Connections between satisfaction, commitment and OCB at the individual level may result because positive attitudes about the job or the organization tend to predispose people toward extra-role behavior (Bateman & Organ, 1983; Van Dyne, Graham & Dienesch, 1994). In addition, high levels of job satisfaction and commitment can create equity pressures that motivate individuals to provide nonrequired helping behaviors as repayment for the fulfillment and belongingness they draw from their work (Schnake, 1991). Because previous theory and research have linked individuals’ satisfaction and commitment with OCB, we considered satisfaction and commitment as relevant individual-level antecedent variables in this study.
OCB and Contextual Effects
The term ‘context’ refers to surroundings associated with a particular phenomenon, and involves units of analysis expressly above those being examined (Cappelli & Sherer, 1991). In turn, the social context is described by relational phenomena that cannot be understood in terms of individuals separately. Characteristics of the environment, organization, and work group are elements of the social context that have an impact on the attitudes and behaviors of individual group members. For individuals working in organizations, perhaps the most prominent social context is their immediate work group (Hackman, 1992).
It is a basic assumption of much contemporary group research that group context influences individual attitudes and behavior. This influence is typically thought to occur directly through interaction with members’ characteristics, as well as indirectly through the subtle but potent effects of the group as a totality. Interest in such effects can be traced as far back as Cattell’s (1948) notion of syntality – those activities that synergistically combine to make a group a unique entity. A work group is a system involving a complex pattern of relations among a set of people using a set of technologies to accomplish mutual goals (Arrow & McGrath, 1995). Whether such a system is recognized as a group depends on the degree that the people involved in the system: a) consider themselves to be members of a group and recognize one another as members, b) have relations shaped by mutual task goals and task interdependence, and c) share a set of tools, rules, or procedures. In sum, at the core of many work groups is the emergence of a social identity among members and a pattern of interdependence-based, recurring relationships. Though elements of these two characteristics may be observed in connection with group processes involving constructs such as conformity, perhaps they are captured most succinctly by the concept of group cohesiveness.
Long considered an important element of group dynamics, cohesiveness describes group members’ affinity for one another and their desire to remain part of the group. In cohesive work groups, individuals tend to be more sensitive to others and are more willing to aid and assist them (Schachter, Ellenson, McBride & Gregory, 1951). Leader behavior exhibiting high consideration toward group members has a greater impact within groups that are cohesive (Schriesheim, 1980). Cohesivehess has been shown to lead to greater intra-group communication, stronger group influence, and more favorable interpersonal evaluations within groups (Cartwright, 1968). It also enhances the potential importance of other members’ evaluation of the behaviors of each group member, due to a desire to preserve the group.
Perhaps the most salient theoretical basis for possible group-level OCB effects stems from research on social exchange and helping. With regard to social exchange, one could expect that cohesive groups would display more positive and frequent social exchanges than noncohesive groups. Some researchers (e.g., Organ, 1990) have suggested that OCB may reflect members’ efforts to maintain exchange relationships within the group that are more social than economic. Work groups characterized by liking and cooperation may encourage trust in the long run that social exchanges will be reciprocated. OCB may act as one medium of exchange in such contexts and may be expected of group members.
Group cohesivehess has been identified as an important situational antecedent to affiliative/promotive behaviors (Van Dyne et al., 1995) like OCB because highly cohesive groups engender a strong social identity that can enhance members’ desires to help one another. In a study conducted at the group level, George and Bettenhausen (1990) found that group cohesivehess correlated with a group measure of prosocial behavior. Additionally, George and Bettenhausen (1990) noted that cohesiveness may impact OCB through its broader effect on group members’ affective states. Members of cohesive work groups experience more positive mood states than do members of noncohesive groups (Gross, 1954; Marquis, Guetzkow & Heyns, 1951). The social psychological research literature on helping behavior has shown that positive mood states may induce or correlate with proclivities to exhibit altruism toward others (cf. Isen & Baron, 1991), at least at the individual level of analysis.
Research on group process variables provides additional general support for potential group-level effects of cohesiveness on OCB. Broadly speaking, the more cohesive a work group, the greater the conformity to group norms. As Hackman (1992) notes, norm conformity is higher because of the pressures exerted by members on one another and the interpersonal rewards that are available through within-group interactions. It should be emphasized that although the group norms literature is supportive of group-level OCB links with cohesiveness, such support is conditional on whether OCB is considered important by members for group functioning. This is the case because group norms generally form only around behavior that is important to group functioning (Cartwright, 1968).
Where the work context is such that citizenship-type behaviors might contribute to group functioning (e.g., where coordinated service demands exist – see George & Bettenhausen, 1990), OCB on the part of work group members could become well established when group cohesiveness is high. Expectations of cooperation and social responsibility may be internalized by group members in the form of values that when practiced would increase feelings of self-worth, and when not practiced would arouse negative feelings and decrease members’ sense of self-worth (Shamir, 1990). The commonality of potential benefits to members increases the possibility that OCB may become a repeated plays occurrence (Axelrod, 1984) and that these behaviors will become increasingly visible as they are reciprocated among group members. In such situations, members may become models for one another in demonstrating appropriate OCB (cf. Schnake, 1991). Logically then, it is reasonable to expect that a “norm of fair dealing” (Stroebe & Frey, 1982, p.127) involving discretionary behaviors (Hackman, 1992) like OCB would be established in groups that are highly cohesive as compared to groups that are not cohesive.
The above logic supports the possibility that across groups, the amount of cohesiveness in work groups will be related to the general amount of OCB shown by work group members. However, work group differences in the amount of OCB may not be the only type of group-level effect associated with work group cohesiveness. Work group cohesiveness may serve a moderating function as well. In general, behavioral norms may not be as well defined in groups with low cohesion (Dobbins & Zaccaro, 1986). Whereas cohesiveness creates expectations stimulating individual members to exhibit extra-role behavior exchanges, absent such expectations, group members may not see a need to respond accordingly. In other words, social exchange relationships are likely to be weaker where cohesiveness is weaker, and individuals may not feel an obligation to reciprocate OCB displayed by other group members. In work groups lower in cohesiveness, members experiencing high satisfaction or organizational commitment may, instead of displaying the OCB that would be expected in highly cohesive groups, choose to behave in other ways. For example, they could concentrate instead on in-role or self-promotional behaviors. In any case, work group cohesiveness may exert a group-level effect by influencing the strength of relations between OCB and its affective antecedents.
There is substantial empirical evidence supporting the notion that certain individual-level variables (e.g., job satisfaction, organizational commitment) correlate with individuals’ OCB. Konovsky and Pugh (1994) have noted that a comprehensive theory of OCB should include group-level situational variables. As argued above and elsewhere (e.g., George & Bettenhausen, 1990; Van Dyne et al., 1995), it also appears that there are sufficient theoretical reasons to expect that work group cohesiveness, a group-level variable, influences individuals’ OCB within work groups. What has yet to be tested, however, is the joint effect of both individual- and group-level variables on OCB. Such an investigation is the fundamental purpose of multilevel analysis.
The potential multilevel effects of cohesiveness on OCB could be manifested in at least two ways. The first would be if group-level variables explained variance in individuals’ OCB beyond that which could be accounted for by the individual-level variables of satisfaction and commitment. The second way that contextual effects could be manifested is in terms of variation in the nature of the relations between individual-level variables and OCB across groups. For example, if the relation between satisfaction and OCB varied according to groups’ cohesiveness, one could surmise that this contextual variable was exerting a systematic influence.
Thus, in order to investigate possible multilevel effects involving individual- and group-level variables, we tested the following two hypotheses:
H1: Work group cohesiveness will have a positive relationship with the amount of OCB exhibited by employees beyond that accounted for by individual satisfaction or commitment.
H2: Work group cohesiveness will moderate the relationships of job satisfaction or organizational commitment with OCB, such that the relations will be stronger as work group cohesiveness increases.
Testing for multilevel effects with HLM
Traditionally, researchers testing multilevel models have had two data analysis options. The first was to assign the higher level measure to each unit at the lower level (e.g., assign group scores to individuals), and then conduct analyses strictly at the lower level. The second alternative was to aggregate measures taken at the lower level of analysis (e.g., aggregating individual-level measures to form group-level composites) and conducting analyses at the higher level only. Each of these options has potential empirical and conceptual weaknesses. With the first option, the researcher must assume that individual responses are not influenced by group characteristics. This approach yields biased estimates of the standard errors and increases the chance of Type I error (Burstein, 1980). With the second option, statistical power often is an issue, as is the appropriateness of inferences concerning relations among the aggregated variables (Klein, Dansereau & Hall, 1994).
HLM allows for the investigation of both within- and between-group effects on an individual-level dependent variable through an empirical Bayesian estimation process in which two different models are estimated iteratively. A within-group or “level-1” analysis is used to estimate two separate parameters describing the relationship between the predictors and the focal dependent variable within each group (i.e., within-group intercept and slope). These intercept and slope parameters obtained from the level-1 analysis serve as the dependent variables in equations used for a between-group or “level-2” analysis. A group-level or contextual effect is suggested by the presence of a significant parameter estimate (gamma coefficient, [Gamma]) for level-2 predictors of the level-1 intercepts. This provides the operational test for our first hypothesis. Further, a significant gamma associated with a level-2 predictor in an equation modeling variance in the slope estimates indicates that the variable moderates the relationship between level-1 independent and dependent variables (cf. Bryk & Raudenbush, 1992). This provides the operational test for our second hypothesis.
The sample for this study consisted of 260 individuals from 49 work groups in eight organizations. All of the organizations could be classified as belonging to the service sector: three community banks, a state bank, a regional bank, a chain of farm implement stores, and organizations specializing in telemessaging and heating and cooling services. Despite their diversity, all of these organizations had employees performing in customer-driven, information-rich task roles requiring degrees of coordination, information exchange, and interpersonal interaction. For example, types of employees responding from the surveyed organizations included banks tellers, loan officers, and customer service specialists from the banks, operators and service specialists from the telemessaging company, and sales, equipment service, and customer service personnel from other organizations in the sample. Thus, the work environments in which the groups operated provided opportunities for meaningful OCB to occur.
The work groups ranged in size from three to nine members (M = 5.69; sd = 1.49), which allows for the meaningful study of small group cohesiveness (cf. Carron & Spink, 1995; Hare, 1981). Employee surveys were administered by one of the authors during work hours in conference rooms away from work areas. Both oral and written instructions specified that respondents answer group-related items in regard to their current work group. Respondents were also told that “work group” referred to those people with whom they worked on a recurring basis, as opposed to people belonging to the same larger organizational units, such as departments or divisions. The supervisors of the work groups responded to questionnaires evaluating the citizenship behaviors of each of their subordinates, completing the employee evaluations in private sessions with one of the authors. All study participants were assured that responses were confidential and would be seen only by the researchers.
Overall, 88 percent of Ire surveyed employees were white (86 percent of the minorities were African-American), 21 percent were male, and 86 percent had less than a college degree. Their ages ranged from 17 to 67, with a mean age of 36. Of the surveyed employees, 95 percent were supervised by whites, 52 percent by men, and 62 percent by someone without a college degree.
Job satisfaction, organizational commitment, and work group cohesivehess were collected through the employee survey. Organizational citizenship behavior was reported by supervisors. A 7-point Likert response format was used in each instance; scales are coded such that a high score indicates a high amount of the focal construct. All scales were summed and divided by the number of items in the scale.
Job satisfaction. Job satisfaction was measured with a seven-item scale ([Alpha] = .85) based on work by Chalykoff and Kochan (1989). Sample items include “I am satisfied with my job,” “I am satisfied with my pay,” and “I am satisfied with the recognition I receive for a job well done.”
Organizational commitment. Commitment to the organization was measured with a six-item scale ([Alpha] = .80) adapted from Allen and Meyer (1990). Sample items include “I would be very happy to spend the rest of my career with this organization,” “I do not feel ’emotionally attached’ to this organization” (reverse coded), and “This organization has a great deal of personal meaning for me.”
Organizational citizenship behavior. Van Dyne et al. (1995) state that Williams and Anderson’s (1991) OCBI/OCBO delineation provides a useful distinction among various OCB operationalizations. In line with this idea, we included in the study two facets from a multidimensional OCB measure that capture this distinction. The Podsakoff, et al. (1990) OCB scale was administered to supervisors to collect their assessment of each employee’s conscientiousness ([Alpha] = .84) and courtesy ([Alpha] = .92). Conscientiousness includes attendance, cleanliness, and punctuality that go beyond normal requirements and represents the more impersonal OCBO facet. Representing the OCBI facet, courtesy involves informational and interpersonal interactions with co-workers. It is especially pertinent given its conceptual connection with cohesiveness. Also, empirical research has uncovered correlations between courtesy and individuals’ perceptions of group cohesiveness (Podsakoff, MacKenzie & Bommer, 1996).
Sample items gauging conscientiousness (OCBO) included “This employee comes to work more often than the average employee,” “This employee does not take extra breaks,” and “This employee believes in giving an honest day’s work for an honest day’s pay.” Sample items tapping courtesy (OCBI) included “This employee takes steps to try to prevent problems with other co-workers,” “This employee does not abuse the rights of others,” and “This employee considers the impact of his/her actions on co-workers.”
Work group cohesiveness. In his review of cohesiveness research and measures, Mudrack (1989) suggested that the field move beyond conceptualizing cohesiveness simply in terms of attraction to – or positive attitudes about – a group. Additionally, he emphasized that mutual commitment among group members to one another and their tasks be considered. Mudrack (1989) cited a group cohesiveness measure developed by Dobbins and Zaccaro (1986) as one that moved in this conceptual direction and was easily adaptable for organizational use. Because identification with the group (i.e., commitment of members to one another) was held to be part of the underlying reason for our hypothesizing group-level effects, it was important for the cohesiveness measure to tap this concept. Therefore, we measured group cohesiveness with an 8-item scale ([Alpha] = 89) modified from Dobbins and Zaccaro (1986).
The items that assessed the identification component of cohesiveness were “I feel that I am really part of my work group,” “I look forward to being with the members of my work group each day,” “The work group I belong to is a close one,” and “I enjoy belonging to this work group because I am friends with many group members.” Other items measured members’ desire to remain part of the group and willingness to defend it. Within each group, responses to this scale were aggregated to form the cohesiveness context measure.
Within- and between-group variance in work group cohesiveness. In conducting a meta-analysis of relationships between cohesiveness and performance, Gully, Devine and Whitney (1995) recently noted the importance of using appropriate group-level measures or aggregation procedures when examining group-level work group cohesion issues. In order to proceed with the multilevel analysis, it was first necessary to assess the appropriateness of our measure of work group cohesiveness. Specifically, prerequisites to the hypothesis tests included demonstrations that (a) there was substantial within-group agreement as to the group’s cohesiveness, and (b) that there was substantial between-group variation in cohesiveness. Without the former, our use of an aggregate measure would be unjustified. Without the latter, no evidence for a group effect exists.
To examine the homogeneity of cohesiveness perceptions within work groups, we computed interrater agreement indices ([r.sub.wg] – James, Demaree, & Wolf, 1984, 1993). James and colleagues have argued that [r.sub.wg] indicates the degree to which group members concur in their individual assessments regarding a contextual variable. Though [r.sub.wg] is not without limitations, on balance it has been suggested as a suitable agreement index within certain bounds (Kozlowski & Hattrup, 1992). To examine between-group variation in perceptions of cohesiveness, we computed eta coefficients (James, 1982).
Hierarchical Linear Modeling. As outlined by Hofmann (this volume), a number of conditions must be met before testing our hypotheses. The degree to which our data meet these conditions is determined by conducting a sequence of models. First, because theory suggests that OCB will be predicted by both individual- and group-level variables, there should be systematic within- and between-group variance in OCB. This is assessed by conducting a one-way analysis of variance (or, in HLM terms, estimating the null model). This model partitions variance into within- and between-group components and provides a statistical test of the between-group variance estimate ([[Tau].sub.00]).
Second, there should be significant between-group variance in the intercepts
([[Beta].sub.0j]) and slopes ([[Beta].sub.1j]) estimated in the level-1 model. In order to consider hypothesis 1, between-group variance in the intercepts is necessary. To consider hypothesis 2, between-group variance in the slopes is required. A random-coefficients regression model is used to assess the degree to which our data meet these requisites.
If it is established that these conditions have been satisfied, then a third model is estimated to actually test hypothesis 1. This model, which we label the intercepts-as-outcomes model, considers whether variance in the intercepts ([[Beta].sub.0j]) from the within-group regressions is associated with the level of work group cohesiveness. A significant t-test for the coefficient associated with cohesiveness ([[Gamma].sub.01]) supports the hypothesis.
As Hofmann (this volume) discusses, an important consideration in hierarchical linear modeling is the method of centering chosen. Since the intercept term [[Beta].sub.0j] is of substantive importance in the level-2 analysis, it is critical that it be interpretable. However, the intercept is difficult to interpret when the level-1 independent variable is not ratio in scale. For this reason, the HLM software provides for a number of centering options (see Bryk & Raudenbush, 1992, for a detailed discussion of centering options in HLM). The exact interpretation of [[Beta].sub.0j] depends on the chosen method of centering. For models used to test hypothesis 1, we centered individuals’ job satisfaction and organizational commitment scores around the grand mean. Therefore, the intercept [[Beta].sub.0j] equals the expected value of OCB when satisfaction (or commitment) is at the sample mean. Grand mean centering has been shown to provide equivalent model fits as raw metric approaches, but usually results in a computational advantage by reducing the covariance between the intercept and slope parameters. It also produces intercept terms that represent the between-group variance in the outcome measure, here OCB, after controlling for the level-1 independent variable (see Bryk & Raudenbush, 1992, p. 26). This is the appropriate choice for testing hypothesis 1.
If between group variance in the slopes ([[Beta].sub.1j]) has been detected in the random coefficients regression model mentioned above, a fourth model is then estimated to test H2. This model, which we label as the slopes-as-outcomes model, considers whether or not variance in the slopes ([[Beta].sub.1j]) from the within-group regressions is associated with work group cohesiveness. The interpretation of [[Beta].sub.1j] is intuitive: [[Beta].sub.1j] represents the strength of the relationship between job satisfaction (or commitment) and OCB in each of the work groups. If the t-test indicates that cohesiveness ([[Gamma].sub.11]) is significantly associated with [[Beta].sub.1j], it would suggest that work group cohesiveness moderates the relationship between job satisfaction or organizational commitment and the focal OCB.
In testing this hypothesis, the use of grand mean centering is no longer appropriate. Raudenbush (1989a, 1989b) has suggested that in testing cross-level moderation models, group mean centering yields a more consistent estimate of the within-group slope [[Beta].sub.1j] – the dependent variable) which, in turn, provides a more accurate assessment of the cross-level moderation. For this reason, we group mean centered individuals’ job satisfaction and organizational commitment in the model testing H2. Because of this action, our model no longer included the between-group variance in these predictors that could explain variance in individuals’ OCBs. To correct for this, in testing H2 an aggregate job satisfaction (or organizational commitment) measure was used to reintroduce this variance back into the model as a predictor of the level-2 intercept term. This aggregate measure is the mean of individual scores for each work group and, as such, reintroduces the between-group variance in the level-1 independent variable as a control in our tests of H2.
[TABULAR DATA FOR TABLE 1 OMITTED]
Table 1 reports means, standard deviations, and intercorrelations for the variables included in the analysis. In terms of the prerequisites for the HLM analysis, the following additional results are relevant. First, the mean [r.sub.wg] for work group cohesivehess was .70. Second, the analysis of variance indicates that 43 percent of the variance ([[Eta].sup.2]) in cohesiveness lies between groups. Though there are no set standards of acceptability for [r.sub.wg] or [[Eta].sup.2], the magnitudes reported here are similar with those reported in other multilevel research (e.g., George & Bettenhausen, 1990; James, 1982). based on these results together, we inferred that the use of our aggregate measure of work group cohesiveness was tenable.
As described above, we conducted a sequence of models using the HLM2L statistical package (Bryk, Raudenbush & Congdon, 1996).
First, we ran null models for conscientiousness and courtesy to assess whether our data met the condition that there be systematic between-group variance in these measures. Both the results for conscientiousness ([[Tau].sub.00] = .25, df = 48, [[Chi].sup.2] = 96.95, p [less than] .001) and courtesy ([[Tau].sub.00] = .46, df = 48, [[Tau].sup.2] = 130.13, p [less than] .001) suggest that this condition has been satisfied. Estimating the null model also produces information necessary for computing intraclass correlation coefficients (p) that indicate the proportion of between-group variance in conscientiousness and courtesy (cf. Bryk & Raudenbush, 1992). These statistics indicated the amount of variance in these variables that could potentially be explained by our level-2 predictor variable, work group cohesiveness. The results of this analysis indicate that 16% of the variance in conscientiousness lies between work groups, 25% of the variance in courtesy lies between work groups.
Next, the random coefficients regression model is estimated in order to assess whether there was systematic between-group variance in the intercept parameter [[Beta].sub.0j] and the slope parameter [[Beta].sub.1j]. The [[Chi].sup.2] tests for the estimates of the intercept ([[Tau].sub.00]) and slope ([[Tau].sub.11]) variances provide this test. Two further pieces of information are available through this analysis. First, a significant result for the [[Gamma].sub.10] parameter indicates that the independent and dependent level-1 variables are related. Second, comparing the estimates of the residual variance ([[Sigma].sup.2]) produced in the null and the random coefficients regression model allows for the computation of the variance explained ([R.sup.2]) at level- 1.
For each OCB measure, analyses were conducted separately for job satisfaction and organizational commitment. So, in total, four random coefficient regression models were estimated. Regarding conscientiousness, the results indicate that for both job satisfaction ([[Tau].sub.00] = .27, df = 48, [[Chi].sup.2] = 91.54, p [less than] .001) and organizational commitment ([[Tau].sub.00] = .25, df = 48, [[Chi].sup.2] = 88.70, p [less than] .001), there was significant variance in the intercept parameters. In other words, the variance in the intercept parameters were significantly different from zero. However, for neither variable were variance estimates for the slope term ([[Tau].sub.11]) significant. In sum, these results suggest that we could test for relationships between work group cohesiveness and conscientiousness (H1), but not the moderating effect of cohesiveness on relations between the individual-level independent variables and conscientiousness (H2).
With regard to courtesy, the results suggest that there was systematic between-group variance in both the intercept and slope parameters. Specifically, the intercept variance in both models is significantly different from zero (job satisfaction: [[Tau].sub.00] = .54, df = 48, [[Chi].sup.2] = 119.75, p [less than] .001; organizational commitment: [[Tau].sub.00] = .48, df = 48, [[Chi].sup.2] = 118.25, p [less than] .001). Further, the slope variance parameter in both models is significantly different from zero (job satisfaction: [[Tau].sub.11] = .10, [[Chi].sup.2] = 48, [[Chi].sup.2] = 80.39, p [less than] .01; organizational commitment: [[Tau].sub.11] = .10, df = 48, [[Chi].sup.2] = 67.46, p [less than] .05).
As mentioned earlier, the random coefficients regression also provides results that address relationships between the level-1 variables. Specifically, the t-test for the [[Gamma].sub.10] parameter assesses whether the pooled level-1 slopes differ from zero. In these data, there is a relationship of conscientiousness with job satisfaction ([[Gamma].sub.10] = 16, se = .06, t = 2.53, p [less than] .05), but not with organizational commitment ([[Gamma].sub.10] = .07, se = .08, t = 1.17). The same pattern emerges for courtesy in that there is a significant relationship between courtesy and job satisfaction ([[Gamma].sub.10] = .13, se = .08, t = 1.65, p [less than] .10), but not with organizational commitment ([[Gamma].sub.10] = .03, se = .08, t = .38). These results are not surprising given the zero-order correlations reported in Table 1. Finally, we computed level-1 [R.sup.2] for those models where a significant [[Gamma].sub.10] was obtained. For the model where job satisfaction predicted conscientiousness, the [R.sup.2] was .05, for the model where job satisfaction predicted courtesy, the [R.sup.2] was. 13.
Next, we estimated the intercepts-as-outcomes models in order to test hypothesis 1, that individuals would display higher levels of OCB in more cohesive work groups than would be expected based on their individual levels of job satisfaction or organizational commitment. These results are reported in the top portion of Table 2. The operative test of the hypothesis is the significance of the [[Gamma].sub.01] coefficient. The results with conscientiousness as the focal OCB were not significant; work group cohesivehess was not associated with the amount of conscientiousness displayed by work group members. However, when examining courtesy, we found support for hypothesized effects for both job satisfaction ([[Gamma].sub.01] = .34, se = .17, t = 2.06, p [less than] .05) and organizational commitment ([[Gamma].sub.01] = .35, se = .18, t = 2.02, p [less than] .05). Additional information relevant to these effects is contained in the variance parameter, [[Tau].sub.00]. A significant [[Tau].sub.00] indicates that there is residual variance in the intercept parameter that could be explained by additional group-level measures. In all four models estimated, the [[Tau].sub.00] was significant. Finally, it is possible to compute the amount of intercept variance explained ([R.sup.2]) by cohesiveness. In these data, work group cohesiveness explained 5.6% of the intercept variance in the job satisfaction – courtesy model and 6.3% of the intercept variance in the organizational commitment – courtesy model.
Table 2. Results of the Level-2 Analyses for Courtesy
Fixed effects Gamma Coefficients Standard error
Cohesiveness, [[Gamma].sub.01] .34(*) .17
Cohesivehess, [[Gamma].sub.01] .35(*) .18
Cohesiveness, [[Gamma].sub.11] 1.05(**) .52
Cohesiveness, [[Gamma].sub.11] .02 .12
Notes: n = 49 work groups; * p [less than] .05, ** p [less than]
Recall that these data did not meet the necessary conditions for modeling a cross-level interaction for work group cohesiveness on the relationship between the level-1 independent variables and conscientiousness. As a result, hypothesis 2 was examined only for the courtesy dimension. Hypothesis 2 stated that the relationship between job satisfaction or organizational commitment and OCB would be stronger in work groups that were more highly cohesive. The results reported in the lower portion of Table 2, show partial support for the hypothesis. Specifically, cohesiveness did moderate the relationship between job satisfaction and courtesy, in the direction predicted ([[Gamma].sub.11] = 1.05, se = .52, t = 2.02, p [less than] .01). Comparing the [[Tau].sub.11] parameter from this model with that obtained in the intercepts-as-outcomes model yields an [R.sup.2] of .13, meaning that cohesiveness accounts for 13% of the variance in the satisfaction-courtesy slopes across groups. Significant results were not obtained for the model involving courtesy and organizational commitment. As was true in the intercepts-as-outcomes model, the residual variance parameter, [[Tau].sub.11], was significant in both models, indicating that additional group-level variables might explain further slope variance.
The findings of this study indicate that social context, as revealed through group cohesiveness, affects the amount of OCB displayed in work groups as well as relationships between affective work reactions and OCB. The study also demonstrates the application of hierarchical linear modeling as a research tool in studying potential multilevel effects. By taking the hierarchical nature of data into account in modeling phenomena that may affect OCB, more information is used to increase the power and precision of data-based estimates (de Leeuw & Kreft, 1995; Hofmann, this volume).
Three of the four hypothesis tests involving the courtesy dimension were supported. Employees in more cohesive work groups were rated by their supervisors as engaging in higher amounts of courtesy than could be explained by their individual levels of job satisfaction or organizational commitment. As a form of OCBI, courtesy involves helping types of behaviors that are directed at co-workers. Using individual-level variables in a social exchange framework, Anderson and Williams (1996) recently confirmed that quality working relationships are integral in bringing about helping behavior among co-workers. We would suggest that work group cohesiveness may serve to establish a context favorable to social exchanges among group members. This increases the chances of quality work relationships developing and encouraging norms or expectations concerning behaviors that would reinforce such relationships. In such a context, courtesy might function as a medium of group-oriented social exchanges that could flow freely among group members without necessitating calculative, quid pro quo arrangements. Work group cohesiveness may also occasion increased empathic concern among work group members, which has been linked theoretically with interpersonal types of citizenship behavior (Settoon & Mossholder, 1996).
A cross-level moderator effect was detected in connection with the relationship between courtesy and job satisfaction, in that the relationship between the two became stronger as work group cohesiveness increased. This result may be explained by placing individual-level OCB research findings within a cross-level perspective. In individual-level OCB research, it has been demonstrated that when people experience positive affective states, such as being satisfied with their jobs, they are more apt to be prosocial (e.g., Isen & Baron, 1991). Higher work group cohesiveness may act as a contextual catalyst for social exchange processes, making it easier for satisfied individuals to act on their tendency to demonstrate courteous citizenship behavior toward other group members. This might occur especially where the task-related activities of work group members were interconnected and required interaction, as was the case with the work groups investigated. We also note that a moderator effect was not found in connection with organizational commitment, however. This may be due to the fact that co-workers rather than organizations are the behavioral referents for courtesy, possibly rendering relations between commitment and courtesy more tenuous even under conditions where social exchanges are prevalent.
None of the hypotheses involving conscientiousness were supported. One explanation for this concerns the target of this OCB dimension. We selected the conscientiousness dimension because it is representative of OCBO, the beneficiary of which is the organization. The OCB facilitating context established by work group cohesiveness may be relevant mostly for citizenship behaviors directed at co-workers. In other words, work group cohesiveness engenders citizenship behavior that is manifested in ways that improve the lot of immediate coworkers rather than the organization as a whole. Speculatively, a different pattern of findings could have emerged if we had considered a group-level characteristic that was more directly tied to broader feelings about the organization (e.g., resources provided to the work group).
It should be noted that the group-level OCB effects in our study may be considered as incremental to those attributable to the individual-level variables of satisfaction and commitment. As discussed at the outset, much research has linked these two variables with OCB. This study was not designed to consider the role that work group cohesiveness may play in affecting individuals’ satisfaction and commitment. Undoubtedly, there are linkages among these variables and further consideration of them is warranted in future investigations. Due to the increased complexity of multilevel investigations, explaining the simultaneous presence (or absence) of individual- and group-level effects is more difficult than explaining either type of effect alone. This underscores the importance of basing multilevel investigations on sound theory (Klein et al., 1994; Mossholder & Bedeian, 1983), and performing confirmatory studies on exploratory studies of multilevel phenomena (Draper, 1995).
Whereas this study focused on OCB, more controversial types of extra-role behavior, such as whistleblowing and principled organizational dissent have been identified and could also be investigated (Van Dyne et al., 1995). Additionally, a complete consideration of OCB and contextual effects may incorporate three levels, the individual members, the group, and the organization, and the possible interchanges among them (cf. Arrow & McGrath, 1995). The relationship of group cohesiveness to individual members’ affect and behavior was the subject of this study. Organization-level variables were not taken into account regarding their influences on groups and their individual members. For example, the average member of the highly cohesive group might be more likely to exhibit OCB if the goals of the group were congruent with those of the organization. If the group and organizational goals were in conflict, this condition might diminish the amount of OCB within the highly cohesive group. Members, however, might compensate by engaging in more discretionary behavior specifically focused on other individuals within the group.
Hopefully, the current study will encourage researchers to continue developing theory relating to OCB in a multilevel framework. Various sources exist that may be useful in developing theory regarding work group dynamics as a reflection of social context (Arrow & McGrath, 1995; Hackman, 1992; House et al., 1995). A full understanding of how OCB operates within work groups may be difficult to reach unless social context is taken into account.
In closing, brief mention of some limitations of this study should be made to place our results in proper perspective. The cross-sectional nature of the study renders it vulnerable to problems typically associated with survey research. It should be noted, however, that our group-level effects are not easily attributable to same source or common method variance, as they stem from an aggregate response rather than that of any particular person. Nevertheless, a longitudinal study might allow for greater insight into the development of cohesiveness within work groups, and whether displays of OCB track this development. Also, as is noted below, issues of power can be a concern when studying small groups or a small number of groups. Smaller group sizes may result in estimates with greater standard errors in level-1 parameters, making prediction by level-2 variables more difficult. Our level-2 variable, cohesiveness, is intrinsically linked with group size (Hare, 1981) and has been studied within the confines of small groups. Though the number of groups used in the study (n = 49) lessens somewhat concerns about power for level-2 equations, we did not feel it appropriate to arbitrarily stipulate minimum group sizes because of the number of studies that examine cohesiveness in groups of three (and greater) individuals (Carron & Spink, 1995; Widmeyer, Brawley & Carron, 1990). Moreover, level-2 effects were detected despite a potential group size bias against finding them, reducing the salience of this issue in the present study. Issues involving the strength of cohesiveness effects within groups of a certain size versus the stability of HLM estimates for groups of a certain size should be studied further, however.
Finally, it should also be noted that only one source of OCB ratings was used in the study: supervisors. This rating source controls for the problem of common method variance often found in survey research. However, it would have been advantageous to gather multi-source OCB ratings, as OCB may be assessed differently by members of the work groups. Co-workers especially may have been in a better position to judge individuals’ OCBI-oriented activities. As Van Dyne et al. (1995) observed, whether a behavior is considered in-role or extra-role is a function of the rater, the actor, and the relationship between these two persons over time.
Though from an analytical perspective HLM has a number of advantages for multilevel research, some cautions should be noted. First, as with other statistical techniques, a number of assumptions are required in making inferences from HLM modeling results. HLM procedures involve certain assumptions about the data (see Bryk & Raudenbush, 1992; Hofmann, this volume) which when violated reduce the confidence with which statistical inferences may be made. The relatively higher complexity of random coefficient models typically used in HLM may make greater demands on these assumptions (Kreft, 1996; de Leeuw & Kreft, 1995). For example, with random independent variables, it becomes more difficult to satisfy the assumption that independent variables are uncorrelated with errors (disturbances) in model equations, which in turn has ramifications for the assumption of multivariate normality underlying HLM. Second, despite its potential advantages, HLM’s random coefficients approach may be unnecessary for some multilevel data structures. Traditional techniques perform as well or better if there are large groups and small intraclass correlations, and the interest is only in fixed-level regression coefficients (de Leeuw & Kreft, 1995). Lastly, with HLM, as with other analytical procedures, researchers must be mindful of power issues when designing and interpreting their results. HLM level-1 parameter estimates require an adequate number of observations, just as other estimation procedures (e.g., ordinary least squares) do. But in addition, level-2 parameters require an adequate number of units (i.e., groups, organizations), with the number of within-units observations being of minor importance (Kreft, 1996). Though a lower bound requisite for the number of units needed depends on several things, Kreft (1996) suggests that 30 might be a good figure, all else being equal.
Conceptual models are not supportable on the basis of elegance alone, as has been shown in other areas where sophisticated methodologies have been employed (e.g., structural equation modeling). By its very nature, multilevel analysis often involves greater complexity than an analysis focusing on one level. For this reason, HLM should be used with thoughtful consideration regarding the hierarchical nature of the theory being investigated and the multiple levels in which correspondingly relevant data may be structured. Under such circumstances, it can provide increased flexibility in modeling multilevel phenomena and for understanding the hierarchical relationships found within organizations.
Acknowledgment: We wish to thank David Hofmann and Larry James for their comments on a previous version of this paper.
Allen, N.J. & Meyer, J.P. (1990). The measurement and antecedents of affective, continuance and normative commitment to the organization. Journal of Occupational Psychology, 63: 1-18.
Anderson, S. E. & Williams, L. J. (1996). Interpersonal, job, and individual factors related to helping processes at work. Journal of Applied Psychology, 81: 282-296.
Arrow, H. & McGrath, J. E. (1995). Membership dynamics in groups at work: A theoretical framework. Research in Organizational Behavior, 17:373-411.
Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books.
Bateman, T. S. & Organ, D. W. (1983). Job satisfaction and the good soldier: The relationship between affect and employee citizenship. Academy of Management Journal, 26: 587-595.
Brief, A. P. & Motowidlo, S. J. (1986). Prosocial organizational behaviors. Academy of Management Review, 11: 710-725.
Bryk, A. S. & Raudenbush, S. W. (1989). Methodology for cross-level organizational research. Research in the sociology of organizations, 7: 233-273.
Bryk, A. S. & Raudenbush, S. W. (1992). Hierarchical linear models: Applications and data analysis methods Newbury Park, CA: Sage.
Bryk, A. S. Raudenbush, S. W. & Congdon, R. J. (1996). Hierarchical linear and nonlinear modeling with the HLM/2L and HLM/3L programs. Chicago: Scientific Software.
Burstein, L. (1980). The analysis of multilevel data in educational research and evaluation. Review of Research in Education, 8: 158-233.
Cappelli, P. & Sherer, P. D. (1991). The missing role of context in OB: The need for a meso-level approach. Research in organizational behavior, 13: 55-110.
Carron, A.V. & Spink, K. S. (1995). The group size-cohesion relationship in minimal groups. Small Group Research, 26: 86-105.
Cartwright, D. (1968). The nature of group cohesiveness (pp 91-109.), in D. Cartwright & A. Zander (Eds.), Group dynamics: Research and theory (3rd ed.) New York: Harper & Row.
Cattell, R. B. (1948). Concepts and methods in the measurement of group syntality. Psychological Review, 55: 48-63.
Chalykoff, J. & Kochan, T. A. (1989). Computer-aided monitoring: Its influence on employee job satisfaction and turnover. Personnel Psychology, 42: 807-834.
de Leeuw, J. & Krefi, I. G. G. (1995). Questioning multilevel models. Journal of Educational and Behavioral Statistics, 20: 171-189.
Dobbins, G. H. & Zaccaro, S. J. (1986). The effects of group cohesion and leader behavior on subordinate satisfaction. Group & Organization Studies, 11: 203-219.
Draper, D. (1995). Inference and hierarchical modeling in the social sciences. Journal of Educational and Behavioral Statistics, 20:115-147.
George, J. M. & Bettenhausen, K. (1990). Understanding prosocial behavior, sales performance and turnover: A group-level analysis in a service context. Journal of Applied Psychology, 75: 698-709.
Gross, E. (1954). Primary functions of the small group. American Journal of Sociology, 60: 24-30.
Griffin, M. A. & Hofmann, D. A. Unpublished Manuscript. Hierarchical linear models in organizational research.
Gully, S. M., Devine, D. J. & Whitney, D. J. (1995). A meta-analysis of cohesion and performance: Effects of level of analysis and task interdependence. Small Group Research, 26: 497-520.
Hackman, J. R. (1992). Group influences on individuals in organizations (pp. 199-267), in M.D. Dunnette & L. M. Hough (Eds.), Handbook of Industrial and Organizational Psychology (2nd ed., Vol. 3) Palo Alto, CA: Consulting Psychologists Press.
Hare, A. P. (1981). Group size. American Behavioral Scientist, 24: 695-708.
Hofmann, D. A. 1997. An Review of the logic and rationale of HLM. Journal of Management, 23(6): 723-742.
Hofmann, D. A., Jacobs R. & Baratta, J. E. (1993). Dynamic criteria and the measurement of change. Journal of Applied Psychology, 78: (194-204).
House, R., Rousseau, D. M. & Thomas-Hunt, M. (1995). The meso paradigm: A framework for the integration of micro and macro organizational behavior. Research in Organizational Behavior, 17:71-114.
Isen, A.M. & Baron, R. A. (1991). Positive affect as a factor in organizational behavior. Research in Organizational Behavior, 13: 1-54.
James, L. R. (1982). Aggregation bias in estimates of perceptual agreement. Journal of Applied Psychology, 67: 219-229.
James, L. R., Demaree, R. G. & Wolf, G. (1984). Estimating within-group interrater reliability with and without response bias. Journal of Applied Psychology, 69: 85-98.
James, L. R., Demaree, R. G. & Wolf, G. (1993). r.sub.wg: An assessment of within-group interrater agreement. Journal of Applied Psychology, 78: 306-309.
Kerr, J. & Slocum, J. W. (1987). Managing corporate culture through reward systems. Academy of Management Executive, 1: 99-108.
Klein, K. J. Dansereau, F. & Hall, R. J. (1994). Levels issues in theory development, data collection, and analysis. Academy of Management Review, 19: 195-229.
Konovsky, M. A. & Pugh, S. D. (1994). Citizenship behavior and social exchange. Academy of Management Journal, 37: 656-669.
Kozlowski, S. W. J. & Hattrup, K. (1992). A disagreement about within-group agreement: Disentangling issues of consistency versus consensus. Journal of Applied Psychology, 77: 161-167.
Kreft, 1. G. G. (1996). Are multilevel techniques necessary? An overview, including simulation studies. Manuscript in preparation.
Marquis, D. G., Guetzkow, H. & Heyns, R. W. (1951). A social psychological study of the decision-making conference, (pp. 55-67), in H. Guetzkow (Ed.), Groups, leadership, and men, Pittsburgh, PA: Carnegie Press.
Morrison, E. W. (1994). Role definitions and organizational citizenship behavior: The importance of the employee’s perspective. Academy of Management Journal, 37: 1543-1567.
Mossholder, K. W. & Bedeian, A. G. (1983). Cross-level inference and organizational research: Perspectives on interpretation and application. Academy of Management Review, 8: 547-558.
Mowday, R. T. & Sutton, R. I. (1993). Organizational behavior: Linking individuals and groups to organizational contexts. Annual Review of Psychology, 44: 195-229.
Mudrack, P. E. (1989). Group cohesiveness and productivity: A closer look. Human Relations, 42:771-785.
O’Reilly, C. & Chatman, J. (1986). Organizational commitment and psychological attachment: The effects of compliance, identification and internalization on prosocial behavior. Journal of Applied Psychology, 71: 492-499.
Organ, D. W. (1990). The motivational basis of organizational citizenship behavior. Research in organizational behavior, 12: 43-72.
Ostroff, C. (1992). The relationship between satisfaction, attitudes and performance: An organizational level analysis. Journal of Applied Psychology, 77: 963-974.
Podsakoff, P.M., MacKenzie, S. B., Moorman, R. H. & Fetter, R. (1990). Transformational leader behaviors and their effects on followers’ trust in leader, satisfaction and organizational citizenship behaviors. Leadership Quarterly, 1: 107-142.
Podsakoff, P.M., MacKenzie, S. B. & Bommer, W. H. (1996). Transformational leader behaviors and substitutes for leadership as determinants of employee satisfaction, commitment, trust, and organizational citizenship behaviors. Journal of Management, 22: 259-298.
Puffer, S. M. (1987). Prosocial behavior, noncompliant behavior and work performance among commission salespeople. Journal of Applied Psychology, 72:615-621.
Raudenbush, S. W. (1989a). “Centering” predictors in multilevel analysis: Choices and consequences. Multilevel Modeling Newsletter, 1(2): 10-12.
Raudenbush, S. W. (1989b). A response to Longford and Piewis. Multilevel Modeling Newsletter, 1(3): 8-10.
Schachter, S., Ellertson, J., McBride, D. & Gregory, D. (1951). An experimental study of cohesiveness and productivity. Human Relations, 4: 229-238.
Schnake, M. (1991). Organizational citizenship: A review, proposed model and research agenda. Human Relations, 44: 735-759.
Schriesheim, J. F. (1980). The social context of leader-subordinate relations: An investigation of the effects of group cohesion. Journal of Applied Psychology, 65:183-194.
Settoon, R. P. & Mossholder, K. W. (1996). Interpersonal citizenship behavior.’ A mid-range theory and model. Paper presented at the 56th Annual Meeting of the Academy of Management, Cincinnati, OH.
Shamir, B. (1990). Calculations, values, and identities: The sources of collectivistic work motivation. Human Relations, 43:313-332.
Smith, C. A., Organ, D. W. & Near, J.P. (1983). Organizational citizenship behavior: Its nature and antecedents. Journal of Applied Psychology, 68: 653-663.
Stroebe, W. & Frey, B. S. (1982). Self-interest and collective action: The economics and psychology of public goods. British Journal of Social Psychology, 21: 121-137.
Vancouver, J. B., Millsap, R. E. & Peters, P. A. (1994). Multilevel analysis of organizational goal congruence. Journal of Applied Psychology, 79: 666-679.
Van Dyne, L., Cummings, L. L. & Parks, J. M. (1995). Extra-role behaviors: In pursuit of construct and definitional clarity (a bridge over muddled waters). Research in Organizational Behavior, 17: 215-285.
Van Dyne, L., Graham, J. W. & Dienesch, R. M. (1994). Organizational citizenship behavior- Construct redefinition, measurement, and validation. Academy of Management Journal, 37: 765-802.
Widmeyer, W. N., Brawley, W. N. & Carcon, A. V. (1990). The effects of group size in sports. Journal of Sport and Exercise Psychology, 12:177-190.
Williams, L. J. & Anderson, S. E. (1991). Job satisfaction and organizational commitment as predictors of organizational citizenship and in-role behaviors. Journal of Management, 17: 601-617.
COPYRIGHT 1997 JAI Press, Inc.
COPYRIGHT 2004 Gale Group