Mediating and moderating effects in job design
Gary Johns
The dominant paradigm in job design research, obtaining self-reports of job characteristics from job incumbents, seems to have weathered the storm provoked by Salancik and Pfeffer (1978). They argued that research concerning incumbent reactions to self-reported job characteristics was susceptible to priming effects, common method variance, and the influence of extraneous social cues.
A literature review by Fried and Ferris (1987) concludes that the evidence for priming effects is nil. Although common method variance might inflate relationships between self-reported job characteristics and worker responses (Roberts & Glick, 1981), it is now clear that the importance of this possibility has been exaggerated. Wagner and Crampton (1990) conducted a meta-analysis that compared “percept-percept” correlations with “multisource” correlations in various areas of organizational behavior, assuming that higher percept-percept associations would be indicative of method variance. In this analysis, job design emerged as one of the least tainted research areas. Finally, regarding extraneous social cues, all well designed studies in this domain (e.g., Griffin, 1983: Griffin, Bateman, Wayne, & Head, 1987) show strong and consistent effects for objective variations in job design in spite of the manipulation of social information.
The most influential model guiding self-report research on job characteristics has been the Job Characteristics Model (JCM, Hackman & Oldham, 1976, 1980). The JCM attempts to specify those job characteristics that lead to favorable work outcomes such as internal motivation, job satisfaction, and effective work behavior (good performance and low absence and turnover). It is expected that these outcomes will occur when workers experience three critical psychological states — meaningful work, personal responsibility, and knowledge of results. In turn, five “core” job characteristics (skill variety, task identity, task significance, autonomy, and feedback) are posited to stimulate particular psychological states. Workers who exhibit high growth need strength, adequate knowledge and skill, and satisfaction with job context factors are expected to respond best to rich or high scope jobs. Measures of most JCM variables are provided in Hackman and Oldham’s (1975, 1980) Job Diagnostic Survey. Although this instrument has some psychometric limitations, it has proven useful in job design research (Taber & Taylor, 1990).
The JCM has stimulated a large body of fragmented research. This work has been very ably summarized in a qualitative and quantitative review by Fried and Ferris (1987). The major purpose of the present study was to address a number of gaps in JCM research identified by Fried and Ferris, who generally concluded that the “JCM has received modest support” (1987:309).
One of the most critical gaps in JCM research involves how infrequently the total model has been tested. Although many studies have factor analyzed the core characteristics or correlated them with outcome variables, the rarity of studies that incorporate the mediating psychological states is remarkable. Hackman and Oldham (1976:255) have described the psychological states as “the causal core of the model,” and the job characteristics were identified to service the states, not the other way around.
Of the previous tests of the complete model, Hackman and Oldham (1976) employed the largest sample, 658 individuals. However, the statistical procedures they used were not always elaborate enough to match the model’s complexity. Wall et al. (1978) and Arnold and House (1980) used more elaborate analyses, but these studies both used small samples (47 and 90 respectively), a potential problem when a complex model with many variables is tested. Griffeth (1985) examined all variables in a field experiment with 57 subjects. Finally, Hogan and Martell (1987) used a sample of 208 and structural equations analysis to test competing versions of the JCM. In all of these studies it is unclear whether the data had a factor structure for the core characteristics that corresponded to that prescribed by the JCM. Fried and Ferris (1987) found that about half of the studies that factor analyzed Job Diagnostic Survey data failed to support a five-factor solution. In the present study, the five-factor solution was appropriate, resulting in a clear and fair test of the JCM.
Drawing on Hackman and Oldham (1976), Fried and Ferris (1987) note the essential questions to be answered in incorporating the psychological states into tests of the JCM:
1. Do the states correspond most closely to the core characteristics that the model specifies will influence them?
2. Do the states occupy a formal mediating role between the core characteristics and outcome variables?
3. Are all three states required to maximize the prediction of outcomes?
Besides mediation, the other critical process specified by the JCM is
moderation. Hackman and Oldham are explicit that this moderation is a dual location process, occurring both between the core characteristics and the psychological states and the states and the outcome variables. This contention has almost never been tested. Rather, most researchers have restricted moderator tests to the relationship between job characteristics and outcomes or overall job scope and outcomes. However, the exact location (or relative strength) of such moderator effects is important for theory development. Do inappropriate levels of the moderators preclude the development of favorable psychological states or do they simply preempt favorable outcomes?
Three meta-analyses of the moderating effects of growth need strength on the relationship between job characteristics and various outcome variables reveal inconsistent results (Fried & Ferns, 1987: Loher, Noe, Moeller, & Fitzgerald, 1985; Spector, 1985). Although each analysis finds some support for moderation, there is disagreement about just which outcome variables are susceptible to this effect. It should be noted that much of this research uses subgroup correlational analysis, not the most appropriate technique for testing interactions (Stone & Hollenbeck, 1989). The present study sought to clarify the moderating effects of growth need strength by exploring its operation at several logical points in the JCM and by using appropriate statistical techniques. In addition, the role of several other seldom examined potential moderators was explored. From the point of view of ease of diagnosis in advance of a job design intervention, context satisfaction might have more utility than growth need strength. Despite this, the few tests of context satisfaction moderation are contradictory. Oldham, Hackman, and Pearce (1976) found support for the moderating effects of context satisfaction, but Katerberg, Hom, and Hulin (1979) generally did not, except for some effects for pay satisfaction.
It appears that the moderating effect of employee skills and knowledge on the receptiveness to a high scope job has never been tested directly. However, a good case can be made that educational level and tenure in the job might be reasonable surrogates for skills and knowledge. The predicted effect for education is straightforward: those with more education would seem better able to cope with high scope jobs and should respond more favorably to those jobs (and their related psychological states) than less educated employees. However, Aldag and Brief (1975) found higher correlations between task characteristics and affective responses among less educated employees.
The case for job tenure as a JCM moderator is even more complicated. Katz (1978a, 1978b) has suggested a three-stage model in which those with intermediate job tenure are most receptive to the touted benefits of high scope jobs. Those new to such jobs and those with long experience are expected to be less likely to respond favorably. According to Katz, newcomers to a job are less concerned with job content and more preoccupied with mastering the job’s social context and technical requirements. On the other hand, employees with considerable job experience may “adapt out” to the challenges of the job. Also, lengthy experience could be accompanied by a change in career stage and a concomitant change in need structure. Using an occupationally diverse sample of public sector employees, Katz presents cross-sectional subgroup results that offer some support for this position. However, he also presents moderated regression results that show linear moderator effects of job tenure on the relationship between job characteristics and outcomes. Clearly, the stage model suggests a curvilinear interaction that Katz did not test for. In a related vein, using some of his own logic, we might predict nonlinear relationships between job tenure and some JCM outcome variables, quite independent of the job in question. For many jobs, for example, those with average amounts of tenure might find greatest growth satisfaction: they have mastered the job, but it has not grown stale. If this is the case, analyses must control for any nonlinear relationships between the moderator (job tenure) and outcome variables.
Kemp and Cook (1983) were not able to replicate Katz’s findings using two samples of blue collar workers. However, Kozlowski and Hults (1986) replicated Katz (using performance as the criterion) for research and development engineers but not staff engineers. All of this suggested a careful examination of the moderating role of job tenure in the current study.
Finally, for the managerial sample studied here, we speculated that the extent to which extrinsic rewards were perceived as being contingent upon performance would serve as a JCM moderator. However, the direction that such moderation might take was unclear given existing theory and research. On one hand, managers generally support a norm of tying extrinsic rewards to performance (Lawler, 1971). If such a norm is violated, the positive intrinsic outcomes of enriched work might be overridden by a concern for more concrete rewards. Aspects of this “need regression” might be predicted by Alderfer’s (1972) ERG theory. On the other hand, although field research in organizational settings has been far from universally supportive (Guzzo, 1979), a continuing line of research has shown that the ready provision of extrinsic rewards can damage intrinsic motivation (Deci & Ryan, 1985). This suggests that contingent extrinsic rewards might neutralize the intrinsic benefits of high scope jobs. To our knowledge, this apparent contradiction has never been examined in job design research.
Method
Subjects and Procedure
A random sample of 605 first and second level managers in a large utility company composed the potential respondents. The jobs themselves represented a wide variety of functions. A questionnaire was mailed to home addresses that included items about their job and some personal characteristics. Three hundred usable questionnaires were returned directly to the researchers’ university address. The tenure distribution of the respondents was similar to that of the nonrespondents. The average respondent was 34 years old, had 11.7 years of tenure with the company, and had been on his or her current job for 2.7 years. Sixty-four percent were men and 39% were college or university graduates. First-level managers composed 81% and line managers 68% of the sample.
Measures
The five core job dimensions, the three psychological states, and several outcome variables (general satisfaction, internal motivation, and growth satisfaction) were measured with the Job Diagnostic Survey (JDS, Hackman & Oldham, 1980). Overall job scope was expressed with Hackman and Oldham’s multiplicative Motivating Potential Score (MPS). Two additional outcome variables not covered by JDS scales were also measured (on 7-point scales). Turnover cognitions were assessed by averaging four questionnaire items that probed frequency of thoughts about quitting, probability of job search, probability of quitting, and the importance attached to remaining in the employ of the utility company. This measure correlated .22 (p |is less than~ .01) with actual turnover measured 28 months later. Also, a single item requested managers to rate their own performance over the past year. This measure correlated .25 (p |is less than~ .01) with single-item supervisory ratings taken from company records. Because both “hard” measures exhibited severe restriction of range (for example, only 19 of 300 managers quit) it was decided to employ the questionnaire measures in the analyses that follow.(1) Also, it was felt that self-rated performance better captured the intrinsic motivational advantages claimed for high scope jobs. Stone (1986) concluded that self-ratings of performance are a particularly useful outcome measure in job design research. Descriptive statistics for the core characteristics, states, and outcomes are presented in Table 1.
The following potential moderator variables were measured with JDS scales: Growth need strength (“job choice” format, |alpha~ = .74), security satisfaction, (|alpha~ = .77), pay satisfaction (|alpha~ = .93), social satisfaction (|alpha~ = .57), and supervisory satisfaction (|alpha~ = .90). Other questionnaire items tapped job tenure, educational level, and the extent to which pay (two items, |alpha~ = .57) and promotions (two items, |alpha~ = .73) were perceived to be instrumentally related to performance.
Results
Table 1 presents the means, standard deviations, alpha reliability coefficients, and intercorrelations for the three classes of variables in the basic (unmoderated) JCM. The standard deviations for the five core job characteristics average only .2 SD less than JDS normative data based on a wide variety of jobs (Oldham, Hackman, & Stepina, 1979). The median correlation among the core job characteristics is .32, and the relationships between the core characteristics and the outcome variables are similar to those that might be expected from the literature. To examine the empirical dimensionality of the items composing the core characteristics, we submitted them to a principal components factor analysis with oblique rotation. TABULAR DATA OMITTED This resulted in an extremely clean and appropriate five-factor solution. Thus, the basic job design data were conducive to a fair test of the JCM.
Many of the analyses to be performed involved self-report data from a single source. To explore the possibility of common method variance, a single-factor test was performed. The logic behind this analysis is attributed to Harman (1976) by Podsakoff and Organ (1986). In this test, the five job characteristics, the three psychological states, and the five outcome variables were factor analyzed simultaneously. Common method variance might be a problem if the unrotated factor solution reveals a single factor or a predominant general factor that accounts for substantial variance. Four factors emerged from this analysis. The first had an eigenvalue of 4.42 and accounted for 34% of the variance. The other three factors accounted for an additional 13% of variance, but had low eigenvalues, ranging between .51 and .61. Varimax and oblimin rotation produced similar four-factor solutions with low residuals. General satisfaction, growth satisfaction, turnover cognitions, and experienced meaningfulness loaded on the first factor. Variety, significance, and autonomy loaded on the second. Experienced responsibility and internal motivation loaded on the third. Feedback, knowledge of results, and task identity loaded on the fourth. (Results of all factor analyses are available from the first author).
The single-factor test results do raise concerns about the impact of method variance on some of the results to be reported. However, as Podsakoff and Organ (1986) note, it is somewhat difficult to know what to make of this test when the theory or model in question explicitly proposes true functional relationships among the variables, as is the case with the JCM. This is particularly the situation when correlations among outcome variables are expected, as occurs here. Also, although common method variance inflates bivariate relationships, it results in conservative tests of moderator effects. Spurious associations between the predictor and the criterion due to method leave less criterion variance for the joint effect to explain.
Correspondence Between Psychological States and Specified Job Characteristics
The JCM suggests that specified job characteristics should account for substantial variance in their corresponding psychological states and that, controlling for this, unspecified job characteristics should not account for substantial additional variance. For example, in a regression equation using variety, identity, and significance to successfully predict meaningfulness, the addition of autonomy and feedback should not have a substantial effect. This analysis is shown in the top part of Table 2. For meaningfulness and responsibility, the addition of the unspecified job characteristics to the respective equations results in statistically significant 5 and 6% increments in explained variance. Only knowledge of results seems to correspond uniquely to its specified core characteristic, feedback.
A slightly different perspective is provided in the lower half of Table 2, where standardized regression coefficients for the full five variable equations for each psychological state are given. The model-specified coefficients for each state are bracketed for clarity. Again, the clearest case of model-specified correspondence TABULAR DATA OMITTED is that between feedback and knowledge of results. The most anomalous case is for experienced meaningfulness, where the regression weight for the model-specified variable of task identity is exceeded by those for the unspecified variables of autonomy and feedback.
Mediating Role of Psychological States
The JCM proposes that higher job scope is translated into favorable outcomes via the psychological states. Wall et al. (1978:188) have set down rigorous criteria for testing this mediating effect:
(a) the critical psychological states should account for sizable proportions of variance in each of the dependent |outcome~ variables; (b) the core job dimensions should add little to this when considered in the same analysis; (c) the core job dimensions alone should account for relatively little of the dependent |outcome~ variable variance; and (d) the critical psychological states should add considerably to this when considered in the same analysis.
As Wall et al. explain, these criteria can be examined using three regression equations for each outcome variable, one using the three psychological states as predictors, the second using the five core characteristics, and the third using all eight predictors together. This analysis is shown in Table 3. If the Wall et al. criteria are taken in a comparative rather than absolute sense, there is good evidence that the psychological states mediate the relationship between the core characteristics and both general satisfaction and internal motivation. The weakest evidence for mediation occurs for the self-rating of performance. The most anomalous result is for growth satisfaction. Here, the job characteristics account for substantial variance in the outcome variable and make a noteworthy incremental contribution, even when controlling for the states. In fact, in the eight-variable regression equation with simultaneous entry, the standardized regression weights for skill variety and autonomy exceeded those for knowledge of results and responsibility.
TABULAR DATA OMITTED
A slightly different perspective was employed to conduct an alternative set of tests of the mediating function of the psychological states. In these tests, MPS was used as a summary measure of job design. In addition, a summary measure of “total psychological states” was constructed by summing the values of the three specific psychological states for each subject. Five tests were then performed to examine whether total psychological states mediated the relationship between MPS and each outcome. With only three variables in each model, it was possible, using regression, to employ Baron and Kenny’s (1986) criteria for mediation. Stated in causal language, these are that MPS should affect total states, MPS should affect the relevant outcome, and the apparent impact of MPS on the outcome should be greatly reduced when psychological states are accounted for. Using the Baron and Kenny criteria that involve both statistical significance levels and the relative size of regression weights, the results of these analyses were gratifyingly similar to those of the previous mediation analyses. As shown in Table 4, for the outcomes general satisfaction, internal motivation, turnover cognitions, and self-rated performance, the psychological states measure met Baron and Kenny’s criteria for “perfect mediation.” That is, the effect of MPS was completely removed when psychological states were controlled. Again, however, growth satisfaction received direct influence from MPS.
Path Analysis
In order to summarize and further clarify the previous analyses, a path analysis was conducted. Like Wall et al. (1978), two models were compared. One of these was the specified JCM, which predicts causal links between certain job characteristics and certain psychological states, links between all states and all outcomes, and no causal links between job characteristics and outcomes. In the present study, using five outcome variables this specified model is represented by 20 causal paths. The specified model was compared with a block recursive model that assumes causal links between all job characteristics and all psychological states, all states and all outcomes, and all job characteristics and all outcomes. This alternative model is represented by 55 causal paths. Standardized regression coefficients from the appropriate regression equations served as path coefficients.
The specified and alternative models were compared with a likelihood ratio statistic, which is distributed as ||chi~.sup.2~ (Land, 1973). This comparison resulted in ||chi~.sup.2~ = 51, df = 35, p |is less than~ .05, indicating that there are additional paths to those specified in the JCM. In inspecting the regression equations for the models, we noted the critical role of the chosen alpha level. With alpha set at .001, 12 of 20 specified paths were identified, whereas only 3 of 35 unspecified paths were identified. With alpha set at .05, identified specified paths totalled 15, whereas identified unspecified paths totalled 14. These results suggest that many of the unspecified paths, though statistically significant by conventional criteria, may not be “substantively meaningful” (Land, 1969) in the overall context of the JCM.
To clarify matters, the data were “trimmed” according to the procedure given by Land (1973). The resulting model is presented as a path diagram in Figure 1. Paths specified by the JCM are depicted with solid lines and parenthesized path coefficients. Substantial alternative paths are indicated by broken lines and/or unparenthesized path coefficients. All in all, the diagram indicates relatively good support for the JCM, with three basic anomalies. First, experienced meaningfulness appears to be a particularly encompassing psychological state, reflecting substantial unspecified associations with autonomy and feedback. Again, growth satisfaction is a particularly sensitive outcome, reflecting direct effects not only from all three states but from variety and autonomy as well. Finally, the self-rating of performance is not a particularly salient outcome variable, showing a connection only with responsibility.
The path analysis provides a good basis for answering Fried and Ferris’s (1987) question about the necessity for the presence of all three psychological states. Although all three states are useful in maximizing the prediction of growth satisfaction, such is not the case for the other four outcome variables. Furthermore, although no psychological state is a clear winner in playing a central role in the prediction of all outcomes, each state has its role as a significant predictor of some outcomes.
Moderator Effects
For sake of parsimony, some decisions had to be made concerning how to test the “three-point” moderator hypotheses. Because discussions of JCM moderator effects generally revolve around the whole job rather than its components, it was decided to perform these tests using the summary MPS to represent the job. On the other hand, the differentiated mediating effects of the three psychological states for the different outcome variables suggested treating them separately. However, it might be argued that an overall or summary measure of the psychological states is more in the spirit of the summary MPS measure. This dilemma was resolved as follows: Moderator effects were tested between MPS and the three states, between the states and the five outcomes, and between MPS and the five outcomes. In addition, the additive total psychological states measure described above was also employed.
Potential moderator effects were tested using hierarchical regression. The results are portrayed graphically in Figure 2, where a line between two variables designates a significant effect for the indicated moderator at p |is less than~ .05 or better. A large number of tests were performed here, and many of them were not independent. The resulting increase in the probability of type II errors (Cohen, 1982) means that the patterns of moderator effects are more indicative than the results of any particular test. A table showing exact variance increments and t values is available from the first author.
Personal characteristics. The moderator effects for the personal characteristics growth need strength, education, and job tenure are shown in the top portion of Figure 2. In the case of growth needs, only two significant interactions occurred, both with meaningfulness in predicting general satisfaction (|delta~|R.sup.2~ = .01, t = 2.07, p |is less than~ .05) and growth satisfaction (|delta~|R.sup.2~ = .02, t = 2.98, p |is less than~ .01). In each instance, those with high growth needs responded more favorably to experienced meaningfulness than those low in growth needs.
Based on the work of Katz (1978a, 1978b), relationships between job tenure and the outcome variables were examined for curvilinearity using hierarchical regression with a squared term. Three significant curvilinear relationships were detected, between job tenure and general satisfaction (|delta~|R.sup.2~ = .02, t = 2.29, p |is less than~ .05), job tenure and growth satisfaction (|delta~|R.sup.2~ = .02, t = 2.38. p |is less than~ .05), and job tenure and turnover cognitions (|delta~|R.sup.2~ = .07, t = 5.16, p |is less than~ .001). The two satisfaction relationships were shallow inverted u’s, indicating that those new to the job and those with extensive job experience tended to be slightly less satisfied. The relationship between job tenure and turnover cognitions was u-shaped, with employees with low tenure revealing the most active thoughts about turnover, followed by those with the most extensive job tenure. All the moderator tests for job tenure that involved growth satisfaction, general satisfaction, or turnover cognitions controlled for curvilinearity. As shown in Figure 2, these tests resulted in three significant linear interactions, all between tenure and knowledge of results in predicting general satisfaction, growth satisfaction, and turnover cognitions. Employees who were higher in job tenure responded especially favorably to knowing the outcomes of their work. The t values ranged from 2.26 to 2.73, with variance increments around 2%. Where appropriate, tests were also run to assess the possibility of curvilinear interactions involving job tenure. None of these tests revealed any significant effects.
Of the three personal characteristics, educational level exhibited the strongest and most pervasive effects. Education interacted with three of the four state measures in predicting several outcomes exclusive of internal motivation and performance. However, the direction of these moderating effects was consistently counter to expectations in that it was the less educated managers who were most sensitive to the psychological states, responding more favorably to enhanced responsibility and knowledge of results and less favorably to decrements in these states. Variance increments for these effects averaged about 2% and t’s ranged from 2.03 to 3.60.
Context satisfaction. The results for context satisfaction moderation are shown in the middle portion of Figure 2. In contrast to the personal characteristics moderators, the largest proportion of significant context satisfaction moderators operate in the “front” part of the model, affecting the relationship between MPS and the psychological states. Forty-four percent of the tests performed here indicated significant moderation. Significant effects in the “back” part of the model tended to center on the more behaviorally-oriented outcomes, turnover cognitions and self-rated performance. This result might be a partial function of the tendency for context satisfaction to be most highly correlated with the more affective outcome variables of general and growth satisfaction, thus reducing the opportunity for moderation. However, turnover cognitions were also fairly well correlated with context satisfaction.
The results for security and social satisfaction ran counter to theoretical predictions, in that there was some tendency for managers who were lower on these facets to report that higher scope jobs induced more positive psychological states. Also, the relationship between experienced meaningfulness and self-rated performance was most negative among those who reported the most social satisfaction. The variance increments for these tests were in the 1-2% range, and the t’s ranged from 1.97 to 2.90.
Significant moderator effects for pay satisfaction were confined to the relationship between the psychological states and turnover cognitions. As predicted by the theory underlying the JCM, those who were most satisfied with pay thought less about quitting when they were experiencing favorable job-related psychological states. Turnover cognitions were essentially unrelated to psychological states among those who were less satisfied with their pay. Again, variance increments were in the 1-2% range, with t’s ranging from 2.02 to 2.42.
The moderator effects for supervisory satisfaction were the strongest, the most pervasive, and the most complicated in pattern. In essence, supervisory satisfaction negatively moderated the relationship between job scope and the psychological states, again running counter to JCM theory. That is, managers who reported less satisfaction with supervision were more likely to describe high MPS jobs as inducing positive psychological states. A similar moderator effect was observed for the relationship between MPS and the outcome variable of general satisfaction. Variance increments for these interactions averaged 2%. and t’s ranged from 2.36 to 3.95. On the other hand, an opposite moderating effect existed between the psychological states and performance (and, in one instance, turnover cognitions). That is, managers who were most satisfied with their bosses were most likely to rate their own performance highly when they reported favorable psychological states. Again, variance increments averaged 2%, with t’s ranging from 2.05 to 2.83.
Instrumentalities. The moderator effects for pay and promotion instrumentality are shown in the bottom portion of Figure 2. For pay instrumentality, significant effects can be observed in both the “front” and the “back” parts of the JCM. All of these moderator effects except those involving self-ratings of performance are negative effects. That is, lower perceived connections between pay and performance result in more positive relationships between MPS and the states and between the states and the outcomes. However, in both cases where pay instrumentality interacts with psychological states to predict performance the moderator effect is the opposite. That is, managers tended to respond to favorable psychological states with enhanced performance if pay was seen to be contingent on performance.
Significant moderator effects for promotion instrumentality were confined almost exclusively to the links between the psychological states and outcome variables. Increased knowledge of results boosted general and growth satisfaction and reduced turnover cognitions when promotions were not perceived to be related to performance. When promotions were seen as performance-contingent, meaningful and responsible work were both positively related to self-ratings of performance. The last interaction was the strongest in the study in terms of variance explained (|delta~|R.sup.2~ = .05, t = 3.88, p |is less than~ .001). Otherwise, variance increments for significant instrumentality interactions averaged about 2.5%, with t’s ranging from 1.95 to 3.38.
Discussion
Returning to Fried and Ferris’s (1987) questions regarding the role of the psychological states, the following conclusions are supported. First, the psychological state knowledge of results is well tied to its corresponding job characteristic, feedback, and the psychological state of experienced responsibility is reasonably well tied to its corresponding job characteristic, autonomy. However, meaningfulness appears to receive some unique contribution from each of the five core job dimensions. It is possible that the strong similarity in the wording of the JDS measures of feedback and knowledge of results is responsible for this being the best supported model-specified connection. Second, the psychological states generally do occupy a formal mediating role between the job characteristics and the various outcome variables. Knowing the psychological impact of various job dimensions often enables us to do a better job of predicting the attitudinal and quasi-behavioral outcomes of those dimensions. One anomaly is growth satisfaction, which appeared to be directly responsive to several job dimensions. Third, for a given outcome variable, all three psychological states arc seldom necessary to maximize prediction. However, given all the outcomes, each state has its place, and none is so inferior as a predictor to warrant exclusion from the model.
Regarding the connection between the core characteristics and the psychological states, Hackman and Oldham (1976) reported results that are parallel but not identical to those seen here. Like Griffeth (1985) and us, they found the most convincing connection to be that between feedback and knowledge of results. However, their “problem” state was responsibility, which was influenced by several of the job dimensions besides autonomy. Although both studies generally support mediation, both also observed unmediated effects of job characteristics on the outcome variable of growth satisfaction. Finally, regarding the utility of all three psychological states, our results are fairly similar to theirs. Wall et al. (1978) reanalyzed Hackman and Oldham’s data and also tested the model on a small sample of factory workers, concluding that the former “results were replicated in general and in detail by the present investigation” (194). This assertion indicates that the psychological states do play a central and useful role in the model, but that the data violate the model in many specific details. We agree with this conclusion. Finally, Hogan & Martell (1987) found reasonable support for the basic JCM, including the mediating role of the states. However, they also found that a single-factor model fit the data well. This corresponds to the concerns raised by our own single-factor test.
The results of all of these studies suggest that researchers and practitioners who are interested in the impact of jobs on employees might consider measuring psychological states more often than is commonly done. Not only are they predictive of important work outcomes, but it is possible that they are responsive to a range of other job characteristics besides those contained in the JDS. Task meaningfulness, for example, might be a product of the ideological context of the job, independent of the other characteristics that are thought to contribute to meaningfulness. Similarly, physical conditions and the presence of status symbols surrounding the job might contribute to perceptions of responsibility over and above opportunity for autonomy. These arguments are in line with recent approaches to job characteristics that extend beyond the JDS core dimensions (Campion & Thayer, 1985; Zaccaro & Stone, 1988). However, method variance concerns suggest that it might be wise to avoid measuring both job characteristics and psychological states or states and outcomes with the same self-report instrument.
This study found very little evidence for the moderating role of growth need strength, with two significant effects occurring between one of the psychological states and outcomes. Using samples similar in occupational level, Arnold and House (1980) found evidence of moderation in the front end of the model only (between job characteristics and states), and Hogan and Martell (1987) found no evidence at any of the linkages. It is possible that these exclusive white-collar samples do not provide the best basis for testing growth need moderation.
Some serious thought should be given to the appropriate sampling strategy for studying the relationship between job design and job tenure. Neither the present study (managerial sample) nor Kemp and Cook (1983) (blue collar samples) were able to replicate Katz (1978a, 1978b), who used an occupationally diverse sample of public sector employees, Kozlowski and Halts (1986) replicated Katz for research engineers but not staff engineers. If there is a pattern to these results, we fail to discern it. In theory, the most interesting and informative sample for studying the interaction between job tenure and job design might sensibly include jobs (a) of a similar occupational level, (b) with a substantial learning curve, (c) that exhibit adequate variance in core characteristics. In addition, the range and variance of job tenure should be great. These characteristics describe the current sample precisely. Significant interactions were observed, however, only between job tenure and knowledge of results in predicting three outcomes. Importantly, simple curvilinear relationships were observed between job tenure and three outcome variables. In future work, such relationships should be controlled.
It is possible that the preponderance of nonsignificant moderator effects for growth need strength and tenure could be due to the presence of method variance, a product of the conservatism of the moderator tests alluded to in the Results section. This is especially true when variables are measured with less than perfect reliability and exhibit correlated error (Evans, 1991). However, in the discussion to follow, a number of significant effects that run counter to JCM theory are reviewed. Such findings suggest that method variance did not pose a serious threat to the detection of moderators in this study and that the theory itself might have some problems.
Among the personal characteristics, the most interesting results occurred for educational level, which moderated the relationship between several measures of psychological states and outcomes. However, contrary to expectations, those with less education responded more favorably to elevated psychological states. It is possible that this result is a function of an expectations mechanism. In the utility company, it was possible to find one’s niche in a lower level (but still high scope) management position without being a university graduate. This ostensible “overemployment” might have exceeded the expectations of less educated employees, whereas better educated managers, still in lower level management, were less enthusiastic about the experience of enhanced responsibility. These arguments suggest that educational level operated here more as an index of needs and expectations than as a simple surrogate for skills and knowledge.
Hackman and Oldham (1976) contend that high satisfaction with job context factors should facilitate the positive effects of job scope on psychological states and the impact of the states on outcomes. There was some evidence for this claim. For example, enhanced psychological states were associated with reduced turnover cognitions among those who were most satisfied with pay. Also, elevated states were associated with enhanced perceptions of one’s own performance when satisfaction with supervision was high. But what induces positive psychological states? There was a tendency for those who were less satisfied with security, supervision, and social factors to respond more favorably to high-MPS jobs. It is tempting to posit a model in which context dissatisfaction enhances the psychological consequences of a high scope job via a compensation mechanism while having a much less beneficial effect on the conversion of these consequences to high performance and low turnover cognitions.
In general, high MPS was associated with enhanced psychological states and outcomes, and elevated states were associated with enhanced outcomes, when instrumentalities were low: that is, when managers perceived that pay and promotions were not contingent on performance. An important exception to this trend was the relationship between the psychological states and self-rated performance. Favorable states were most likely reflected in high performance among those who perceived stronger links between pay or promotions and performance. Perhaps tying extrinsic rewards to performance blunts the psychic rewards of good job design but boosts self-efficacy for performance.
On the surface, these results would seem to simultaneously support and contradict Deci and Ryan’s (1985) cognitive evaluation theory regarding the impact of contingent rewards. Deci and Ryan would almost certainly agree that high MPS jobs and enhanced psychological states would stimulate intrinsic motivation, especially via autonomy and self-control mechanisms. In fact, this finding is illustrated in Table 1 by significant correlations with the JDS variable internal motivation. Summarizing the implications of the theory for work behavior, Deci and Ryan (1985:310) state:
Whenever rewards are used to motivate people — in other words, to control them — it is probable that they will have a negative effect on the people’s intrinsic motivation. Competitively-contingent rewards were said to be the most detrimental. However, rewards that are appropriately linked to performance, representing positive feedback in an informational context, ought not to be detrimental.
The large bureaucratic organization in which the current research was conducted had in place systems to tie pay and promotions to performance. The general culture was far from exhibiting cutthroat competition. As a result, there was substantial variance in the perceptions of the extent to which performance was instrumental in achieving rewards. This “eye of the beholder” effect, which is probably typical of many large organizations, might reflect personality or experience differences that moderate the connection between job variables and various outcomes. This situation can be contrasted with the many experimental tests that form the bulk of support for cognitive evaluation theory, in which reward contingencies are generally manipulated unambiguously and dependent responses are restricted to free-choice performance of a stimulating task. Interestingly, in the present study, the raw correlations between the two forms to instrumentality and internal motivation were both identical in size and positive (r = .11, p = .03), rather than negative, as Deci and Ryan (1985:48) might expect to see in field data.
In conclusion, we feel that the results provide relatively good support for the basic (unmoderated) JCM, and especially for the mediating role of the psychological states. Thus, the states might be used as more sensitive criteria in job design efforts and as preliminary criteria in studies of job characteristics other than those specified by the JCM. We emphasize, though, that it would be wise to separate methodologically the measurement of the states from the measurement of job design and/or outcomes. Our results for the less-investigated moderators (education, context satisfaction, and reward instrumentalities) are interesting both because many go against JCM predictions and because of their practical implications. Many jobs are “designed” via technological constraints, but organizations have more degrees of freedom in terms of who staffs those jobs, the surrounding context, and the reward systems put in place.
The authors gratefully acknowledge the support of FCAR Quebec grant number 91ER0506.
1 For exploratory purposes, the analyses for mediation and moderation to be reported in Tables 3 and 4 and Figure 2 were conducted using actual performance and turnover. These analyses yielded only one significant moderator effect.
References
Aldag, R.J., & Brief, A.P. 1975. Impact of individual differences on employee affective responses to task characteristics. Journal of Business Research, 3:311-322.
Alderfer, C.P. 1972. Existence, relatedness, and growth: Human needs in organizational settings. New York: The Free Press.
Arnold, H.J., & House, R.J. 1980. Methodical and substantive extensions to the job characteristics model of motivation. Organizational Behavior and Human Performance, 25:161-183.
Baron, R.M., & Kenny, D.A. 1986. The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51:1173-1182.
Campion, M.A., & Thayer, P.W. 1985. Development and field evaluation of an interdisciplinary measure of job design. Journal of Applied Psychology, 70:29-43.
Cohen, P. 1982. To be or not to be: Control and balancing of Type I and Type II errors. Evaluation and Program Planning, 5:247-253.
Deci, E.L., & Ryan, R.M. 1985. Intrinsic motivation and self-determination in human behavior. New York: Plenum.
Evans, M.G. 1991. On the use of moderated regression. Canadian Psychology, 32:116-119.
Fried, Y., & Ferns, G.R. 1987. The validity of the job characteristics model: A review and meta-analysis. Personnel Psychology, 40:287-322.
Griffeth, R.W. 1985. Moderation of the effects of job enrichment by participation: A longitudinal field experiment. Organizational Behavior and Human Decision Processes, 35:73-93.
Griffin, R.W. 1983. Objective and social sources of information in task redesign: A field experiment. Administrative Science Quarterly, 28:184-200.
Griffin, R.W., Bateman, T.S., Wayne, S.J., & Head, T.C. 1987. Objective and social factors as determinants of task perceptions and responses: An integrated perspective and empirical investigation. Academy of Management Journal, 30:501-523.
Guzzo, R.A. 1979, Types of rewards, cognitions, and work motivation. Academy of Management Review, 4:75-86.
Hackman, J.R., & Oldham, G.R. 1975. Development of the Job Diagnostic Survey. Journal of Applied Psychology, 60:159-170.
Hackman, J.R., & Oldham, G.R. 1976. Motivation through the design of work: Test of a theory. Organizational Behavior and Human Performance, 16:250-279.
Hackman, J.R., & Oldham, G.R. 1980. Work redesign. Reading, MA: Addison-Wesley.
Harman, H.H. 1976. Modern factor analysis (3rd ed.). Chicago: University of Chicago Press.
Hogan, E.A., & Martell, D.A. 1987. A confirmatory structural equations analysis of the job characteristics model. Organizational Behavior and Human Decision Processes, 39:242-263.
Katerberg, R., Jr., Hom, P.W., & Hulin, C.L. 1979. Effects of job complexity on reactions of part-time employees. Organizational Behavior and Human Performance, 24:317-332.
Katz, R. 1978a. Job longevity as a situational factor in job satisfaction. Administrative Science Quarterly, 23:204-223.
Katz, R. 1978b. The influence of job longevity on employee reactions to task characteristics. Human Relations, 31:703-725.
Kemp, N.J., & Cook, J.D. 1983. Job longevity and growth need strength as joint moderators of the task design-job satisfaction relationship. Human Relations, 36:883-898.
Kozlowski, S.W.J., & Hults, B.M. 1986. Joint moderation of the relation between task complexity and job performance for engineers. Journal of Applied Psychology, 71:196-202.
Land, K.C. 1969. Principles of path analysis. In E.F. Borgata (Ed.), Sociological methodology: 3-37. San Francisco: Jossey-Bass.
Land, K.C. 1973. Identification, parameter estimation, and hypothesis testing in recursive sociological models. In A.S. Goldberger & O.D. Duncan (Eds.), Structural equation models in the social sciences: 19-49. New York: Seminar Press.
Lawler, E.E., III 1971. Pay and organizational effectiveness: A psychological view. New York: McGraw-Hill.
Loher, B.T., Noe, R.A., Moeller, N.L., & Fitzgerald, M.P. 1985. A meta-analysis of the relation of job characteristics to job satisfaction. Journal of Applied Psychology, 70:280-289.
Oldham, G.R., Hackman, J.R., & Pearce, J.L. 1976. Conditions under which employees respond favorably to enriched work. Journal of Applied Psychology, 61:395-403.
Oldham, G.R., Hackman, J.R., & Stepina, L.P. 1979. Norms for the Job Diagnostic Survey. JSAS Catalog of Selected Documents in Psychology, 9:14. (Ms. No. 1819).
Podaskoff, P.M., & Organ, D.W. 1986. Self-reports in organizational research: Problem and prospects. Journal of Management, 12:531-544.
Roberts, K.H., & Glick, W. 1981. The job characteristics approach to task design: A critical review. Journal of Applied Psychology, 66:193-217.
Salancik, G.R., & Pfeffer, J. 1978. A social information processing approach to job attitudes and task design. Administrative Science Quarterly, 23:224-253.
Spector, P.E. 1985. Higher-order need strength as a moderator of the job scope-employee outcome relationship: A meta-analysis. Journal of Occupational Psychology, 58:119-127.
Stone, E.F. 1986. Job scope-job satisfaction and job scope-job performance relationships. In E.A. Locke (Ed.), Generalizing from laboratory to field settings: 189-206. Lexington MA: Lexington Books.
Stone, E.F., & Hollenbeck, J.R. 1989. Clarifying some controversial issues surrounding statistical procedures for detecting moderator variables: Empirical evidence and related matters. Journal of Applied Psychology, 74:3-10.
Taber, T.D., & Taylor, E. 1990. A review and evaluation of the psychometric properties of the Job Diagnostic Survey. Personnel Psychology, 43:467-500.
Wagner, J.A., III, & Crampton, S.M. 1990. Percept-percept methods in psychological research on organizations: A meta-analytic investigation of artifactual effects. In L.R. Jauch & J.L. Wall (Eds.), Academy of Management Best Papers Proceedings 1990:400-404.
Wall, T.D., Clegg, C.W., & Jackson, P.R., 1978. An evaluation of the job characteristics model. Journal of Occupational Psychology, 51:183-196.
Zaccaro, S.J., & Stone, E.F. 1988. Incremental validity of an empirically based measure of job characteristics. Journal of Applied Psychology, 73:245-252.
COPYRIGHT 1992 JAI Press, Inc.
COPYRIGHT 2004 Gale Group
