Vous êtes sur la page 1sur 9

Journal of Applied ftychology 1986, Vol. 71. No.

1, I02-HO

Copyright 1986 by the American Psychological Association, Inc. Q02I-90K>/86/$W.75

Role of Social Support in the Experience of Stress at Work


Daniel C. Ganster
Department of Management University of NebraskaLincoln

Marcelline R. Fusilier
Department of Management University of NebraskaOmaha

Bronston T. Mayes
Department of Management California State UniversityFullerton

It has been hypothesized that the positive relation between stress and strain responses is stronger for individuals who have low levels of social support than for those who have high levels of support. This hypothesis that social support buffers (moderates) the negative effects of stress has been tested extensively in a variety of setting with highly conflicting results. Some theorists have recently proposed that the moderating effect of social support is itself buffered by other variables such as sex or social class. The present study was designed to examine the role of social support in the experience of work stress with a sample large enough to provide statistically powerful tests of models of social support that specify two-way and three-way interactions. No support for higher order interactive models was found. In addition, no evidence emerged demonstrating any buffering effect for social support. Arguments are advanced for a parsimonious model in which social support has a modest direct effect of lowering experienced strain.

Researchers in the social sciences have sought to identify factors reducing or eliminating negative effects of stress in the work setting (Ganster, Mayes, Sime, & Tharp, 1982). The primary social factor hypothesized to mitigate these effects, or strains, is the degree of social support that an individual receives. The nature of the effect of social support on strains, however, is presently unclear (see Gore, 1981;House, 198 l;Kessler, Price, &Wortman, 1985;Leavy, l983;Thoits, 1982; for reviews). This lack of clarity has implications for the development and refinement of models of the phenomenon of stress as well as for managers and counselors seeking methods for coping with the problem. Until the role of social support in work stress is identified, its potential benefits cannot be fully used. The present study, therefore, attempts to delineate and investigate the relations among stressors, strains, and social support in the work setting. The dominant social support hypothesis has been that it buffers the impact of stressors on manifestations of strain. However, a great deal of confusion and imprecision has accompanied the use of this term. Most authors have used the term in a way consistent with the explicit definition of LaRocco, House, and French (1980). Social support is hypothesized to interact with stressors such that the relation between stress and strain is stronger for persons with low levels of social support than for those with high levels of support. In other words, social support moderates

This research was supported by a grant from the National Institute of Mental Health (1-R01-MH34408), Daniel C. Ganster, Principal Investigator. We would like to thank Mary Barton and Pamela Perrewe for their assistance in data collection and coding. We would also like to acknowledge the helpful comments of Robert Guion and two anonymous reviewers. Correspondence concerning this article should be sent to Daniel C. Ganster, Department of Management, University of Nebraska, Lincoln, Nebraska 68588.

the stress-strain relation. It is important to distinguish between the terms moderate and mediate, because both have been used in reference to the buffering effect. Following James and Brett (1984), the buffering role of social support refers to a moderating effect; the terms buffering and moderating will be used interchangeably. Findings have been inconsistent. Several studies report evidence of the moderating effect (Abdel-Halim, 1982; Gore, 1978; House, McMichael, Wells, Kaplan, & Landerman, 1979; Karasek, Triantis, & Chaudry, 1982; Kobasa & Puccetti, 1983; LaRocco et a]., 1980; Lefcourt, Martin, & Saleh, 1984; Sandier & Lakey, 1982; Seers, McGee, Serey, & Graen, 1983; Wilcox, 1981). Many of these investigations, however, did not find consistent effects across different (a) stressors and indexes of strain, (b) sources of support, and (c) personal characteristics of the subjects. For example, LaRocco et al. (1980) reported that social support moderated the effects of stressors on health outcomes such as depression and somatic complaints, but they found no evidence of the effect on job-related strains such as job dissatisfaction and boredom. Kobasa and Puccetti (1983) reported that support from the boss buffered the effect of critical life events on illness symptoms but that support from the family did not. Sandier and Lakey (1982) found social support buffered the impact of critical life events on depression and anxiety for persons with an internal locus of control but not for those with an external locus of control. In sum, the evidence of moderating effects is equivocal, suggesting that their existence may depend on the source of support, the recipients, and the stressors and strains being examined. In addition to the studies that to some extent support buffering effects, other investigations have not (Aneshensel & Stone, 1982; Blau, 1981; Ganellen & Blaney, 1984; Lin, Simeone, Ensel, & Kuo, 1979; Turner, 1981). Still others have reported what might be termed opposite buffering effects. That is, social support ap-

102

SOCIAL SUPPORT

103

peared to exacerbate the effects of stressors on strains. For example, Beehr (1976) noted that work group support tended to increase the impact of role ambiguity on job dissatisfaction. Similar opposite buffering findings were reported by AbdelHalim (1982) and Kobasa and Puccetti (1983). In summary, the literature is unclear about the generality of a buffering effect of social support on stress. Reasons and research strategies are discussed below. Methodological Issues Several recent reviews of the social support literature have suggested that some of the discrepant results may be accounted for by methodological shortcomings (Gore, 1981; House, 1981; Thoits, 1982). To provide a rigorous examination of the role of social support in the workplace, the present study was designed to meet three methodological concerns that generally have not received adequate treatment in past research. The first issue concerns the construct of social support itself. Many operationalizations have been used in the literature, differing in objectivity, dimensionality, and meaning. Social support can be broadly defined as "the availability of helping relationships and the quality of those relationships" (Leavy, 1983, p. 5). This definition connotes social ties of a positive nature. Although objective indicators such as marital status, the size of the individual's social network, and the number of social contacts have all been used to make inferences about the amount of social support one receives, such measures are deficient because they omit assessment of the quality of such relations. Also neglected is assessment of social support dimensions such as who provides the support and what form it takes. Sources of support include co-workers, supervisors, friends, and families. Support might take the form of emotional reassurance or assistance in meeting goals. A knowledge of the effects of support dimensions may have implications for structuring the work environment as well as for better understanding the dynamics of social support in the stress process. The present study uses a measurement approach that attempts to assess directly individuals' perceptions of various forms of positive support from three different sources. Second, much of the research investigating social suport at work has relied on a limited number of stress measures (e.g., role conflict and ambiguity); clearly, other organizational and job characteristics may be related to subsequent strains. Furthermore, social support has been found to act as a moderator for some strains but not others (LaRocco et al., 1980). In particular, it appears to matter whether the strains reflect health outcomes, such as somatic complaints, or whether they are explicitly job-related or attitudinal. The present study is multivariate regarding both stressors and strains. Finally, size and heterogeneity of the sample are important methodological issues. To obtain statistical power sufficient to detect interactions of typically small effect size, a large sample is needed. Large samples have not been the norm in the social support literature (see Thoits, 1982), so much of the apparent variability in reported findings might be a function of sampling error (see Hunter, Schmidt, & Jackson, 1982). In addition, an ideal sample would represent a range of personal and work setting characteristics. Such a sample would allow the testing of higher order interactive effects as are now frequently being proposed

(Mitchell & Trickett, 1980). For example, social support might buffer the stress-strain relation in the presence of some work setting characteristics but not others. Such characteristics could be involved in a three-way interaction with social support and stressors. Caplan, Cobb, French, Harrison, and Pinneau (1975) reported that holders of blue- versus white-collar positions had different exposure and response to work stressors. This finding may be due to the different socioeconomic status levels that tend to be associated with such positions. Social class has been shown to influence vulnerability to stress and also may determine whether social support moderates the stress-strain relation (Turner & Noh, 1983). Furthermore, other characteristics such as sex and education might alter the buffering effect. Women tend to have more and different types of supportive ties than men (Leavy, 1983). Etzion (1984) reported that the work stress-burnout relation was moderated by work sources of support for men and by nonwork sources for women. Finally, because a purpose of education is to improve individuals' abilities to cope, it may influence stress reactions. Furthermore, it is a component of social class. Its role with regard to stress, social support, and strain is therefore explored. The present study, then, examines (a) the main effects of social support on strain outcomes, (b) the interactive, or moderating, effect of social support in combination with work stressors, and (c) higher order interactions involving social support, stressors, and personal and job variables. The personal and occupational factors examined are sex, educational level, and job type, as indexed by blue- versus white-collar position. Moreover, the study examines multiple sources of social support and stress and both work and non-work related strains.

Method
Subjects
Employees of a large contracting firm were recruited to participate in the study. The sample was contacted at two of the firm's job sites and at its corporate headquarters. In all, 326 employees provided complete data. Eighty-four percent of them were male, they had an average education of 14.4 years, and averaged 32.5 years in age. Approximately 60% occupied construction trade jobs such as electrician, welder, and plumber, while the rest were employed as accountants, engineers, and clerical support staff, and middle to upper level managers. Subjects were recruited as volunteers for a "work stress study." Participants received a personal report of their responses and also received time off from their job to provide data. This sample represented about 80% of the employee populations at these sites. This relatively high response rate was probably due in part to the employers allowing data collection sessions during regular working hours.

Measures
Social support from three sources (supervisor, co-workers, and family and friends) was measured with the three subscales used in the Caplan et al. (1975) survey. Each subscale consists of four 5-point Likert-scaled items. Unlike some procedures, which measure social support indirectly (for example, as number of social contacts), these scales were chosen because they directly assess the subject's perception regarding the level of social support received. The types of support assessed by these scales are illustrated by the following items: "How much does each of these people go out of their way to do things to make your life easier for you? How easy is it to talk with each of the following people? How much can each of these people be relied on when things get tough at work? How

104

D. GANSTER, M. FUSILIER, AND B. MAYES

much is each of the following people willing to listen to your personal problems?" Subjects responded to these items in terms of support received from "your immediate supervisor," "other people at work," and "your spouse, friends, and relatives." Six job stressors were assessed. They were chosen because they are often found to be the strongest correlates of individual strain (see Caplan et al., 1975). Role conflict and ambiguity were measured with the scales of Rizzo, House, and Lirtzman (1970). The items use 7-point response formats. The other stressors were measured with scales from the Caplan et al. (1975) survey. Quantitative work underload refers to how much work one has to do and was measured with an 11-item scale. High scores on this variable indicate the subject perceives he or she has too little to do. Lack of Variability refers to the extent that the level of one's workload remains constant rather than changes from low to high levels and was measured with 3 items. Skill underutilization refers to the degree to which
4 OO (N P^ OO

one's job does not require the worker to use his or her valued skills, and was indexed with 3 items. Responsibility for others refers to the level of responsibility that a worker has for the welfare and future of others, and was assessed with 4 items. The preceding four scales employed the same items as reported by Caplan et al. (1975). They involved 5-point response formats. For ease of interpretation, some of the scales were simply reverse scored and renamed so that "stressor" variables were all operationalized so that high scores were associated with high levels of stress. For example, what we refer to as "skill underutilization" is simply the original "skill utilization" scale with reverse scoring. The outcome variables consisted of depression, job dissatisfaction, life dissatisfaction, and somatic complaints. Depression was assessed with the Caplan et al. (1975) scale. Job dissatisfaction was measured with a sexless form of the "faces" scale (Kunin, 1955), and general life dissatisfaction with the Quinn and Shepard (1974) scale. The somatic complaints scale asked the respondent to list the frequency of 17 symptoms such as headaches, nausea, and sweaty palms. An overall value for somatic complaints was obtained by averaging the scores on the 17 individual symptoms. All of the measures used in this study are fairly standard scales that have demonstrated acceptable reliability in past studies. Reliability estimates for the present sample are reported in Table 1 and are consistent with those reported previously. The job dissatisfaction scale, although consisting of a single item, has shown high convergence with other multi-item satisfaction scales. Oanster(1978) reported correlations ranging from .80 to .90 with a multi-item Likert-scaled measure and a semantic differential measure.
i<N
3 '* \ O \ \ \

a^ o* o o p S ON r r f

o (N o ^H

' r

oo ts

en <N p -; *

fN

pppCJ'N

- - ^

pr

(-

Procedure
In most studies of job stress, measures of stressors and strains are obtained from the self-reports of respondents in one questionnaire administration. This practice can lead to inflated correlations between stressors and strains because of the influence of common method variance and response consistency effects. In an attempt to minimize such artifacts, the task and role stressor scales and the social support scales were administered in one questionnaire, and the outcome measures were administered in another questionnaire several days later. Although we know of no data that empirically support the effects of temporal separation on these measurement artifacts, that such separation might lessen them seemed a plausible assumption. Such temporal separation should also serve to reduce the fatigue associated with long questionnaire administration sessions. All data were obtained in group administrations supervised by members of the research team during working hours in facilities provided by the organization.
j -c oo c
rn O

f r f

oo rn rq f

-SB

Results Main Effects of Stressors on Strains

03 Si

III

1,

The means, standard deviations, reliabilities, and intercorrelations of all study variables are listed in Table I. As can be seen

II

- .s,

""{llflfl

I^I8^

SOCIAL SUPPORT from the zero-order correlations, the stressor variables generally show small, but statistically significant, relations with the outcome variables. To get a more accurate picture of the combined main effects of the task and role variables on the outcome variables, a canonical analysis was computed. As is the case in most work stress studies, the various stressor variables are intercorrelated. The same is true for the strain variables. Canonical analysis provides a multivariate assessment of the overall relation between stressors and strains while considering the correlations within the two sets of variables (see Cohen & Cohen, 1983 for an explanation of canonical analysis). Results are displayed in Table 2. Two of the four canonical correlations were statistically significant. As indicated by a redundancy analysis (see Cohen & Cohen, 1983), a total of 13% of the variance of the strain measures was explained by the stressors through these two sets of canonical variates, with the first root explaining almost all of this (12%). An examination of the structural coefficients, which are simply the correlations of the variables with the canonical variates, provides a method of interpreting the variates themselves. The two strain variables with the highest structural coefficients for Variate 1 are job dissatisfaction (.89) and depression (.63), whereas the coefficients for life dissatisfaction (.32) and somatic complaints (.24) are of considerably less magnitude. On the stressor side, skill underutilization has the highest structural coefficient (.77), but role conflict (.48), role ambiguity (.53), and underload (.54) are also fairly strongly related to the variate. Thus, high scores on this variate appear to reflect high levels of skill underutilization and work underload, as well as high levels of role conflict and ambiguity. Because the two variates are correlated positively (canonical r = .59), jobs with high levels of role conflict, ambiguity, skill underutilization, and underload are associated with both job dissatisfaction and depression. The second strain variate essentially reflects somatic complaints, whereas its associated stressor variate primarily reflects work underload, which is negatively loaded (.82). Because the second set of variates is orthogonal to the first set, it appears that underload is associated with lower levels of somatic complaints, and this effect is independent of its association with job dissatisfaction and depression. Variable Life dissatisfaction Job dissatisfaction Somatic complaints Depression Redundancy Total redundancy Role conflict Role ambiguity Underload Lack of variability Underutilization Responsibility Structural Coefficients Canonical Variate 1 Canonical Variate 2 Table 2 Canonical Analysis of Stressors and Outcomes

105

.32 .89 .24 .63 .12 .13 .48 .53 .54 .38
.77

.02 .16 .82


-.12

.01 .49 .20


-.82 -.42

-.21

.32 .26

Note. For Root No. 1, canonical correlation = .59, F(24, 1107) = 7.38, p < .01. For Root 2, canonical correlation = .22, fl(15, 878) = 1.73. Structural coefficients are the correlations between the individual variables and their associated canonical variates.

from family and friends makes a smaller contribution (.39). Thus it appears that a lack of social support from individuals at work, and in particular, from the supervisor, is most strongly related to workplace strain (job dissatisfaction), but is also related to strains not specific to work. The second variate set indicates that a lack of support from family and friends is associated with higher levels of somatic health complaints. Moderating Effects of Social Support

To say that social support moderates the effects of stressors on strains is to assert that there is a significant interaction between social support and stressors such that the effects of stressors are less pronounced when accompanied by high levels of social support. In this study there are multiple measures of all three factors, thus somewhat complicating the analysis of interactions. Although hierarchical multiple regression using product terms to carry interactions is the appropriate analytical model (Cohen & Cohen, 1983), one has some choice in deciding how many stressors and social supports will be examined in each regression analysis. This decision involves making some trade-off between sacrificing power by putting all variables into one analysis and tolerating some amount of Type I error inflation by examining separate models. The lower power option is to regress each strain variable on all 6 stressors, all 3 social support measures, and all 18 product terms (Each Social Support X Each Stressor). This type of omnibus buffering test consumes 27 degrees of freedom. An analysis that considers each stressor variable separately along with the 3 social support measures will yield higher power tests but also more Type 1 error inflation. Each of these regressions consumes only 7 degrees of freedom, but with 6 stressors and 4 dependent variables, 24 different regressions need to be computed. Beginning with the more conservative approach, 4 omnibus regressions were computed, each of which involved a different strain outcome as the dependent variable and all the stressors

Main Effects

of Social Support on Strains

The same analytical strategy was followed in assessing the main effects of social support on strain outcomes. A canonical analysis used the three social support variables and the four strain variables. Results are displayed in Table 3. Two significant canonical correlations were yielded with a total of 6% of the strain variance being explained through the two sets of variates. The first variate set accounted for 5% of the strain variance, whereas the second accounted for 1%. The structural coefficients suggest that the first strain variate primarily reflects high job dissatisfaction (.87). However, the other strains are also moderately represented in the variate, with life dissatisfaction (.56), somatic complaints (.44), and depression (.65) positively related. On the social support side, support from the supervisor (.91) is clearly the dominant factor, with 80% of its variance accounted for by the variate. Co-worker support is also significantly related (.61), though at a lower level; and support

106
Table 3

D. GANSTER, M. FUSILIER, AND B. MAYES interaction between social support from family and friends and lack of variability. Lack of variability has a slight positive effect on job dissatisfaction for those with low support. For those with high support, lack of variability has a relatively stronger, and negative, impact on job dissatisfaction. This result is not consistent with our definition of the buffering effect, but it does suggest that those with high support suffer less dissatisfaction when exposed to a lack of variability than those with low support. In light of the total lack of interactions in the omnibus tests,

Canonical Analysis of Social Support and Outcomes Structural Coefficients Canonical Variate 1
.56 .87 .44 .65 .05 .06 -.91 -.61 -.39

Variable Life dissatisfaction Job dissatisfaction Somatic complaints Depression Redundancy Total redundancy Supervisor support Coworker support Friends support

Canonical Variate 2
.46 -.45 .65 .40 .01 .32 -.20 -.87

Table 4 Social Support InteractionsTotal Sample


B1 F

Note. For Root No. 1, canonical correlation = .34, f\l2, 844) = 5.12, p < .01. For Root No. 2, canonical correlation = .23, F{6, 640) = 3.47, p < .01. Structural coefficients are the correlations between the individual variables and their associated canonical variates.

Dependent variable: Life dissatisfaction Stepl Lack of variability (LV) Supervisor support (SS) Co-worker support (CS) Friends support (FS) AS2 = .05 Step 2 L V X SS LVXCS LVXFS Intercept AJ?! = .03 Total R3 = .08 Adj. R* = .06

and social supports. These 4 regressions yielded no significant interactions between social support and stressors. Thus, although social support is associated with lower strain, as indicated by the main effects analyses, support does not appear to have a significant buffering effect. Following the analytic logic discussed above, however, it is possible that no interactions were discovered because the tests were of low power. To give every benefit of the doubt to the buffering hypothesis, the alternate strategy, involving the computation of 24 regressions, was also implemented. In each of these regressions, a strain variable was regressed on a stressor and the 3 social support variables, then the set of 3 product terms was entered into the regression. If the interaction set made a significant contribution to R, then the significance of each of the 3 product variables within the set was considered. Of these 24 regression analyses, only 3 yielded evidence of interactions between social support and a stressor. Table 4 provides summary results of the 3 regression analyses that yielded significant interactions. In Case 1, lack of variability and support from coworkers significantly interact to affect life dissatisfaction. A graph of this interaction appears in Figure 1. For illustrative purposes, in Figure 1 and in the following figures, a value of 4 (on the 5-point scale) was chosen to represent high support, and a value of 2 was chosen to represent low support. Thus, in the first case, lack of variability in workload leads to greater dissatisfaction for those with high levels of co-worker support, but not for those with low support. This case is clearly not consistent with the buffering hypothesis. In fact, it might even be termed opposite buffering. In Case 2, skill Underutilization interacts with support from the supervisor in affecting life dissatisfaction (see Figure 2). Again, opposite buffering seems evident in that skill Underutilization leads to more dissatisfaction for workers with high support from the supervisor than for workers with less support. Figure 3 illustrates one of the two significant interactions in Case 3yet another case of apparent opposite buffering. A lack of variability leads to greater job dissatisfaction, and this effect is more pronounced for workers with high support from the supervisor than for those with low support. Finally, Figure 4 demonstrates the

-.85 -.31 -.80 .05

2.48 2.02 7.16* .03 4.51"


.62 6.89" .86 20.17 3.00' 3.91"

.06 .27 -.08 7.16

Dependent variable: Life dissatisfaction Step 1 Underutilization (SU) Supervisor support (SS) Co-worker support (CS) Friends support (FS) AJ?2 = .06 Step 2 SUXSS SUXCS SUXFS Intercept AJ?2 = .03 Total R2 = .09 Adj. R2 = .07

-.10 -.44 -.21 .15

.07 7.09" .86 .42 4.97"

.11 .07 -.12 4.83

4.49* .82 3.28 19.11 3.18* 4.27**

Dependent variable: Job dissatisfaction Step 1 Lack of variability (LV) Supervisor support (SS) Co-worker support (CS) Friends support (FS) AJ?2 = .14 Step 2 LVXSS LVXCS L V X FS Intercept &R2 = .03 Total R2 = .17 Adj. 2 = .15
.77 -1.15 -.23 .98 .96 12.99" .27 6.45* 13.04"

.24 .01 -.31 2.75

4.70* .00 5.80* 1.41 3.26* 9.22**

*p<.05.**p<.01. " B values are unstandardized regression coefficients.

SOCIAL SUPPORT

107

High Coworker Support Y=.23(LV)+4.19

'

Low Supervisor Support Y=1.23(LV)t.45

Low Coworkar Support Y=-.31(LV)t5.56

Q O
High Supervisor Support Y=1.73(LV)-1.85

5 1 2 3 4 5

Lack of Variability Figure 1. Interaction between co-worker support and lack of variability.

Lack of Variability
Figure 3. Interaction between lack of variability and supervisor support.

we are very reluctant to attribute much importance to the few interactions that were indicated by these fairly liberal tests. Moreover, of the interactions uncovered, none supports the hypothesized buffering effect for social support.

Sttbsample Analyses
In the present study a sufficiently large and heterogeneous sample exists to allow for the examination of higher order interactions involving social support. In particular, the following analyses examine interactions between stressors and social support within subsamples across which the buffering effect might differ. These variables consist of sex, education, and blue- versus white-collar position. For each variable two subgroups were

formed, and the interactions between stressors and social support were examined within each subgroup. The interaction tests consisted of the same type of regression analyses reported previously. However, because subgrouping produces smaller samples, the lower power omnibus tests were not performed, but only the more liberal tests which examined one stressor variable at a time. Thus 24 regressions were performed on each of 6 subgroups formed by the 3 classification variables. Tables 5, 6, and 7 display some of the results of the subgroup regressions. All significant interaction effects found in any subgroup are listed in these tables along with the regression coefficient for that interaction in the corresponding subgroup, even if it was not significant. The lower and upper bounds of the 95%

Low Supervisor Support Y=.12(SU)+3.95

c .2 1 o>
5

Low Support Y=.15(LV)+4.71

High Supervisor Support Y=.34(SU)+3.07

High Support Y=-.47(LV)+6.67

-i

Skill Underutilization
Figure 2. Interaction between skill Underutilization and supervisor support.

Lack of Variability
Figure 4. Interaction between lack of variability and support from friends and family.

108

D. GANSTER, M. FUSILIER, AND B. MAYES

Table 5
Subsample Interaction Tests: White Collar versus Blue Collar
B"
Dependent variable Job dissatisfaction Job dissatisfaction Job dissatisfaction Life dissatisfaction Depression Interaction term Underload X SS Lack of Variety XSS Underutilization X SS Lack of Variety X CS Responsibility X CS White collar
.03 -.19 -.12 .36* -.16*

95% confidence interval of B Blue collar White collar Blue collar


.22 to .13 to .15 to -.12 to -.09 to

.65* .45* .37* .16 .05

-.37 to -.59 to -.40 to .04 to -.28 to

.43 .21 .16 .68 -.04

1.08 .77 .59 .44 .19

Note. SS = support from supervisor; CS = support from co-workers. White collar n = 124; Blue collar n = 202. " B values are unstandardized regression coefficients.

confidence interval for each coefficient are also displayed in the tables. Several interactions are evident in the groups formed on the basis of a blue- versus white-collar distinction (see Table 5). When graphed, some of these interactions are consistent with the buffering hypothesis and some are not. Note that none of the interactions found in one subgroup are replicated in the other subgroup. This pattern is also true with the subgroups formed on the basis of education (Table 6) and sex (Table 7). That is, interactions found in one group are not evident in the other group. At first glance one might be tempted to conclude that these results suggest that models positing three-way (or higher) interactions are needed to explain the apparently complex effects of social support. Performing similar analyses on groups formed on the basis of personality variables, Lefcourt et al. (1984) reached this same conclusion. However, at least in the present case this conclusion is wrong. If one notes the confidence intervals listed in Tables 5 through 7, it can be seen that they overlap across subgroups. Thus, although a coefficient might be significant at p < .05 in one subgroup and not significant in the other subgroup, they are not significantly different from each other. This conclusion was confirmed, of course, when we computed hierarchical regressions that tested three-way interactions between stressors, social support, and each of the three classification variables (following the procedures of Cohen & Cohen, 1983). The conclusion that higher order interaction models are necessary to explain the effects of social support receives no validation here. Discussion This investigation attempted to make a comprehensive examination of the main and interactive effects of social support

in the experience of stress at work. Major work and demographic subgroups were examined to assess the generality of any moderating effects that social support might have. A number of methodological considerations were addressed in this study to provide more rigorous tests than have been previously reported concerning the effects of social support at work. A summary and discussion of the findings follows.

Main Effects
The stressors explain 13% of the total variance of the strain measures, as indicated by the canonical analyses. Moreover, when each strain is regressed on the six stressor variables, it is evident that they are not all equally affected by the stressors. Only 4% of the variance in life dissatisfaction is accounted for, whereas the stressors explain 29% of the variance in job dissatisfaction. Similarly, the stressors explain 16% of the variance in depression and only 6% of the variance in somatic health complaints. If the temporal separation of stressor and strain measures had its intended effect of reducing common method and consistency effects (which is impossible to assess empirically), the main effects reported here might be viewed as being more conservative than most published results. Social support also shows a significant multivariate association with strain. Although job dissatisfaction has the strongest relation with social support, significant amounts of the variance in life dissatisfaction, depression, and somatic complaints are also explained by support. Of the different sources of support, those sources from the workplace, especially the supervisor, are the most important in affecting strains. Furthermore, support from family and friends is significantly associated with lower levels of somatic health symptoms. Overall, however, less of the total variance in strain (6%), as indicated by the canonical analysis, is accounted for by social support than by the stressors.

Table 6 Subsample Interaction Tests: High School Diploma or Less versus More than High School
95% confidence interval of B Dependent variable Job dissatisfaction Interaction term Lack of Variety XSS < high school > high school
.31'

< high school

> high school


.03 to .59

.23

-.15 to .61

Note, n = 156 for high school diploma or less; n = 170 for more than high school. SS = support from supervisor. * B values are unstandardized regression coefficients. * p < .05.

SOCIAL SUPPORT

109

Table 7 Subsample Interaction Tests: Males versus Females


95% confidence interval of B Dependent variable Somatic complaints Depression Depression Interaction term Underutilization X FS Responsibility X CS Ambiguity X SS Males .02 .00 .03 Females -.33** -.31" .27* Males Females

-.09 to. 13 -.11 to.11 -.05 to. 11

-.59 to -.07 -.53 to -.09 .01 to .53

Note, n - 281 for males and 45 for females. SS = support from supervisor, CS = support from co-workers, FS = support from family and friends. ' B values are unstandardized regression coefficients. *p<.05. **/><.01.

Moderating Effects The general hypothesis that social support moderates the impact that job stresses have on strain responses was examined in a series of regression analyses. The first approach was to simultaneously examine all the interactions between each of the stressors and each source of social support. Such an "omnibus" test was replicated for each of the strain measures. No evidence of a buffering effect emerged. This analytical strategy was followed by one that tested the interaction of each of the stressors with the three sources of support separately. This approach should yield tests of higher power, but because many nonindependent tests were run the overall level of Type I error might be considerably inflated. Despite the liberality of these latter tests, only four significant interaction effects were found, and none is consistent with the buffering hypothesis. We would argue that the evidence is thus very weak regarding the buffering hypothesis of social support. Higher Order Interactions One of the methodological advantages of the present study is that the sample is large enough and diverse enough to accommodate the examination of higher order interactions involving social support. If such higher order interactions do exist they might obscure two-way interactions involving social support. Threeway interactions involving support, stressors, and sex, education, and blue- versus white-collar position were tested, but none was found to be significant, despite the tests being liberal with regard to overall Type I error level. Thus we find little reason to advocate the complex models regarding social support that have recently been suggested. To fit the present negative findings on the buffering hypothesis with previous conflicting evidence, three questions might be considered. First, is the buffering effect specific with regard to particular stressors, strains, and sources of social support? Second, is the effect general across different stressors, sources of support, and strains for some population groups but not others? And third, is the effect specific to particular stressors, sources of support, and strains in certain population groups and specific to different stressors, sources of support, and strains in other groups? With regard to the first question, there is no obvious theoretical basis for expecting social support to act as a moderator of some stressstrain relations and not others. Although statistically significant moderator effects have been found for certain variable combinations, this does not mean that these effects necessarily differ from the nonsignificant effects found for other combinations of variables. With regard to the second and third questions, if the existence of a moderating effect varies across population groups, such dif-

ferences can only be ascertained by testing the three-way interactions among stress, social support, and group characteristics. Unfortunately, reporting of such tests, as well as interaction effect sizes, is rare (for exceptions, see Etzion, 1984; and Ganellen & Blaney, 1984). For example, both Turner and Noh (1983) and Lefcourt et al. (1984) concluded that group characteristics dictated the existence of the moderating effect of social support. Lefcourt et al. investigated locus of control, and Turner and Noh dealt with social class. In both studies, the Social Support X Stress interaction was significantly different from zero in some groups but not in others. These interaction terms, however, were not compared across groups, therefore the size and statistical significance of the three-way interaction of Social Support X Stress X Group was not assessed. Because so few investigators have explicitly tested higher order interactions involving social support, the only way to determine whether the apparent variability of social support buffering effects really does reflect the presence of complex processes, as opposed to sampling error, is to perform a meta-analysis (Hunter et al., 1982). However, we are not hopeful that such an analysis will be completed soon, given the great variety of statistical tests that have been used to assess the buffering effect and the propensity of authors to omit the reporting of effect sizes of interactions. Until a meta-analysis can be performed, attempts to reconcile conflicting findings can only proceed by first identifying major differences between supportive and nonsupportive studies, then seeking explanations for why these differences should influence the buffering effect. With regard to the first point, support for the buffering hypothesis appears to be more prevalent in studies concerning life events as sources of stress than in studies concerning work stressors. A possible explanation for this difference hinges on the fact that many life events directly reduce social support, whereas work stressors do not. Examples of such life events are divorce, death of a spouse, and change of residence. Stressors that are causally related to a particular type of support may have a different impact on strain depending on one's overall level of social support. The work stressor-strain relation, however, may not vary with regard to one's overall support level, as work stressors typically are not causally related to support. Hence the buffering effect might be found for certain types of life event stressors but not for work stressors. This suggests that when constructing models of the joint effects of stress and social support, the causal nature of the stress-support relation should be considered. Conclusions From the foregoing discussion several conclusions seem warranted. First, social support, especially from one's supervisor,

110

D. GANSTER, M. FUSILIER, AND B. MAYES House, J. S. (1981). Work stress and social support. Reading, MA: Addison-Wesfey. House, J. S., McMichael, A. J., Wells, J. A., Kaplan, B. R, & Underman, L. R. (1979). Occupational stress and health among factory workers. Journal oj Health and Social Behavior, 20. 139-160. Hunter, J., Schmidt, F, & Jackson, G. (1982). Mela-analysis: Cumulating research findings across studies. Beverly Hills, CA.: Sage. James, L. R., & Brett, J. M. (1984). Mediators, moderators, and tests for mediation. Journal of Applied Psychology, 69, 307-321. Karasek, R. A., Triantis, K. P., & Chaudhry, S. S. (1982). Coworker and supervisor support as moderators of associations between task characteristics and mental strain. Journal of Occupational Behavior, 3, 181-

shows a consistent relation with a variety of affective and somatic outcomes. It appears that interventions designed to provide this type of support might have a beneficial impact on the mental and physical welfare of workers, and should be encouraged. Second, work stressors such as role conflict, role ambiguity, and skill underutilization are associated with indices of mental and physical poor health. Third, the effects of job stressors appear to exist independently of the level of social support. That is, although social support may have main effects on strains, it does not moderate the effects of stressful conditions in the workplace. The distinction between main effects and buffering effects has important implications for practice. Social support appears to have beneficial (albeit small) effects, but it does not seem to reduce the impact that work stressors have on strains. Moreover, there is little convincing evidence that certain groups of individuals will derive benefit from the receipt of social support, whereas others will not. Thus, certain groups of workers should not be denied increases in social support on the assumption that it will not help them. Finally, a main effects model of social support implies the implementation of a stress management strategy that focuses on both the reduction of harmful job and role characteristics, such as conflict and ambiguity, and the augmentation of such positive factors as use of skills and social support.

200.
Kessler, R. C., Price, R. H., & Wortman, C. B. (1985). Social factors in psychopathology: Stress, social support, and coping processes. Annual Review of Psychology, 36, 531-572. Kobasa, S. C. Q, & Puccetti, M. C. (1983). Personality and social resources in stress resistance. Journal of Personality and Social Psychology, 45, 839-850. Kunin, T. (1955). The construction of a new type of attitude measure. Personnel Psychology, 68, 334-337. LaRocco, J. M., House, J. S., & French, J. R. P. (1980). Social support, occupational stress, and health. Journal of Health and Social Behavior, 21, 202-218. LaRocco, J. M., & Jones, A. P. (1978). Coworker and leader support as moderators of stress-strain relationships in work situations. Journal of Applied Psychology, 63, 629-634. Leavy, R. L. (1983). Social support and psychological disorder: A review. Journal of Community Psychology, II, 3-21. Lefcourt, H. M., Martin, R. A., & Saleh, W. E. (1984). Locus of control and social support: Interactive moderators of stress. Journal of Personality and Soda! Psychology, 47, 378-389. Lin, N., Simeone, R. S., Ensel, W. M., & Kuo, W. (1979). Social support, stressful life events, and illness: A model and an empirical test. Journal of Health and Social Behavior, 20, 108-119. Mitchell, R. E., & Trickett, E. J. (1980). Task force report: Social networks as mediators of social support. Community Mental Health Journal, 16, 27-44. Quinn, R. P., & Shepard, L. J. (1974). Quality of Employment Survey. Ann Arbor, MI: Institute for Social Research, University of Michigan. Rizzo, J., House, R., & Lirtzman, S. (1970). Role conflict and ambiguity in complex organizations. Administrative Science Quarterly, 15, 150-

References Abdel-Halim, A. A. (1982). Social support and managerial affective responses to job stress. Journal oj Occupational Behavior, 3, 281 -295. Aneshensel, C. S., & Stone, J. D. (1982). Stress and depression: A test of the buffering model of social support. Archives of General Psychiatry, 39, 1392-1396. Beehr. T. A. (1976). Perceived situational moderators of the relationship between subjective role ambiguity and role strain. Journal of Applied Psychology, 61, 35-40. Blau, G. (1981). An empirical investigation of job stress, social support, service length, and job strain. Organizational Behavior and Human Performance, 27, 279-302. Caplan, R. D., Cobb, S., French, J. R. P., Harrison, R. V., & Pinneau, S. R. Jr. (1975). Job Demands and Worker Health. Washington D.C.: H.E.W. Publication No. NIOSH 75-160. Cohen, J., & Cohen, P. (1983) Applied multiple regression/correlation analysis for the behavioral sciences (2nd ed). Hillsdale, NJ: Erlbaum. Etzion, D. (1984). Moderating effect of social support on the stress-burnout relationship. Journal of Applied Psychology, 69, 615-622. Ganellen, R. J., & Blaney, P. H. (1984). Hardiness and social support as moderators of the effects of life stress. Journal of Personality and Social Psychology, 47. 156-163. Ganster, D. C. (1978). The effects of individual differences and objective task scope on task perceptions and satisfaction: A laboratory investigation. Unpublished PhD. dissertation, Purdue University. Ganster, D. C., Mayes, B. X, Sime, W., & Tharp, G. (1982). Managing organizational stress: A field experiment. Journal of Applied Psychology, 67. 533-542. Gore, S. (1978). The effect of social support in moderating the health consequences of unemployment. Journal of Health and Social Behavior, 19, 157-165. Gore, S. (1981). Stress-buffering functions of social supports: An appraisal and clarification of research models. In B. S. Dohrenwend & B. P. Dohrenwend (Eds.), Stress life events and their contexts (pp. 202-222). New York: Prodist.

163. Sandier, I. N., & Lakey, B. (1982). Locus of control as a stress moderator:
The role of control perceptions and social support. American Journal of Community Psychology, 10, 65-80. Seers, A., McGee, G. W., Serey, T. T., & Graen, G. B. (1983). The interaction of job stress and social support: A strong inference investigation. Academy of Management Journal, 26, 273-284. Turner, R. J. (1981). Social support as a contingency in psychological well-being. Journal of Health and Social Behavior, 22, 357-367. Turner, R. J., & Noh, S. (1983). Class and psychological vulnerability among women: The significance of social support and personal control. Journal of Health and Social Behavior, 24, 2-15. Thoits, P. A. (1982). Conceptual, methodological, and theoretical problems in studying social support as a buffer against life stress. Journal of Health and Social Behavior, 23, 145-159. Wilcox, B. L. (1981). Social support, life stress, and psychological adjustment: A test of the buffering hypothesis. American Journal of Community Psychology, 9, 371-386.

Received December 14, 1984 Revision received April 26, 1985

Vous aimerez peut-être aussi