Psychologkal Bulletin Copyright 1985 by the American Psychological Association, Inc. 1985, Vol. 98. No. 2, 31 0033-2909/85/$00.75
Stress, Social Support, and the Buffering Hypothesis
Sheldon Cohen Thomas Ashby Wills Carnegie-Mellon University Cornell University Medical College
The purpose of this article is to determine whether the positive association between social support and well-being is attributable more to an overall beneficial effect of support (main- or direct-effect model) or to a process of support protecting persons from potentially adverse effects of stressful events (buffering model). The review of studies is organized according to (a) whether a measure assesses support structure or function, and (b) the degree of specificity (vs. globality) of the scale. By structure we mean simply the existence of relationships, and by function we mean the extent to which one's interpersonal relationships provide particular resources. Special at- tention is paid to methodological characteristics that are requisite for a fair com- parison of the models. The review concludes that there is evidence consistent with both models. Evidence for a buffering model is found when the social support measure assesses the perceived availability of interpersonal resources that are re- sponsive to the needs elicited by stressful events. Evidence for a main effect model is found when the support measure assesses a person's degree of integration in a large social network. Both conceptualizations of social support are correct in some respects, but each represents a different process through which social support may affect well-being. Implications of these conclusions for theories of social support processes and for the design of preventive interventions are discussed.
During recent years interest in the role of experiments, and prospective surveys suggest social support in health maintenance and dis- that social support is a causal contributor to ease etiology has increased (e.g., G. Caplan, well-being (cf. S. Cohen & Syme, 1985b; 1974; Cassel, 1976; Cobb, 1976; Dean & Lin, House, 1981; Kessler & McLeod, 1985; 1977; Gottlieb, 1981, 1983; Kaplan, Cassel, & Turner, 1983; Wallston, Alagna, DeVellis, & Gore, 1977; Sarason & Sarason, 1985). Nu- DeVellis, 1983). merous studies indicate that people with The purpose of this article is to consider the spouses, friends, and family members who process through which social support has a provide psychological and material resources beneficial effect on well-being. Although nu- are in better health than those with fewer sup- merous studies have provided evidence of a portive social contacts (Broadhead et al., 1983; relation (i.e., a positive correlation between Leavy, 1983; Mitchell, Billings, & Moos, 1982). support and well-being), in theory this result Although the many correlational results do not could occur through two very different pro- by themselves allow causal interpretation, cesses. One model proposes that support is re- these data in combination with results from lated to well-being only (or primarily) for per- animal research, social-psychological analogue sons under stress. This is termed the buffering model because it posits that support "buffers" (protects) persons from the potentially patho- Authorship of this article was equal. Order of authors genic influence of stressful events. The alter- was determined alphabetically. Preparation of this article native model proposes that social resources was partially supported by National Science Foundation have a beneficial effect irrespective of whether Grant BNS7923453, National Heart, Lung and Blood In- stitute Grant HL29547, and National Cancer Institute persons are under stress. Because the evidence Grant CA38243 to the first author and National Heart, for this model derives from the demonstration Lung and Blood Institute Grant HL/HD21891 and Na- of a statistical main effect of support with no tional Cancer Institute Grants CA25521 and CA33917 to Stress X Support interaction, this is termed the second author. Requests for reprints should be sent to Sheldon Cohen, the main-effect model. Understanding the rel- Department of Psychology, Carnegie-Mellon University, ative merits of these models has practical as Pittsburgh, Pennsylvania 15213. well as theoretical importance because each 310 SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 311
has direct implications for the design of inter- standpoint of health psychology, several pro- ventions. spective epidemiological studies have shown This article reviews evidence relevant to a that social support is related to mortality. This comparative test of the main effect and buff- was shown in 9- to 12-year prospective studies ering models. We consider in detail the meth- of community samples by Berkman and Syme ods used to measure social support and the (1979) and House, Robbins, and Metzner methodological issues relevant to providing a (1982) and in a 30-month follow-up of an aged fair comparison of these models. Because sample by Blazer (1982). In these studies, much of the pioneering research in this area mortality from all causes was greater among was not theoretically designed, considerable persons with relatively low levels of social sup- diversity across different investigators exists in port. Similarly, several prospective studies us- the conceptualization and measurement of so- ing mental health outcome measures have cial support. Hence, results have disagreed, and shown a positive relation between social sup- previous discussions of the social support lit- port and mental health (Aneshensel & Frer- erature arrived at different conclusions about ichs, 1982; Billings & Moos, 1982; Henderson, whether social support operates through a Byrne, & Duncan-Jones, 1981; Holahan & buffering or main effect process (e.g., Mitchell, Moos, 1981; Turner, 1981; Williams, Ware, & Billings, & Moos, 1982;Broadheadetal., 1983; Donald, 1981). Gottlieb, 1981; Leavy, 1983). We posit that, The mechanisms through which social sup- in addition to conceptual inconsistencies, dif- port is related to mental health outcomes and fering results are often attributable to aspects to serious physical illness outcomes, however, of methodology and statistical technique. remain to be clarified. At a general level, it can Hence, we examine those characteristics of be posited that a lack of positive social rela- method and analysis that are relevant to an tionships leads to negative psychological states adequate test of the alternate models of the such as anxiety or depression. In turn, these support process. Detailed consideration is psychological states may ultimately influence given to issues that are particularly important physical health either through a direct effect for tests of interaction effects which have not on physiological processes that influence sus- been extensively discussed in previous sum- ceptibility to disease or through behavioral maries. patterns that increase risk for disease and We begin by presenting conceptual models mortality. In the following section, we outline of stress and support and discussing method- how social support could be linked to health ological and statistical issues that are crucial outcomes on a main effect basis and the mech- for comparing the main effect and buffering anism through which stress-buffering effects models. We then review literature published could occur. This framework is then used for through 1983, classifying studies according to reviewing literature on stress and support. the type of support measure used. The review is limited to studies of informal support sys- Support as a Main Effect tems such as family, friends, or co-workers and excludes studies of professional helpers (see, A generalized beneficial effect of social sup- e.g., DePaulo, Nadler, & Fisher, 1983; Nadler, port could occur because large social networks Fisher, & DePaulo, 1984). Because most of the provide persons with regular positive experi- studies reviewed in this article index symptoms ences and a set of stable, socially rewarded roles of psychological or physical distress rather than in the community. This kind of support could extreme disorder such as clinical depression be related to overall well-being because it pro- or chronic physical illness, we use the term vides positive affect, a sense of predictability symptomatology to refer to criterion variables. and stability in one's life situation, and a rec- ognition of self-worth. Integration in a social Models of the Support Process network may also help one to avoid negative experiences (e.g., economic or legal problems) Numerous studies have shown that social that otherwise would increase the probability support is linked to psychological and physical of psychological or physical disorder. This view health outcomes. Most important from the of support has been conceptualized from a so- 312 SHELDON COHEN AND THOMAS ASHBY WILLS ciological perspective as "regularized social that the potential for serious disorder occurs interaction" or "embeddedness" in social roles (cf. Wills & Langner, 1980). Mechanisms link- (Cassel, 1976; Hammer, 1981; Thoits, 1983, ing stress to illness include serious disruptions 1985) and from a psychological perspective of neuroendocrine or immune system func- as social interaction, social integration, rela- tioning, marked changes in health-related be- tional reward, or status support (e.g., Levinger haviors (e.g., excessive alcohol use, poor diet & Huesmann, 1980; Moos & Mitchell, 1982; or exercise patterns), or various failures in self- Reis, 1984; Wills, 1985). care. This kind of social network support could It is important to note that our psychological be related to physical health outcomes through definition of stress closely links appraised stress emotionally induced effects on neuroendo- with feelings of helplessness and the possible crine or immune system functioning (see loss of self-esteem. Feelings of helplessness Jemmott & Locke, 1984) or through influence arise because of the perceived inability to cope on health-related behavioral patterns such as with situations that demand effective response. cigarette smoking, alcohol use, or medical help Loss of esteem may occur to the extent that seeking (see Krantz, Grunberg, & Baum, 1985; the failure to cope adequately is attributed to Wills, 1983). In an extreme version, the main one's own ability or stable personality traits, effect model postulates that an increase in so- as opposed to some external cause (cf. Garber cial support will result in an increase in well- & Seligman, 1980). being irrespective of the existing level of sup- Following these propositions, possible stress- port. There is some evidence, however, that buffering mechanisms of social support are the main effect of support on major health depicted in Figure 1. As indicated by Figure outcomes occurs for the contrast between per- 1, support may play a role at two different sons who are essentially social isolates (i.e., points in the causal chain linking stress to ill- those with very few or no social contacts) and ness (cf. S. Cohen & McKay, 1984; Gore, 1981; persons with moderate or high levels of support House, 1981).' First, support may intervene (Berkman & Syme, 1979; House et al., 1982). between the stressful event (or expectation of Although the evidence is not conclusive, the that event) and a stress reaction by attenuating suggestion is that there may be a minimum or preventing a stress appraisal response. That threshold of social contact required for an ef- is, the perception that others can and will pro- fect on mortality to be observed, with little vide necessary resources may redefine the po- improvement in health outcomes for levels of tential for harm posed by a situation and/or support above the threshold. bolster one's perceived ability to cope with im- posed demands, and hence prevent a particular Support as a Stress Buffer situation from being appraised as highly stressful. Second, adequate support may in- For the purpose of this article, we posit that tervene between the experience of stress and stress arises when one appraises a situation as the onset of the pathological outcome by re- threatening or otherwise demanding and does ducing or eliminating the stress reaction or by not have an appropriate coping response (cf. directly influencing physiological processes. Lazarus, 1966; Lazarus & Launier, 1978). As Support may alleviate the impact of stress ap- noted by Sells (1970), these situations are ones praisal by providing a solution to the problem, in which the person perceives that it is impor- by reducing the perceived importance of the tant to respond but an appropriate response is not immediately available. Characteristic ef- 1 fects of stress appraisal include negative affect, As noted earlier, support may also prevent the occur- rence of objective stressful events. We discussed this mech- elevation of physiological response, and be- anism in a main effect context because we feel that (a) it havioral adaptations (cf. Baum, Singer, & occurs because of the feedback and structure provided by Baum, 1981). Although a single stressful event embeddedness in a social network and (b) it is independent may not place great demands on the coping of the hypothesis that support protects (buffers) persons abilities of most persons, it is when multiple from the potentially pathogenic effect of experiencing stressful event(s). Asdiscussed later, the lack of correlation problems accumulate, persisting and straining between stress and support found in many of the reviewed the problem-solving capacity of the individual, studies indicates that this is not a prevalent mechanism. SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 313
SOCIAL SUPPORT may result in reappraisal, inhibition of maladjustlv SOCIAL SUPPORT responses, or Hay prevent facilitation of str sss adjustlve counter app •alsal responses
POTENTIAL SPPBAISAL EVENT(S) EMOTIONALLY ILLNESS STRESSFUL -.> PROCESS —> APPRAISED -> LimtED -^3 &/OR ILLNESS EYEHT(S) AS PHYSIOLOGICAL BEHAVIOR | STRESSFUL | BEHAVIORAL ADAPTATION
Figure 1. Two points at which social support may interfere with the hypothesized casual link between stressful events and illness.
problem, by tranquilizing the neuroendocrine for affiliation and contact with others, by help- system so that people are less reactive to per- ing to distract persons from worrying about ceived stress, or by facilitating healthful be- problems, or by facilitating positive affective haviors (cf. House, 1981). moods. This dimension has also been referred to as diffuse support and belongingness. Fi- Social Resources That Operate as nally, instrumental support is the provision of Stress Buffers financial aid, material resources, and needed services. Instrumental aid may help reduce In the following sections we provide brief stress by direct resolution of instrumental definitions and discussions of four support re- problems or by providing the recipient with sources. Our purpose is not to provide a com- increased time for activities such as relaxation prehensive typology of support functions but or entertainment. Instrumental support is also rather to represent those functions measured called aid, material support, and tangible sup- by the instruments used in studies reviewed in port. this article. The terms and functional catego- Although support functions can be distin- ries used here are consistent with social sup- guished conceptually, in naturalistic settings port typologies presented in various discus- they are not usually independent. For example, sions of support (e.g., Antonucci & Depner, it is likely that people who have more social 1982; Barrera & Ainlay, 1983; R. D. Caplan, companionship have more access to instru- 1979; Gottlieb, 1978; House, 1981; Moos & mental assistance and esteem support. Empir- Mitchell, 1982; Silver & Wortman, 1980). ical studies sometimes show an appreciable Esteem support is information that a person intercorrelation between measures of different is esteemed and accepted (e.g., Cobb, 1976; functional support dimensions (e.g., Norbeck Wills, 1985). Self-esteem is enhanced by com- & Tilden, 1983; Schaefer, Coyne, & Lazarus, municating to persons that they are valued for 1981; Wethington, 1982), although some their own worth and experiences and are ac- studies have used scales that provide relatively cepted despite any difficulties or personal independent measures of different support faults. This type of support has also been re- functions. As discussed later, development of ferred to as emotional support, expressive more specific instruments is important for support, self-esteem support, ventilation, and studying how support operates. close support. Informational support is help How do these functions mitigate the effects in defining, understanding, and coping with of stressful events? We suggested earlier that problematic events. It has also been called ad- appraising events as stressful often results in vice, appraisal support, and cognitive guid- feelings of helplessness and threat to self-es- ance. Social companionship is spending time teem. Under these conditions, esteem support with others in leisure and recreational activi- may counterbalance threats to self-esteem that ties. This may reduce stress by fulfilling a need commonly occur as a response to stress ap- 314 SHELDON COHEN AND THOMAS ASHBY WILLS praisal. Informational support that helps one of well-being, they do not directly tap the as- reappraise a stressor as benign or suggests ap- pects of social support that would be responsive propriate coping responses would counter a in the event of highly stressful experiences. perceived lack of control. Hence, esteem and Thus, measures of this type should not gen- informational support are likely to be respon- erally show buffering interactions. Global sive to a wide range of stressful events. In con- functional measures that tap a general avail- trast, instrumental support and social com- ability of resources, without assessing specific panionship functions are assumed to be effec- resources, would similarly result in main ef- tive when the resources they provide are closely fects without buffering interactions. The ex- linked to the specific need elicited by a stressful ception is a global measure that happens to be event. For example, if stress is created by a loss heavily loaded by a type of function (e.g., es- of companionship, then it would be best re- teem support) that was highly relevant for the duced by social companionship. If stress is dominant stressor. This analysis is the basis created primarily by economic problems, then for our review. it would best be alleviated by instrumental support. Measurement of Stress, Social Support, Although stressful events may elicit needs and Symptomatology for multiple resources, it is reasonable to as- As an aid to understanding the variety of sume that specific events elicit particular sa- measures of stress, social support, and symp- lient coping requirements. We posit that there tomatology used in the reviewed literature, the must be a reasonable match between the cop- following section briefly describes the mea- ing requirements and the available support in surement procedures used for each. order for buffering to occur. This analysis pre- dicts that buffering effects will be observed Measuring Stress when the support functions measured are those that are most relevant for the stressors faced The majority of studies examining the oc- by the person (for further discussion see S. currence of potentially stressful events used Cohen & McKay, 1984). As discussed earlier, some version of a life events checklist. These because of the nature of the stress process, in- measures are based on the assumption that ill- formational and esteem support are likely to ness is related to the cumulative impact of be relevant for a broad range of stressful events. events requiring substantial behavioral adjust- The effectiveness of social companionship and ment (Holmes & Rahe, 1967). Typical check- instrumental support will depend on a more lists assess the occurrence of interpersonal specific match between stressful event and stress (e.g., problems with spouse or children), coping resource. financial difficulties, occupational events (e.g., Most existing buffering studies do not mea- job demotion), unemployment, and legal sure discrete stressful events but rather use cu- problems. Usually, the stress score is simply mulative stress measures such as life event the total number of items checked as having scales. If we assume that high levels of cu- occurred in the recent past (e.g., the past year). mulative stress generally represent needs for In some studies, scores were based on the sum the broadly useful esteem and informational of normative (as determined by judges) stress support, the stress-support matching model weights assigned to each item. Others summed can be applied to this literature. Hence, sup- scores based on individual respondents' per- port measures that provide a reliable index of ceptions of the impact of each event. There is, these functions should show buffering effects. however, little difference between the predictive When structural measures of social support validities of scores based on simple event are used, we predict that only main effects will counts, normative weights (e.g., Lei & Skinner, be observed. By definition, structural measures 1980; Ross & Mirowsky, 1979), and respon- of social integration assess only the existence dent assigned weights (e.g., S. Cohen & Hob- or number of relationships and do not provide erman, 1983; Sarason, Johnson, & Siegel, sensitive measures of the functions actually 1978). provided by those relationships. Although such Other studies measured chronic strains. measures may be correlated with overall levels These are persistent objective conditions that SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 315
require continual behavioral adjustment and whether a measure assesses social network are assumed to repeatedly interfere with ade- structure versus function, and (b) the degree quate performance of ordinary role-related of specificity versus globality of the scale. behavior (Pearlin & Lieberman, 1979). Ex- Structural measures are those that describe the amples of ongoing strains include poverty, existence of relationships. Functional measures marital conflict, parental problems, work are those that directly assess the extent to overload, and chronic illness. which these relationships may provide partic- Finally, a number of studies used specialized ular functions. Specificity is a generic term in- stress scales to assess levels of perceived oc- dicating whether a measure assesses a specific cupational stress. The term perceived stress is structure/function or combines a number of used in this literature to refer to worker's rat- structural/functional measures into an undif- ings on such issues as job satisfaction, occu- ferentiated global index. Global structural pational self-esteem, role conflict, and inequity measures typically combine a variety of items of pay. about connections with neighbors, relatives, and community organizations. Global func- Measuring Symptomatology tional measures similarly combine functional indexes such as informational, instrumental, Psychological symptoms are usually mea- and esteem support into a single, undifferen- sured with standard epidemiological screening tiated measure. instruments, brief self-report scales in which A central premise of this article is that subjects report the occurrence of depression, structural measures provide only a very indi- anxiety, concentration difficulties, physical fa- rect index of the availability of support func- tigue, and a variety of psychosomatic symp- tions. Although it might seem that the number toms. These instruments have been validated of social connections would be strongly related in epidemiological and clinical research and to functional support, studies that have ex- produce generally similar results (see B. P. amined the issue consistently find rather low Dohrenwend et al., 1980). They are highly in- correlations, between .20 and .30 (Barrera, tercorrelated, indexing a general dimension 1981; P. Cohen et al., 1982; Sarason, Levine, that has been termed demoralization (see B. P. Basham, & Sarason, 1983; Schaefer et al., Dohrenwend, Shrout, Egri, & Mendelsohn, 1981). These results are not, however, difficult 1980). Subsets of psychosomatic symptoms to understand. Adequate functional support from these instruments (headache, stomach may be derived from one very good relation- upset, tension, insomnia) are sometimes ana- ship, but may not be available to those with lyzed as separate scales (e.g., Billings & Moos, multiple superficial relationships. 1981; Miller & Ingham, 1979). Measures of From the perspective of our model, the im- physical health typically focus on the presence plications of specificity versus globality are dif- of serious illness or chronic conditions. Some ferent for the two types of measures. For tests investigators (S. Cohen & Hoberman, 1983; of main effects, global structural (social inte- Monroe, 1983) used fairly extensive checklists gration) measures are predicted to be necessary of physical symptoms, which are validated for showing strong main effects because they through indexes of health services utilization. tap the existence of a wide variety of stable Because there may be considerable correlation community connections. In contrast, struc- between psychological and physical symptom- tural measures tapping the existence of only atology measures (e.g., Garrity, Somes, & one relationship (often with a single item) are Marx, 1978; Tessler & Mechanic, 1978), it is presumed to have low reliability for tapping currently not clear whether there is a sharp social integration, and thus to be weak and conceptual distinction between the two di- inconsistent for producing main effects. For mensions. functional measures, our matching model predicts that buffering will be observed when Measuring Social Support a functional measure is well matched to the stressful events under study, implying that only Our review is organized with respect to two specific (and appropriate) functional measures characteristics of the support measures: (a) will show buffering effects. The model suggests 316 SHELDON COHEN AND THOMAS ASHBY WILLS that global (undifferentiated) functional mea- support may partially reduce the effect of stress sures will be less successful because the com- on symptomatology (Figure 2B). Alternatively, posite index may obscure the relevant function. support may totally ameliorate the effect of stress on symptomatology (Figure 2C). In ei- ther case, if the sample is large (allowing for Methodological Issues sufficient statistical power), there would be a significant Stress X Support interaction, to- A necessary characteristic for a comparative gether with statistical main effects for stress test of the main effect and buffering models is and social support. a factorial design including at least two levels An adequate comparative test of the main each of stress and social support. If the main effect and buffering models depends on several effect model is correct, then the pattern of re- methodological and statistical considerations. sults would resemble that diagrammed in Fig- We emphasize important issues in providing ure 2A. In this case, there would be a main a sensitive test of the buffering interaction be- effect for support but no Stress X Support in- cause this test is particularly affected by design teraction. Two patterns of results would be weaknesses. A number of the issues we raise consistent with the buffering model. First, are also important in testing for a main effect.
High [ A B P ' Low Social / Low Social f Support ' Support
5 O High Social High Social 0. Support Support 2
Low Low High Low High STRESS STRESS
High P c
/ Low Social ' Support in S o H a_ / HighSocial , Support
Low Low High STRESS
Figure 2, Depictions of main effect of social support on symptomatology and two forms of the Stress X Social Support buffering interactions. SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 317
Statistical Analyses multiple regression (Dawes, 1969; Reis, 1984). As a result, reasonably reliable measurement The most common statistical procedure instruments, and a large number of subjects used in social support studies when the crite- are required to distinguish a significant mono- rion variable is continuous (e.g., level of de- tone interaction from a main effect. Thus, in pressive symptomatology) is a two-way analysis studies where measurement procedures or of variance, with stress and social support as sample size are suboptimal, one may find a factors, or equivalently a multiple regression significant main effect and a pattern of means analysis with the cross-product term (Stress X consistent with a buffering effect, together with Support) forced into the equation after the an interaction term that does not reach statis- main effect terms for stress and support. When tical significance (cf. Kessler & McLeod, 1985). appropriate data are available, the regression Low reliability or validity of support mea- analysis is preferred because it treats the pre- sures also reduces the probability of showing dictor variables (stress and support) as well as either main or interaction effects. Unfortu- the criterion (symptomatology) as continuous. nately, many investigators used scales that were Regression and analysis of covariance models created post hoc from large data sets or created also provide a means of partialing out con- their own scales without psychometric testing founding variables such as socioeconomic sta- or development. Others used single-item mea- tus (SES) and can be used to control for initial sures that almost necessarily have low reli- symptom level in prospective data analyses. ability. Although most of these scales have Tests for dichotomous criterion variables some face validity, formal psychometric data (e.g., clinically depressed vs. not clinically de- are seldom reported. In a number of cases, we pressed) include chi-square analysis, log-linear argue that a particular scale is more sensitive analyses (especially logit analysis), and analysis to recent stress or to stable personality factors of variance. Analysis of variance is usually than to social support. Such psychometric de- considered unacceptable for categorical cri- ficiencies reduce the probability of demon- terion data, whereas log-linear analysis is con- strating a buffering effect if it truly exists in sidered by many to be the appropriate ap- the population (cf. Reis, 1984). proach (see discussions by Cleary & Kessler, It is desirable to have a sample with broad 1982; Dooley, 1985). A number of investiga- ranges of stress, social support, and symptom- tors dichotomized or trichotomized continu- atology. From a conceptual standpoint, it is ous data and then used categorical analytic advantageous for the persons with relatively procedures. Unless there are good conceptual high and low stress and support levels to be reasons for treating data in this way, this pro- sufficiently different from one another that the cedure is not optimal because it can severely difference is psychologically significant. Statis- decrease statistical power due to the loss of tically, the probability of finding relations be- metric information, and hence reduce the tween these variables increases as variability probability of finding an effect that truly exists in stress, support, and symptomatology in- in the population. creases. This requirement tends to be violated in samples where all subjects are under rela- tively high stress to begin with, such as clinical Characteristics of Studies samples. Results for more homogeneous pop- ulations tend to be less marked than those Several methodological issues are relevant found with general population samples, where for an adequate test of buffering effects (cf. the range of stress is usually considerable. Gore, 1981; Heller, 1979; House, 1981; Kessler Another methodological requirement for & McLeod, 1985). Two of these issues are re- testing a buffering model is a significant rela- lated to sample and measurement parameters. tion between stress and symptomatology. Such First, the relevant interaction (Figures 2B and a relation suggests that there is minimally ad- 2C), termed a monotone interaction, is difficult equate measurement and range of scores for to demonstrate statistically because effects are these factors within the sample. Without a divided between main effect terms and the in- meaningful effect of stress on symptomatology, teraction term in the analysis of variance or there is little possibility of finding a monotone 318 SHELDON COHEN AND THOMAS ASHBY WILLS
Stress X Support interaction.2 In general, for tomatology relations using a modified life studies with large samples this should not be events score where social exit events are ex- a problem because the relation between stress plicitly removed from the total stress score or and symptomatology has been replicated many to restrict data analysis to a subgroup of sub- times with established life event measures. jects whose support level remained constant A subtle but important methodological issue over a longitudinal measurement period. The derives from a possible overlap of stress and latter solution, however, produces a sample se- support measures (Berkman, 1982; Gore, lection bias, arbitrarily excluding persons 1981; Thoits, 1982a). As noted earlier, most whose support level has changed. These ap- studies of the buffering hypothesis have mea- proaches have been used by some investigators sured stress with a checklist of negative life and are noted in the review. events. These measures typically include items about interpersonal discord (e.g., marital Prospective Analyses problems) and social exits (e.g., moving, sep- aration/divorce, family death). By definition, Finally, we should note the importance of such events result in at least temporary loss of using prospective analytic models for data on support resources. Hence, there might be some social support when possible. Concurrent cor- confounding of stress and support measure- relations between support and well-being are ment because the stress and social support in- amenable to three alternative causal interpre- struments might, to some extent, be measuring tations. They may reflect social support caus- the same thing, namely changes in social re- ing changes in symptomatology, symptom- lationships. atology causing changes in support level, or a To understand the problem that confound- third factor (e.g., social class or personality) ing creates for buffering tests, assume that in- causing changes in both support and symp- creased symptomatology is caused by inter- tomatology. When two-wave (Time 1 and Time personal loss or discord. Persons with events 2) longitudinal data are available, the most de- based on these problems would be categorized sirable model is an analysis using Time 2 as having a high level of life events and a low symptomatology as the criterion with Time 1 level of social support. Recall that the buffering life events and social support as the predictors hypothesis predicts that persons with high and Time 1 symptomatology included as a stress and low support will show dispropor- control variable. By focusing on changes in tionately elevated symptomatology (Figures 2B symptomatology that occur as a function of and 2C). If serious confounding between life Time 1 stress and support, this analytic model events and social support measures exists, there helps rule out the possibility that results are would be a marked elevation of symptom- attributable to preexisting symptomatology atology for subjects in one cell (high stress/low causing subsequent life events and loss of sup- support), occurring not necessarily because of port. The use of multiple regression analysis an interaction effect but entirely because of an (or analysis of covariance) also makes it pos- elevated, loss-related stress level. In principle sible to control for third variables (e.g., age, there are several ways to deal with the potential sex, social class) that may be correlated with confounding issue. On a post hoc basis, one can examine the correlation between stress and support measures when these data are re- 2 Pinneau (1975) argued that when testing the buffering ported. If the stress and support measures are hypothesis, one should not test the Stress X Support in- uncorrelated, then the confounding issue teraction if there are not significant main effects for both would not seem important, whereas if there is stress and social support Such a position assumes that one is only looking for a "perfect" monotone interaction (i.e., a substantial negative correlation between both low stress groups have identical mean outcomes) and stress and support indexes, then there was there is a large sample. We do not agree with Pinneau's probably some confounding. In the review position (also see House, 1981; House & Wells, 1978). We section, we give specific attention to this issue are merely arguing that in the absence of a Stress X Support whenever relevant data are available. Alter- interaction, the existence of a main effect for stress provides reasonable evidence that psychometrically sound measures native procedures for dealing with the con- were used, and that there is a reasonable range of scores founding issue are to analyze stress-symp- within the sample. SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 319
the predictors and symptomatology and hence tions, providing a match with the needs elicited affect interpretation of the results. by the stressful events under consideration, are Prospective analysis of social support data most appropriate for testing a buffering model, is not always appropriate. The prospective de- whereas studies using global measures of social sign assumes that predictor variables remain networks would be most appropriate for testing relatively stable over the period of prediction. a model of social support as a main effect pro- Measures of some conceptions of support, cess. especially perceived availability, may fluctuate In addition to considering the extent to considerably over long periods. Moreover, for which a study possesses desirable method- some populations, such as college freshmen ological properties, we give specific consider- and armed forces recruits, support is likely to ation to the nature of the effects that are found. fluctuate dramatically as people enter a new A pure main effect is one in which there is a environment. In these cases, prediction from main effect of social support but no evidence an initial assessment of support to an outcome of an interaction. If there is a significant in- occurring a year or more later would not pro- teraction, we determine (when relevant data vide a true prospective analysis. The time are available) whether social support has any course of the disease under study must also be beneficial effect under low stress.3 We follow considered. Short intervals would be appro- this procedure because the finding of a statis- priate when studying health outcomes with tically significant main effect together with an short developmental periods such as colds, in- interaction does not necessarily provide evi- fluenza, and depression. However, long inter- dence for a main-effect model: A statistical vals are required when predicting a disease main effect typically occurs as an artifact of a with a long developmental period such as cor- significant monotone interaction (Dawes, onary heart disease. Hence, it is critical to 1969; Reis, 1984). A pure buffering effect is consider the correspondence between longi- when mean symptomatology level for low- and tudinal intervals, the time course of the cri- high-support subjects is not significantly dif- terion disease, and the stability of social sup- ferent under low stress (but quite different un- port in the population under study when eval- der high stress); it indicates that support is rel- uating prospective support research (S. Cohen evant only for subjects under stress. Alterna- & Syme, 1985a). Although there are a reason- tively, a combined main effect and interaction able number of studies that have collected lon- is when supported and unsupported subjects gitudinal data, relatively few investigators have differ significantly under low stress with even used a true prospective analysis. This issue is greater difference under high stress. The latter noted in the review section when appropriate. pattern is seldom found in the existing litera- ture. Properties for Adequate Tests Research Review It is worth emphasizing the points outlined In the following section, we review studies previously because they provide a first ap- of the relations among stress, social support, proximation of those issues to be considered and well-being. We limit the review to studies in designing comparative tests of the main ef- published through 1983 that report statistical fect and buffering models. The optimal study would use a large sample with reasonable dis- tributions of stress and support, instruments 3 In the review section, a number of instances are prob- with acceptable psychometric characteristics, lematic because investigators reported inferential statistics stress and support measures that are not con- for main effects and interaction effects without reporting cell means. In cases where there was a significant interac- founded, and optimally a longitudinal design tion, it is problematic to interpret findings of a significant with appropriate prospective analyses. More- main effect, because the main effect may be largely or over, a significant relation between the stress completely artifactual. For cases where appropriate data and symptomatology measures is necessary to are available, we report evidence of a main effect only when there was clear evidence of a difference between support provide a fair test of the buffering hypothesis. groups under low stress. Otherwise, we report findings of Studies using a social support measure that significant main effect and interaction, realizing that the taps one or more independent support func- meaning of the statistical main effect is ambiguous. 320 SHELDON COHEN AND THOMAS ASHBY WILLS analyses testing for an interaction between cial networks. Studies using specific structural stress and social support. The literature is or- measures (summarized in Table I) are pre- ganized with respect to whether the support sented first, followed by global structural mea- measure is (a) structural versus functional, and sures that combine a number of different (b) specific versus global. In some cases two or structural items into a single structural index more kinds of support scales were used in the (summarized in Table 2). As noted previously, same study. In such instances the findings for studies using global structural measures pro- each type of scale are discussed separately in vide a reliable index of social integration and the appropriate sections. There are a few am- are predicted to show consistent support for biguous cases in which a scale includes both the main effect model. Those using specific structural and functional items or items that structural measures provide a less reliable in- are difficult to categorize. We classify studies dex of social integration and should not pro- using these scales in the section that seems vide consistent support. most appropriate but discuss why classification is problematic. Because roughly 90% of exist- Specific Structural Measures ing studies use cumulative life event measures, an organization of the review section based on Several studies in Table 1 used single-item the match between the needs elicited by the measures of the number of social connections stressful event and the coping resources pro- (Husaini, Neff, Newbrough, & Moore, 1982; vided by the support measure would be im- Monroe, Imhoff, Wise, & Harris, 1983; Thoits, practical. However, in cases where specific 1982a; Warheit, 1979). Measures used in these stress measures are used, the relation between studies include number of friends living different stress and support measures is dis- nearby, frequency of visiting, number of rel- cussed. atives living nearby, frequency of church at- The basic details of the studies are included tendance, whether the respondent lived at in Tables 1 through 7. The tables include de- home with the family, and whether the re- scriptions of sample characteristics, design spondent belonged to any social or religious (cross-sectional, longitudinal, or longitudinal- organizations. In general, neither main effects prospective), stress measures, social support nor interactions are found for these measures. measures, results, and special remarks. In the Monroe et al. (1983) observed a main effect text, we provide a summary and integration for living at home and a buffering effect for of the findings for each type of support mea- number of best friends. However, given the sure. We focus on consistencies in results from large number of statistical tests performed in studies with a particular type of measure and Monroe et al.'s study, these results could easily discuss the methodological characteristics of have occurred by chance. studies that are distinguished in some way from In contrast to quantitative counts of social the general trend. Conclusions and theoretical relationships, indexes of significant interper- issues are subsequently considered in the Dis- sonal relationships such as marriage sometimes cussion section.'* show main effects and/or buffering interac- For the sake of brevity, we often refer to a tions. Using measures of enduring strains in main effect for stress or a main effect for sup- three specific role areas, Kessler and Essex port without indicating the direction of the ef- (1982) found that marital status showed sig- fect. Unless otherwise noted, a main effect for stress refers to higher stress associated with in- 4 There are two studies not reviewed in this article that creased symptomatology, and a main effect for are often cited as supportive of the support buffering hy- support refers to greater support associated pothesis (deAraujo, van Arsdel, Holmes, & Dudley, 1973; with decreased symptomatology. Nuckolls, Cassel. & Kaplan, 1972). Both studies used scales that combine, in a single index, items assessing support functions with others measuring nonsocial coping re- Studies Using Structural Measures sources. Both of these studies provide some evidence sug- gesting a buffering role of psychosocial assets. However, of Support because the effect of social support items cannot be sep- arated from the effect of items tapping other resources, it The studies reviewed in this section measure is impossible to interpret the relevance of these studies for a wide range of structural characteristics of so- support hypotheses. SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 321
nificant buffering interactions for all three role Variables based on simple quantitative counts areas. Warheit (1979) found a main effect for such as number of social contacts or organi- marital status, but no interaction with a social zational memberships produce no buffering ef- exit scale. fects and in some cases no main effects (Eaton, Studies by Eaton (1978) and Thoits (1982b), 1978; Husaini et al., 1982; Monroe et al., 1983; based on analyses of items from the New Thoits, 1982a; Warheit, 1979). Our interpre- Haven data set collected by Myers, Lindenthal, tation is that such measures do not reliably and Pepper (1975), similarly suggest effects of index the dimension of social integration that marital status. Eaton's analysis used a regres- is relevant for well-being. Variables that reflect sion model in which Time 2 data were ana- the presence of a significant interpersonal re- lyzed controlling for Time 1 life events and lationship such as marriage or close friendship symptomatology, as well as relevant demo- often produce significant buffering interactions graphic characteristics. This analysis indicated (Eaton, 1978; Kessler & Essex, 1982; Linn & buffering effects for marital status and cohab- McGranahan, 1980; and marginal effects in itation versus living alone (which were not in- Thoits, 1982b). Although the interpretation is dependent); no interactions were found for five less clear for Sandler's (1980) finding that the other single-item support measures tested. presence of an older sibling produced buffering Thoits (1982b) used a similar regression effects, this may also be an example of a sig- model, but restricted the analysis to subjects nificant interpersonal relationship operating as whose marital status was stable from Time 1 a buffer. to Time 2 (i.e., continuously married or con- Another important aspect of this literature tinuously unmarried). As noted earlier, this is that two studies (Kessler & Essex, 1982; procedure biases the sample and may elimi- Sandier, 1980) specifically noted that their nate respondents with greater needs for sup- stress and support measures were uncorre- port. This analysis indicated a higher regres- lated. The finding of significant buffering in- sion coefficient for stress on symptomatology teractions in these studies helps to rule out the among individuals with low support, but the possibility that buffering effects may be arti- interaction was only marginally significant. factual. Buffering effects have been found with two As a general conclusion, the observed find- other single-item structural measures that may ings are consistent with our theoretical model be interpreted as assessing the existence of in- in that specific structural measures that pro- timate relationships. Linn and McGranahan vide a quantitative count of social connections (1980) reported interactions with a 9-point typically do not show significant main effects. scale indicating how often the respondent Somewhat contrary to prediction, measures would talk to close friends about three specific that index the presence of a significant inter- stressors: health problems, divorce, and personal relationship such as marriage or close unemployment. Their support measure in- friendship do show buffering interactions, cluded an assessment of qualitative aspects of particularly when analyzed in relation to spe- relationships ("close friends") and may have cific stress measures. The reason, we believe, been interpreted by the respondents as talking is that on the average an enduring and intimate to close friends about concerns and problems. relationship such as marriage is likely to pro- In a study of maladjusted children, Sandier vide several kinds of functional support. (1980) similarly found that the presence of an older sibling showed a significant interaction, Global Structural Support Scales which was a pure buffering effect. In contrast, the number of parents in the home (1 vs. 2) The studies reviewed in this section used showed only a nonsignificant trend toward a support measures that combine a number of buffering effect. (This was a sample of mal- structural items into a global structural sup- adjusted children, who almost by definition port index. Studies discussed in this section are having problems with their parents.) are summarized in Table 2. In summary, results from studies using spe- In contrast to specific structural measures, cific structural measures are dependent on the these global scales consistently show strong nature of the relationships that are assessed. (lext continued on page 326) p p ~ f= t'"'" Z Z Z 0 -; P 0 s:: '" '" :I: :I: :Il VJ W tv -< 0 '" '" I) '" :I: :I: 8 322tv SHELDON COHEN AND THOMAS ASHBY WILLS for T I T "8 tcsts for £ v white, classify buffer all 10 If support l mwried performed performed measure sample Remarks control and uncorrelated mostly measures entire multiple with difficult symptomatology: Analysis Analyses Stress Subsample Support Physical effect b aU Buffering and for b a all all for for for for only stresson> stressors Psychological None Yes Yes Yes Yes Physical support- of effect for Main fora
>• GHQ Psychological NR Non,
& y" y" Yes STAI) xtety scale scale scale variable an 801, depressive depressive II happiness. ploms satisfactioo symptoms life (CES-D) symptoms sym symptomatology (OHQ, Dependent II-item 2O-ilem Depressive! Depressive Overall of * "of 5 u- I visits of to (e) ijfsl cl|(f) f very no. often, no.
liSS-i talking s(e) a for alone, social! belonging status, and measure
I1.|l -g5si«, of (b) relatives friends (b) visit, attendance nearby. ll to oul friend (d) with status friends, of others having living belonging frequency church, |1close Marital No. Live relatives with nearby, club, going friends (g) (c) to close ebureh friends (b) (e) belong parents, to close Support 8 (a) Marital (a) Frequency I (a) role J I 9 check- life life life on fi .8 (nealth measure •I ,L-1, events 8 'S i '8 S1 checklist events ~train problems, unemployed) homemaker checklist eventS strain. parental divorce. strain, rated list; events items Stress tt2-item 52-item Economic 3 95-item Design Measures interval) (2-year interval) (6-wcek longitudinal Cross-sectional Cross-sectional Longitudinal Cross-sectional = = Structural N N (1983) Haven & Essex I 167 965 2.300 Neff, Imhoff, & Moore = (1978) & 720 study), 1.423 N= Newbrough, McGranahan (New & Harris (1982) (1982) (rural), N (1980) N= Wise, (college students), Study/sample Community Community, Community Specialized Community, Specific Table Husaini, Eaton Linn K5'Sler pt^l Monroe, SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 323 for of terms set on
support i*| B-S& S-3 test performed supported/ support ambiguous. Remarks and measure . Ills 111 ji is performed uncorrdated measures performed problematic unsupported stably stable subjects Analysis Bulfering Stress Stress
ll effects Im/entor)'
iimain such Anxiety of Physical Trait effect State meaning = the STAI for for Buffering that b d for for Scale. health· related ag&reS.'Iion evenu PsychoJogical None Yes y" Yes Iiassumed is il Depression of noted, Physical ?Study 1 support- •ftotherwise of b for Unless tffi:ct c Epidemiologic b. a for Main for for inhihition reponed. Psychological NR NR Yes Yes Center nOl = arc
S J! CES-O elmeans s scale scale scale variable cell •illgroups. depresslve depressive depressive
behavior III because support problems Louisville inhibition), (aggression, symptoms symptoms symptoms c~k1ist Questionnaire. Dependent 18-item 20-item 20-item Child i" III between Health ambiguous (c) to of is how
lived _ "t> 5 - (e) of (e) clubs in no. (d) home, General attend to of visiu, difference upset! of services status, relatives person -= (b) measure in close if on, congruence no people atlend sibling., status (f) of community to often so is inleraccion of i- presence close comfort GHQ Marital Number No. neighbors religious nearby parents with how meetinp often depressed and older ethnic exchange friends. (b) (b) no. religious groups, weaknesses., with, lum discussing (d) 1. Support there MaritaJ (a) (a) (a) Time significant stress
m= of low Tl life health and scale life socia] life under presencc measure event subject 3^0 1 the Inventory. nonhealth related loss checklist related checklist events. events events dimensions ••Si by effect; in Stress 23-item Undesirable. 26-item 32-item i-sl buffer effect Depression pure
21main a is Beck the -= of interval); effect interwl) interval) anily1is (3.year em,,· .o,;ec1ional buffer (2-year (2-year
111 111 8DI Longitudinal Longitudinal Longitudinal Cross-sectional meaning the interaction Mreported. s = = s|5 the not N .III (continued) . Haven Haven city studies = ~80) 517 71 1 982a) case (1979) (I = (J (1982b) NR 720 N study). inner children), elementary (maladjusted (New (New study) N= .:Itool .83 this many Community Community, Specialized Community Study/sample In In Note. \I Warllei! • Thaits Thoils Sandler Table ~ r- '" » r ~ ~ tT1 Z Z Z r- 0 0 -; (') '" '" ::c » » :t :t w ~ tT1 t-.> " " '" 324 SHELDO:t N COHEN AND THOMAS ASHBY WILLS of outcome i.tems set structural! analysis terms sca1e with. Remarks sex differences; buffering functional support predictors. measure done combines No Dichotomous SuPPOrt Physical effect CF in Buffering parental a for for sample Psychological None None Yes Nu Yes Physical support" of effect yes L yes only (set for on a married) both b Mai.n negative clear forc for for for feelings; feelings (non samples; positive on fOTa sample Psychological Nooe Yes Yes y", Not 5- of of variable (cut (GHQ), index index, well-being score) scale depressive positive negative depressive item GHQ feelings symptoms IGWB) overaU feelings Dependent 18-item II-item Depression 8·itern Depres.'iiOli 5- 5- of of visits (b) of with of status (b) of to nearby, being church, relatives., (b) relatives support measure index index & index index index marital neighborhood friends Neighborhood 6-item 6-item participation married. item community including & belonging neighborhOCJd zations interaction, frequency attending c1ubs/organi- neighbors. interaction cohesion, friends participation to item community (c) Support 8-item (a) (a) Integration (a) of of self- strain, crime), (el scale (d) checklist scale life life measure life chronic Neighborhood events events perception problems, respondenl's illness, evaluation financial status. violent deprivation conditions economic events (social (b) Stress 63-ilem 3D-item Economic (a) 63-itcm Indexes Design Support Cross-sectional Cross-sectional Croswectional Cross-sectional CrOM-sectional & with & of al. Structural 863 E~x 2,029 2 602 et (1981) ~ = eo & LeRoy. Vaillanl Hewson, Tennant, (1978) N Stephenson (1982) N N Parents LorCF;N= (1982) 220 children Illi (1982) Study/sample Community, Community, Cohen Community, (J U Specialized: Community, Global Table Andrews, Bell, P. t"rydman Kessler W -< rn '" N ::t: ::t: iii ." ~ ~ VI t
difficult *"lf functional work is criterion; enters stress igS.2 Is Isample Regression Weak Diffuse Small Physical None None effect Buffering onJy anxiety, for strain depression Psychological No None No Yes for headache only Ph,,;caI None Yes support- of in effect b to all Main main positively for mullivariate related anal_ depression of effect> entered)
Pwchological lfK-8 ' No Yes, y" Yes of of (HRS) scale (SCL- & feelings variable i||| LE morale, health of ratings rating sca.lc
fillpositivel . Ill physical negative tiredness, headache, irritability, Slatus depression 90). severity symptoms backache, loneliness depression. dizziness) (anxiety, symptoms depression/ Dependent Depressive Oinical 24-item Symptom
s & cO- in index in •| .= 'a at rela- of of relatives, marital with with feeJ- or measure neishbors. meetings & with and no. includes includes includes ownership, attendance support support index about network !M«i IlllPIp community ^SIIM clubs membership workers, relati'les, meetings neighbors, index organizations attendance clubs/church friends neighbors, neighoors, includes involvement index contacts co-workers, clubs/community contacts frequency status, home relatives neighborhood, the talkiT1f!i il18S index includes lives. empjoyment, church friends, friends at Support Illll Diffuse Social Diffuse »9-itr:m Q of SM'essors parental checldist stressors, checklist life life measure strain event measure major mmor interview homemaker events role events strain, Stress Life 24-item 14-item InterView for is Design measure) sectional) analysis eros&- (cross- (buffering sectional structural Longitudinal Longitudinal Cros,s.sectional Cross-sectional 100 Kuo N"" yean (continued) & health 80 170 800- Coyne, Ingham 2 (1980) il Laza",,, & No: inpatients), 1,060 (depressed oId),N~ center), from & (1981) (45-64 (1979) (recruited N'" N~ Americans), 2,271 Ensel. (1979) (Chinese- Simeone. inlill Community Oinical Community Community Study/sample Surtees Schaefer, Miller Lin, Table t: ,," ~ <:>- ;;;- ~ (table~ continued) 326 SHELDON COHEN AND THOMAS ASHBY WILLS
main effects for support without interactions when analyzed with continuous outcome measures in community samples. This was found by Lin, Simeone, Ensel, and Kuo (1979) for a scale of neighborhood and community involvement, by Frydman (1981) for a scale indexing community participation, and by Bell, LeRoy, and Stephenson (1982) for a composite scale that included questions about contacts with relatives, neighbors, friends, and community participation. A main effect of a similar composite support scale was also found in a prospective study by Williams et al. (19 81). A main effect of support without a Stress X Support interaction is also reported by P. Cohen et al. (1982). Although we felt the com- posite support scale used in this study was pri- marily structural in nature, it included a 6- item subscale that might assess functional support. The subscale indexed the extent to which persons in the neighborhood were per- ceived as friendly and helpful. A regression model that included multiple factors, and the apparent lack of matching between the stress- ors (crime, poverty, and physical illness) and the support measure, may have reduced the likelihood of finding the buffering effect one l|e| Ill might expect if the scale were functional. Data from a study by Schaefer et al. (1981) ^HeS K.aSf •fllll Jli are mixed in regard to evidence of a main effect of support for their structural measure. A so- cial network index including questions about friends, relatives, and community participation was analyzed in a complex multivariate anal- ysis where it was entered in the regression equation with two life events measures and m three functional support measures. On a zero- order basis, the social network measure was unrelated to depression or negative morale but was positively correlated with positive morale. In the regression analysis, the network index showed a positive main effect on depression, but this is probably an artifact of the inter- correlations between the network index and the other variables in the regression model. Failures to find main effects for compound structural measures occurred in Andrews, li* "° § Tennant, Hewson, and Vaillant (1978), where both predictor variables and outcome mea- sures were dichotomized, and Surtees (1980), with a clinical sample and a dichotomized !"!' outcome measure. As previously noted, these I methodological characteristics greatly reduce SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 327 the sensitivity of an analysis for testing either munity organizations typically show main ef- main effect or buffering hypothesis. It is note- fects for support but not buffering effects worthy that Frydman (1981) found a main ef- (Frydman, 1981; Bell et al., 1982; P. Cohen et fect of support for the same community par- al., 1982; Lin et al., 1979; Williams et al., ticipation scales used by Andrews et al., but 1981). In cases where dichotomized predictor with an analysis of continuous predictor vari- and criterion variables were used (Andrews et ables. al., 1978;Surtees, 1980), not even main effects Frydman (1981) reported evidence that is were found, a fact that illustrates the statistical inconsistent in terms of implications for the disadvantage of dichotomous variables. main effect and buffering models. This study Two studies found mixed evidence for a included data from two different samples: par- buffering model. Frydman (1981) found buff- ents of children with cystic fibrosis (CF group) ering effects in one of four analyses, and Kes- and parents of children with leukemia (L sler and Essex (1982) found evidence for buff- group). Here measures were scales from An- ering in one of three analyses. These results drews et al. (1978) for neighborhood interac- may be explicable in terms of a particular tion and community participation. There was global structural measure being associated with a pure buffering effect for neighborhood in- a support function that matched up well with teraction in the CF group but not in the L a particular stressor or sample. group, where a weak main effect was observed. Finally, buffering effects for anxiety and The community participation scale, which in- depression were found by Miller and Ingham cluded items about participation and special (1979) with their diffuse support scale, which, interest groups, showed no buffering effects, in contrast to other measures discussed in this but did show a strong main effect for the L section, is based on an interview procedure in sample only. These results are puzzling and which the existence of friendly relationships is may represent differences in measure vari- specifically determined. As noted, it is likely ability across samples or in the roles that com- that this procedure taps components of func- munity and neighborhood contacts play for tional support that are responsive to the needs persons in different samples. elicited by the stressor. In cases where signifi- Kessler and Essex (1982) reported evidence cant interactions were found (Frydman, 1981; of buffering for one of the three stressors they Miller & Ingham, 1979), they were pure buff- studied. A 9-item integration scale based on ering effects, with virtually no difference be- items such as employment, home ownership, tween supported and unsupported subjects and number of community involvements was under low stress. analyzed with respect to three specific stressors. We conclude that combined information The analysis tested individual cross-product from a variety of social relationships measures terms but did not test the independent main the extent of imbeddedness in a social network, effect contribution of the integration scale. No which is important for overall well-being but buffering effects were found for economic does not reliably index the availability of spe- strain or homemaker strain, but a significant cific support functions that are relevant for interaction was found for parental strain. stress buffering. The observed findings thus are Finally, Miller and Ingham (1979) con- generally consistent with our theoretical pre- structed a support score termed diffuse support dictions. based on answers to interview questions about respondents' contacts with co-workers, neigh- Studies Using Functional Measures bors, and relatives, plus church and social club membership. The interview included probes Within the functional support section, about the existence of friendly relationships studies are organized with respect to the spec- and thus may have tapped functional com- ificity of the support measure because this is ponents. These investigators found significant the most relevant methodological character- interactions (pure buffering effects) for both istic in accounting for differing results across anxiety and depression outcome measures. studies. Separate sections review studies using In sum, global structural measures indexing (a) specific functional scales that measure one connections with friends, neighbors, and com- or more specific aspects of support, and (b) 328 SHELDON COHEN AND THOMAS ASHBY WILLS global functional indexes that assess an undif- protection from severe negative life events. ferentiated mixture of support functions, and Similar results were found by Paykel et al. in a number of cases structural characteristics (1980) in a sample of postpartum mothers. A as well. A third section reviews a set of studies rating of communication with husband about on perceived occupational stress. Unlike other worries and problems showed a significant studies reviewed in this article, this work fo- buffering effect, whereas a measure of confi- cuses on perceptions of stress rather than ob- dants other than the spouse did not show an jective life events and assesses support for a interaction effect (but did have a main effect). particular type of stressor, hence deserving Husaini et al. (1982) used questionnaire items separate attention. As noted previously, our tapping respondents' marital satisfaction (not model predicts that valid measures of esteem marital status) and perception of the spouse or informational support are most likely to as a confidant who was "good at understanding show buffering effects irrespective of the my problems." Buffering interactions were stressor. Measures of instrumental support and found for these items for women, but not for social companionship are predicted to operate men. Warheit's (1979) study is again an ex- as buffers when they match the coping re- ception; a main effect but no buffering inter- quirements elicited by particular stressful action was found for a dichotomous item con- events. cerning "friends to talk with about problems." Here, the methodological characteristics pre- Studies Using Specific Functional Measures viously discussed work against the demonstra- tion of a buffering interaction. Confidant measures. Because a number of Consistent buffering effects are also found studies have assessed whether a person per- with multi-item scales of confidant relation- ceives the availability of a confidant, these ships. This includes Miller and Ingham's studies are treated separately. These studies use (1979) interview measure of having close con- interview procedures or questionnaires to de- fidants to talk with about personal problems, termine whether the respondent has access to Habif and Lahey's (1980) brief scale of con- a confiding relationship with another person. fiding relationships, Kessler and Essex's (1972) We have classified this as a functional measure interview measure of feelings of acceptance because, by definition, the existence of such a and ability to talk about problems with spouse relationship (having a confidant) implies the (or nonspouse confidant for nonmarried sub- availability of esteem and probably informa- jects), and Fleming, Baum, Gisriel, and tional support. Studies using confidant mea- Gatchel's (1982) 6-item scale of availability of sures are summarized in Table 3. As discussed confiding relationships. The latter study found earlier, we expect that esteem and informa- buffering interactions for psychological symp- tional support provide needed resources for toms but not for physiological measures of those under stress and thus are likely to pro- stress. The exception is again the study by Sur- duce buffering effects. tees (1980) in which interview ratings of con- Of the studies listed in Table 3, most pro- fidant support showed a main effect but no vided evidence of buffering effects, and the interaction effect when a logistic regression finding of such effects does not seem greatly analysis was used. Here the use of a clinical dependent on the measurement procedure sample and the dichotomization of predictor used to index confidant relationships. Four and criterion variables greatly reduces the sen- studies used either interview procedures sitivity of the design for testing interaction ef- (Brown, Bhrolchain, & Harris, 1975; Paykel, fects. Emms, Fletcher, & Rassaby, 1980) or ques- Buffering effects were observed in cross-sec- tionnaire items (Husaini et al., 1982; Warheit, tional and longitudinal analyses of data from 1979) to determine whether or not a confidant a community sample in Australia (Henderson, relationship existed. Brown et al. (1975) found 1981; Henderson, Byrne, Duncan-Jones, Scott, buffering effects in a sample of women that & Adcock, 1980). In this study an interview were accounted for primarily by their confiding procedure provided an index termed attach- relationship with their husband or boyfriend: ment, indicating the availability of relation- confidants other than these failed to provide ships that provide esteem support (e.g., some- SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 329
one to lean on, someone who feels very close because the support measure, termed strong to respondent, and someone to confide in, to tie support, was based on items assessing the share one's most private feelings with). This frequency of problems the respondent had with procedure provided measures of both the not having enough close companions or close availability of such relationships and their ad- friends. This approach raises questions about equacy as perceived by the respondent. Cross- construct validity because it seems more likely sectional analyses indicated significant buff- to measure social isolation or social skills than ering effects only for women, and primarily to index available functional support. Dean for adequacy of confidant support rather than and Ensel (1982) found a main effect for this availability. For men, only main effects were measure in all subgroups tested but no inter- found. In the longitudinal analysis, the effect actions. Other problems in this study are the of confidant support was tested prospectively, lack of a main effect for stress in two of three using Time 1 support and stress as predictors subgroups tested and a significant correlation of psychological outcomes at Time 2 and con- between the life events and support scales, trolling for Time 1 symptomatology. This pro- suggesting considerable confounding. spective analysis indicated a significant buff- In summary, results from measures of con- ering effect for the total sample, but separate fidant support provide consistent evidence of analyses for men and women were not re- a buffering model of the support process. Of ported. the 13 studies listed in Table 3,10 showed sig- The other confirmatory study is reported nificant Stress X Support buffering interac- by Pearlin, Menaghan, Lieberman, and Mullan tions. Of the remainder, 1 (Warheit, 1979) used (1981), who demonstrated the effect of con- a single-item support measure and a social ex- fidant relationships for buffering the spe- its stress measure that was severely confounded cific stressor of unemployment. A 3-point with support, 1 (Surtees, 1980) used a clinical measure was created by first establishing sample and dichotomized predictors and cri- whether the spouse was perceived as a confi- terion, and 1 (Dean & Ensel, 1982) used a sup- dant, and an additional point was added if port measure whose construct validity is ques- there was a confidant other than the spouse. tionable and additionally found no main effect The regression analysis used a Time 2 support of stress for the majority of the sample. Of the measure to predict a Time 2 criterion variable; 10 confirmative studies, Henderson (1981) relevant demographic variables and Time 1 found a buffering effect using a true prospec- symptomatology were controlled for. Signifi- tive analysis, and Pearlin et al. (1981) found a cant Stress X Support interactions were found buffering effect with a concurrent analysis that for the criterion variables of self-esteem and controlled for demographic variables and Time mastery. For predicting depression, there was 1 symptomatology. In studies where relevant no significant buffering interaction when the data are reported, it is evident that the inter- effects of support on self-esteem and mastery actions represent pure buffering effects, with were partialed out. This suggests that the buff- a large effect of support under high stress but ering effect of confidant support occurs no significant difference between low- and through its influence on feelings of self-esteem high-support subjects under low stress (Flem- and self-efficacy. The interactions that were ing et al., 1982; Henderson, 1981; Henderson, found were slightly crossed; subgroup analyses Byrne, Duncan-Jones, Scott, & Adcock, 1980; (using trichotomized scores) showed a strong Miller & Ingham, 1979; Paykel et al., 1980). relation between stress and symptomatology The exception is Habif and Lahey (1980), for subjects with low support, a slight relation whose support scores were arbitrarily cut so for subjects with average support, and a slight that 88% of subjects were classified as having reverse relation for subjects with high support. high support. Consequently, this study does not The crossing may have occurred because of seem to bear strongly on the general conclu- the use of a highly skewed stress measure; only sion. It is noteworthy that buffering interac- 88 subjects from the large sample were clas- tions were reported in studies that found no sified as having had a job disruption. correlation between the support and stress A final longitudinal study (Dean & Ensel, measures (Habif & Lahey, 1980) or between 1982) differs from other studies in this section (lex! continued on page 332) ~ ~ tr1 tr1 Z -I 0 '" '" :s:: r r r » » :t " w " Z Z 0 0 0 '" '" () :I: » :t :t 33w 0 SHELDON COHEN AND THOMAS ASHBY WILLS only stress for of support support three with round uncorrelated of support of effect Remarks stronger effect and spouse two prospectives main for for measures subgroups compared adequacy availability analysis Buffer No Results Stres;. True Physical None effect
S for Zung; on Buffering Yes and
iL None global
BOI i|il e SO fc for Psychological and symptomatology Men: adequacy GHQ Yo. No Yes y", yd' Women: Physical for norepinephrine support" Yes of effect Main men all"
J women)" far for Si. (for and adequacy (cross- £Ps)'cnoJogicaJ NR y" Yes y" Yes Yes (SCL- (DOl) scale Q , variable y m (DOl). symptom symptom symptom measures (GHQ , (GHQ, (GHQ, v e sym p physioJogica1 interview ncphrine) (CES-D) (caseness) symptoms psychological 90). symptoms symptoms stress (epi/norepi- Zung ) Zung) scae s Zung) scales scales Dependent Depressive Depressive Depression Depressive Depressive I1epressive
8 not of of of or a of with intimate of of
existence 3VS £ •I^'o of mc:a.sure husband, J,l of rating rating dose scale scale Sill! Ill
with 3 111 confiding tie problems boyfriend. having other confiding friends/ relationslJips confidant companions relationships assessing of availability Support relationshps confidant availability/ adequacy availability/ mi III! HI adequacy Existence Frequency 4-item 6-item Interview
5 " " Interview of life event life life areas) ('Is. life near measure S'E S e |= ^ ^
6events S I S « presence ~re interview. nuclear Island 8-item events plant cneeldist Three-Mile other events checklist events Relationships checklist checklist events Life II Stress Living H-item 60-item 71-itcm Confidant Design of interval) (I-year interval) (!-year T_ongitudinal CrOSS-se<'tionai Cross-sectional Cross-sectional Longitudinal Cross-sectional eI Measures & Byrne, Scon, (1981) ) ni t 3]4 871 109 3 2.52 Saum, 75 6 7.56 Lahe), Ensel = '" Harris = Adcock & 1!J7S) & Bhrolchain. & ( (women), ;.l N 1'1/= Gisriei, Gatchel (coJlege students), (1982) {]982} N (1980) Duncan- (1980) ,'11= Henderson Jones, & (!>ameas
Studj-/samp!c in Community Community, Community, Specialized Community, Community Specific Table Brown, Dean Aeming, Habif Henderson, Henderson X tTl -I ~ ~ ~ X ttl Z CO c::: en -< ;
Hi. main for for dear for prospective sectiooal women; only of effects for entered) Psychological Yes Yes Not Yes of scale scale & variable rating of |gl|ratings j depressive symptoms depressive djzzi~) severity backache, tiredness,. headache, irritability, depression, (anxiety. symptoms symptoms (CES-D) II Dependent Interviewer II-item Symptom Dep<=ive at as one (b)
§3 of of other as other (a;ood measure least or support assessing confidant spouse al a I husband spouse spouse communication Marital relationships than availability confident (confidant), confidant with good index availability Support spou~ understanding confidant confidant (b) (c) satisfactioo, satisfaction, I! Hi problems) !!!!] (a) Having Intimacy (a) of for life measure strain negative rafed occurrence/ impact checklist; major minor measure stressors, stresson role strain. parental homemaker checklist events stram, Stress 64-itcm Interview 52·item Economic Design CrOM--SeCtionai Cross-sectional Cross-sectional Cross-sectional = N (continued) 1,060 231 Essex 3 965 Ingham 1980), Neff. Mullan Emms, == & = Moore & 120 Menagban, Lieberman, (1981) & women), Retehco-,8< aI., (postpartum N Rzsaby (1980) /!.'= (1979) N·800- Newbrough, 2,271 8< N (1982) (1982) Study/sample Specialized Community, Community Community, Pearlin, Paykel, Table Miller Kessler I Husaini, 1. t: s· "" ~ 8 ;
(tableS continued) 332 SHELDON COHEN AND THOMAS ASHBY WILLS
social exits and depression (Paykel et al., 1980), II which again argues against an artifactual in- terpretation of buffering effects. II Several important points about this litera- •8s ture should be noted. First, confidant measures 11 II assess the availability of a close, confiding re- lationship with one or more other persons, and it is in such relationships that support func- tions such as self-esteem enhancement and in- formation support are most likely to occur. Second, there is a suggestion of sex differences; women showed more benefit from confidant support. This was found by Husaini et al. (1982) and Henderson etal. (1980). Brown et al. (1975) and Paykel et al. (1980) also found buffering effects of confidant support for women but had no data on men. Third, in two cases (Brown et al., 1975; Paykel et al., 1980) confiding husbands or boyfriends served stress- protective functions for women, but other confidants did not. This may occur because such intimate relations are conducive to the adequate provision of information and esteem support. It is also possible that some aspect of the companionship and sexual relationship, implied when one's spouse is a confidant, op- erates to reduce stress impact. Finally, there is a suggestion that the effects of confidant sup- port are mediated by self-esteem or self-effi- fo! cacy; this was shown by Pearlin et al. (1981) through hierarchical regression analyses and by Husaini et al. (1982), who showed a buff- ering effect only for subjects classified as low in personal competence. Overall these results are wholly consistent with our theoretical analysis. These studies provide strong evidence for a buffering model of the support process and provide some data indicative of psychological mediators of this process, which suggest that confiding relation- 1. ships may counteract stressors by increasing ff! feelings of self-esteem and personal efficacy. The only limitation in this literature is that it is not clear from knowing the existence of a confiding relationship exactly what supportive functions are provided in such a relationship. Measures of individual support functions. The following section reviews studies using measures that index specific, conceptually dis- tinct support functions. As previously noted, buffering effects are predicted to be particularly likely to occur with specific functional mea- sures that match the needs elicited by stressful SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 333
events. Esteem and informational support are An instrumental support scale was used in assumed to provide needed resources for a the Paykel et al. (1980) study of depressive broad range of stressful events, and hence are symptoms in postpartum mothers described expected to match up well with cumulative earlier. For this measure the subject rated the stress measures. Social companionship and degree to which the husband provided help instrumental support are expected to operate with household chores, shopping, and other as buffers only when they are closely matched children since the baby's birth. There was a to a particular stressful event. Because little pure buffering effect for instrumental help attention has focused on functional support from the husband. A noteworthy aspect of this measurement, relatively few studies have tried finding is the specific match between the rated to obtain measures of specific functions, and support function and the needs of the mother. measurement approaches have varied consid- Schaefer et al. (1981) showed no buffering erably, a factor that we think accounts for the effects. In this study an interview procedure rather diverse findings in this area. Studies with provided separate measures for instrumental, specific measures of one or more support esteem, and informational support. Instru- functions are listed in Table 4. mental support was assessed by determining Studies showing consistent buffering effects how often, in nine hypothetical situations, are those by S. Cohen and Hoberman (1983) there was someone whom the respondents and Paykel et al. (1980). In S. Cohen and Hob- could count on to provide instrumental assis- erman's (1983) study, the support measure— tance. For the remaining two measures, the the Interpersonal Support Evaluation List respondents were asked to list their spouse, (ISEL)—was an extensive inventory that pro- close friends, relatives, co-workers, neighbors, vided multiitem scales assessing perceived and supervisors and rate each person on the availability of appraisal (confidant/informa- list on the extent to which they provided in- tional) support, tangible (instrumental) sup- formational and emotional support during the port, self-esteem (esteem) support, and be- last month. Separate scores for informational longing (social companionship). The social and emotional support were calculated by companionship subscale was moderately cor- summing across ratings for each of the target related with the instrumental and esteem sup- persons. Unlike other measures used in the lit- port subscales, but otherwise the scales were erature, this one weighs the availability of sup- relatively independent. The overall ISEL and port by the number of persons from whom it subscales were shown to have adequate internal is available. For example, persons having close consistency and test-retest reliability. Regres- confidant relationships with their spouses but sion analyses indicated that there was a sig- not with others would score relatively low on nificant interaction in predicting depressive these measures. symptomatology for the total ISEL score, which In Schaefer et al.'s (1981) study, data analysis was a pure buffering effect. The interaction for was performed through multiple regression physical symptomatology was similar but analyses in which the three functional support showed a slight crossover effect. Analyses of scales were entered together with a social net- subscales indicated that social companionship, work index and life events measures. The cri- informational, and esteem support showed terion variables were psychological and phys- buffering interactions, and the interactions for ical symptomatology. A main effect of stress the latter two scales were independent of all on psychological outcomes occurred in only 4 other subscale interactions. Interactions for of 12 possible comparisons, and both stress social companionship and informational sup- measures were unrelated to physical health. port were pure buffering effects, whereas the Although some main effects for the support esteem support subscale showed a strong main measures were noted in predicting depression, effect in addition to an interaction. These in- no significant interactions were observed. In vestigators found no correlation between the addition to the problematic nature of the sup- life events measure and the ISEL. Recent rep- port measures, the general absence of a main lications of this work are reported by S. Co- effect for stress, and an analysis that entered hen, Mermelstein, Kamarck, and Hoberman multiple structural and functional terms prior (1985). (text continued on page 336) 334 SHELDON COHEN AND THOMAS ASHBY WILLS
3 as !*!i!; isiii'
-II
II
?g t s& its flllli
5 s i I s I « III!
1 •«•<>; SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 335
*-l i .| 1 1 1 1
3 II •£ 1j> O g £ 1!
III I? I £ E * 336 SHELDON COHEN AND THOMAS ASHBY WILLS to testing the interaction, all seem to work (1983), breakdowns by social class indicated against the demonstration of a buffering in- that the support-well-being relation was sig- teraction. nificant only for upper and middle-class sub- These same functional measures were used jects; for lower class subjects, support and stress in a study of pregnant women by Norbeck and were highly correlated, and the regression Tilden (1983). In this study the informational weight for support was nonsignificant. Addi- and esteem support scales were so highly cor- tional post hoc regroupings of the data by related that they were analyzed as a single Turner and Noh (1983) suggested that a buff- measure. A cross-sectional analysis used sup- ering effect might be detectable for lower class port and stress measures as predictors of de- subjects, but the small cell sizes and the post pressive symptomatology assessed at the third hoc nature of the analysis do not seem to pro- month of pregnancy. This analysis showed a vide very conclusive evidence on this issue. main effect for the esteem support scale but In summary, the results from studies clas- no interaction effect. A longitudinal analysis sified in this section are strongly dependent on used support and stress measures as predictors the particular measures used. The instrument of medical complications scored from medical used by S. Cohen and Hoberman (1983) had records after delivery. There were no significant good psychometric characteristics and com- main effects for support on medical compli- parable measurement procedure for different cations, and stress before pregnancy showed a scales and was specifically designed to measure main effect for only one of four indexes of a broad range of perceived functional support complications. Analyses of stress during preg- availability. This measure showed pure, inde- nancy indicated no main effects of stress or pendent buffering of negative life events in a support, and interactions of stress with support college student sample for the total scores as were nonsignificant for overall complications. well as for esteem support and informational Some inconsistent interactions for instrumen- support (but not for instrumental support). tal support were noted for specific types of The continuity of S. Cohen and Hoberman's complications, but given the number of tests results with findings from studies of confidant performed—208 separate tests in the various relationships reviewed in the previous section discriminant analyses—these could easily be is noteworthy, suggesting that esteem and in- chance results. formational support are important elements Studies by Turner (1981) and colleagues in confidant relationships. Paykel et al.'s (1980) used a vignette technique for assessing support. study found evidence for a pure buffering effect Subjects were presented with seven sets of de- of instrumental support that was well matched scriptions of persons with varying levels of to the stressful event (pregnancy) under study. support and were asked to rate their own sim- In S. Cohen and Hoberman's (1983) and Pay- ilarity to the persons in the vignettes. This ap- kel et al.'s (1980) studies, the stress and support proach differs considerably from other mea- measures were uncorrelated, so again a social sures that tap the availability of functional exit interpretation of the interaction results is support and introduces questions about con- ruled out. The failure of Schaefer et al. (1981), struct validity. It seems likely that affirming Norbeck and Tilden (1983), and Turner (1981) that one is similar to persons who are popular to find buffering effects seems attributable to and well-liked reflects stable personality char- methodological problems, especially in regard acteristics such as social competence or self- to measurement of functional support. image rather than perceived availability of As in other areas, findings concerning phys- functional support. In fact this measure con- ical symptomatology are less consistent, pos- sistently showed strong main effects for pre- sibly because of the diversity of criterion vari- dicting depressive symptomatology in several ables used. Norbeck and Tilden (1983) found different samples, including a longitudinal no effect of support on pregnancy complica- analysis that predicted Time 2 well-being from tions, and Schaefer et al. (1981) found no effect Time 1 support with control for Time 1 sup- of support on number of chronic physical port and well-being. There were, however, no conditions, although this seems a rather ex- significant interaction effects. In subsequent treme measure of physical symptomatology. analyses of the same data by Turner and Noh As noted, both of these studies suffer from SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 337
methodological problems that would work The measure social integration, used by against the demonstration of a buffering effect. Henderson (1981; Henderson et al., 1981, Buffering was suggested by the physical symp- 1980), is somewhat difficult to classify but toms measure used in S. Cohen and Hober- seems most similar to a complex functional man's (1983) study, but these measures con- measure.5 This index was derived from inter- sistently produce crossover interactions; there view questions asking about persons who pro- was a slight relative elevation of reported vided relationships with a "sense of social in- symptomatology among persons with low tegration, reassurance of personal worth . . . stress and high support. As yet this crossover a sense of reliable alliance, and the obtaining effect has not been adequately explained by of help and guidance" (Henderson et al., 1980, any investigator. p. 576). This description suggests that the measure is indexing a combination of esteem, Studies Using Global Functional Measures informational, instrumental, and social com- panionship support functions. In cross-sec- Internally Consistent Compound tional analyses (Henderson et al., 1980), this Functional Measures measure showed a significant interaction—a Several studies used apparently complex pure buffering effect—for men; for women functional measures that in fact prove to have only a strong main effect was noted. Prospec- high internal consistency or which have been tive analyses (Henderson, 1981) replicated the construed by the authors as measuring an in- cross-sectional result, indicating a pure buff- dividual support function even though they ering effect of Time 1 stress and support on appear compound in nature (see Table 5). Time 2 symptomatology controlling for Time These studies are difficult to categorize in 1 symptomatology, and showed that adequacy terms of our theoretical perspective because of support was more important than avail- they appear to be global but have adequate ability. This analysis was performed for the internal consistencies. If we assume that these entire sample, and sex differences were not an- measures are tapping unitary elements of sup- alyzed. port, whether they operate as buffers should Complex results were obtained by Monroe depend on the extent to which these scales re- (1983) in a work site sample using a 4-item flect broadly useful coping functions such as measure intended to assess marital or signifi- esteem and informational support. cant other emotional support. Although items Perhaps the clearest example of an internally for this scale were not reported, it had adequate consistent compound measure is that used by internal consistency (alpha = .77), and so it Wilcox (1981). This investigator derived a total seems reasonable to classify it as a compound functional support score from respondents' functional measure. The scale was highly cor- answers to an 18-item checklist assessing related with respondents' ratings of general whether they had support available (from any- satisfaction with their relationship over the past one) during periods of stress. Support was as- 2 years. (A single-item measure of crisis sup- sessed for each of three functional categories port was also obtained but was not discussed (esteem, instrumental, and informational in the article because it was unrelated to any support), but for analysis the subscales were of the outcome measures.) Prospective analyses summed to provide a single score for func- using depressive symptomatology as the cri- tional support. The overall scale had a high terion variable showed no interaction effects, level of internal consistency (alpha = .92). but interpretation of this result is qualified be- With measures of depression and anxiety/ten- cause there was no main effect for the stress sion as criterion variables, the functional sup- measure on depressive symptomatology (see port score showed significant interactions, Monroe, 1982). Prospective analyses for a which were pure buffering effects. A support physical symptom measure did show signifi- index based on the total number of persons (text continued on page 340) indicated by the respondent as being available to provide any of the three types of functional 5 This is not a measure of "social integration" as we support during periods of stress also resulted have denned it (i.e., global structural), but rather a term in a pure buffering effect. developed independently by the author. ~ t= t"" ~ Z t"Il Z t"" til Z -I 81 ::t: ::t: '" :.- :.- :.- ;t 8 8 t"Il 00 " 0 '-" '" ::t: '" 33'-" 8 SHELDON COHEN AND THOMAS ASHBY WILLS for are main with with rather (no stronger ilVailability events events; Remarks prospective support poSitively positively buffering than correlated life interactions crossed life adequacy effect) of analysis conelated Results ISSB True ISSB Physical No No effect yes Buffering only d: for c, men e b, a. (for for (positive Psychological for only)b depression correlation) No No Yes No y" e; C, (positive b, d a, for Physical for (positive no support' corre1ation) correlation) No :'>Jo No of effect yes e; men Main tI yes a, b, c; for (for women) for
{positive i I b for fur (positi\'e carrdation) correlation) adequacy & •UT! J (prospective) Psychological No No No Ye~ Yes & Zung) Zung)
-. a(BSI) variable (3(j Ed physical symptom symptom symptoms £•£ symp- depression. (OHO. sS(GHQ. global somatization, tomatology items) (CES-D). symptoms scales scales Dependent • id i! Anxiety, Depressive Depressive Depressive (d) of of (a)
on I « °f J! « °f Consistency for size, based based worth size, size worth need, measure pen;ons integration integratioo of (based conflicted total support reliable jilflj reliable support unconflicted network no. network network 6 functions): (b) available (e) (e) support satisfaction, providing peI'3Onal providing relationships measure reassurance measure relationships on on personal slisitsi reassurance Support 'SSB ASSIS Social Social .ss. Internal
High . I
> S S B 3 life life life measure events With
negative HI victimization life &S·item =1, checkJist checklist checklist (CSLES) events events checklist ilevents l Stress 144-item 27-item 71-item 71-item Measures Design Cross-~tional Longitudinal Cross-sectional Cross-sectionaJ Functional et 86 years le:.s '" Byrne, Scott, as (1981) 20 i ° >,N 5 756 & 64 •- B " £ ° 231 (1981) = I S^ Adcock women (pregnant than old). Hoberman Duncan- (1983) JIj- (college students), Henderson & ('980) N Jones. (same N= jjij; al..1980), Study/sample Specialized Cohen Community Community, Community Table Compound Barrera S. Henderson, Henderson SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 339 of prior low for but I Ji College of includes tor = effect persons effect inde X on with support high Remarks main ambiguous. CSLES prospective stress buffers Hardiness depression; occurs symptomatology pendent with hardiness those hardiness control increases analysis symproms Stress is No Family Buffering . effects Inventory. State main such Symptom mofMoro Physica1 of (prospective analysis)
IIBrief ! Yes etrecl Profile = = meaning No' v,,' BSI the ror roMS Buffering only that locus Schedule. both (crossed control (or (prospective fn\'eIltory. Psychological interaction) inlmlal of analysis) g .S S -3 assumed No is Yes None it Interview Anxiety noted, Trait Support State Social = utherwise: in Ph,,;caI (concurrent); yes prospective analysis support- STAt No of Arizona Unless No' No' = etTect rnterview. ASSIS ... reported. Main both"
"itnot in ; S e in for no prospect Research concurrent analysis; analysis are Ps:yt'hological Behaviors. Yes Yes None means cell i •- groups. Supportive Epidemiologic of mood checkJis.t variable I-l! symptoms symptoms because support physical anxiety Survey) SociaUy Mil & « S 5 ! (roMS) items) iS\l or Psychiatric psychologica1 (GHQ), (STAJ) symptoms (83 (DDI), Illness (Langner), state (Seriousness symptoms between = Dependent If I Depressive Depressive symptoms Depr~ve Physical ambiguous is PERI JilInventory l == difference r of - Scale. ISSB no as staff which 8 (4 the. is interaction to measure rin on support total items for support ""'" informa support esteem. there based perceive feelings supportive measuring helpfulness measu and cltpression extent support subscales superior.; ""pia,.., perceived tional measure Family Environment items) instrumental. of and support score Scale !medon •3 3 S.S availability Inventory. Suppon IS-item Boss Family Relationship ISS8 ofDepressioD significant stress of low Study Depression life life under life presence measure life Beck the '" effect; checklist events che<:k.1ist checklist checklist (PERI) events ."."ts events in Stress 102-item Ill-item ?-item 6O-item Epidemiologic iliHOI l symptoms. buffer ror jj tS *£effect -° pure (4. main a physical Center is the • = and interval) of Questionnaire. Design effect mo. Cl:'S-D Longitudinal Cross-sectional Cross.sectional Cross-sectional 1 Health .S meaning Scale. the psycilological interaction General the Event = (continued) Puccetti studies Lakey .. 170 93 320
75 L Is 5 case Life (1983) & & (1981) = large combined GHQ executives. (1983) N- N~ (college students), N N~ corporation, of (1982) many this Rusines. Employees Community, Study/sample Community In In Scale Note. Student C Kobasa b • Wilcox Monroe Table Sandler 340 SHELDON COHEN AND THOMAS ASHBY WILLS cant buffering interactions. The interactions haviors (ISSB; Barrera, Sandier, & Ramsay, were found not only for undesirable life events, 1981), a 40-item inventory that presents re- but also for neutral ambiguous events and de- spondents with a list of transactions in which sirable events. support was given, and asks them to rate each In a study of a sample of middle-aged male one for how often it had occurred during the executives, Kobasa and Puccetti (1983) used past month. Although the ISSB appears to in- two support scales that can be classified as clude supportive behaviors representing a wide functional although neither appears to index variety of support functions, it has high inter- specific functions. First, a "family support" nal consistency (alpha = .93). The conceptual composite was created from the cohesion and difficulty with this measure is that it confounds expressiveness subscales of Moos's Family En- the availability of support with the need for vironment Inventory; we classify this as inter- and use of support. In fact, results obtained nally consistent because Billings and Moos with this measure are inconsistent with those (1981) reported an alpha of 0.89 for a similar obtained with other measures. Barrera (1981) composite. This measure indexes the extent to found that the ISSB was positively correlated which family members are perceived as gen- with a life events measure and positively re- erally supportive and expressive of their feel- lated to symptomatology. There were no in- ings. Second, a "boss support" measure was teraction effects. Sandier and Lakey (1982) derived from an analogous inventory of work found significant Stress X ISSB interactions for environments; this measure taps the extent to subjects high (but not low) on internal locus which work organizational superiors are per- of control, but these were crossed interactions, ceived as supportive of employees. A buffering and social support was associated with lower interaction was found for the boss support levels of symptomatology for high-stress per- measure, probably attributable to the fact that sons and higher levels of symptomatology for the stress scale was strongly weighted by work- low-stress persons. S. Cohen and Hoberman related life events, so a reasonable matching (1983), who included the ISSB in their study, of stress and support measures was obtained. found that it was positively correlated with life For the family support composite, there was events measures and noted significant inter- no main effect for support and no Stress X actions that were not consistent with the buff- Family Support interaction. Complex inter- ering model, but that instead reflected a neg- actions were found with a measure of person- ative relation between the ISSB and depressive ality hardiness—an amalgam of three separate symptomatology under low stress, but not un- traits: control, commitment, and challenge. A der high stress. These results suggest that past buffering interaction with some crossing was use measures may to some degree reflect psy- found for the high-hardiness group. However, chological distress, which leads to increased for the low-hardiness group, the family support use of support. was positively related to symptomatology at The remaining compound functional sup- all levels of stress. The complex results for this port measure is the Arizona Social Support measure may be partly attributable to the gen- Interview Schedule (ASSIS; Barrera, 1981). In der composition of the sample, because several this interview method, the respondents are studies (Billings & Moos, 1982; Holahan & asked about persons who could provide six Moos, 1981) found that family support showed types of support functions: material aid, phys- beneficial main effects primarily for women ical assistance, intimate interaction, guidance, but not for men. Also, it is possible that some feedback, and social participation. The scoring element of the support measure (e.g., emo- procedure provides variables indexing total tional expressiveness) was differentially rele- network size, the number of persons who have vant for individuals with different personality supplied support during the past month, the characteristics. number of network members who are sources Studies using support instruments assessing of interpersonal conflict, and satisfaction with support received in the recent past rather than support received in the six functional areas. the perceived availability of support have found Analyses by Barrera (1981) for this measure rather different results. Several studies have indicated that the score for conflictual network used the Inventory of Socially Supportive Be- members was positively correlated with symp- SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 341
tomatology, and only the index for satisfaction buffering effect. The lack of a main effect of with support showed the typical beneficial stress on depression may derive from a rela- main effect. In performing interaction tests for tively selected sample (only 17% of the solicited the various support indexes, Barrera (1981) group participated), which contrasts with the noted significant buffering interactions for the representative community samples used in total number of nonconflictual support per- most other studies. sons in predicting the criterion variable of Complex-global indexes of functional sup- depression, although, given the considerable port. The studies in this section either ask number of statistical tests performed, it is pos- global questions about available functional sible that this is a chance result. support or combine a variety of functional In summary, work with compound but in- (and sometimes structural) support items into ternally consistent functional measures pro- single indexes without evidence of internal vides evidence for the buffering model when consistency. It appears that they combine items measures assess perceived availability of sup- assessing very different aspects of social sup- port, but not when they assess use of support port. A summary of studies reviewed in this in the recent past. Studies by Wilcox (1981) section is found in Table 6. Because these and Henderson (1981; Henderson et al., 1980) measures lack the specificity of other func- provide strong evidence of pure buffering ef- tional measures, we predict that they will not fects; there were no differences in symptom- consistently show buffering interactions. In- atology for low- and high-support subjects un- teractions may occur for those indexes heavily der low stress. The buffering effects found in weighted by esteem or informational support these studies are probably attributable to in- items, and main effects without interactions ternally consistent support scales tapping a may occur for indexes that include structural broadly useful support function (e.g., esteem items. or informational support). Kobasa and Puc- Single-item measures were used in several cetti's (1983) study of male executives found studies. Warheit (1979) included a dichoto- that support from supervisors resulted in a mous item about whether there were "close buffering interaction with a stress scale that friends nearby to help" (what kind of help was primarily assessed work stress, although sup- not specified). A significant main effect was port from family only buffered persons high found for this item and, although the pattern- in hardiness. The relative effectiveness of su- ing of means was consistent with a buffering pervisor support in this context is probably effect, the interaction was not significant prob- attributable to the match between stressful ably because of the low reliability of such a events and support functions provided in the measure. Husaini et al. (1982) included work setting. Studies using the Inventory of items concerning the frequency with which the Socially Supportive Behaviors (Barrera, 1981; respondent called on relatives or friends for Cohen & Hoberman, 1983; Sandier & Lakey, help "when you have a real problem" (again, 1982)—a measure of support received during the kind of problem was unspecified). The item the past month—are not supportive of buff- for friends produced a significant interaction ering. The failure of the ISSB to find a buffering effect but it was a reverse interaction (i.e., effect, we think, is at least partly due to asking greater symptomatology among persons with about support received in the past. We argue a high support score). It is likely that this mea- that the ISSB confounds the availability of sup- sure reflects recent use (rather than availability) port functions and the recent need for support of support, comparable with the ISSB scores (i.e., stress), an argument that is supported by discussed previously. the positive correlation between life events and The remaining studies in this group used the ISSB (Barrera, 1981; Cohen & Hoberman, multiple-item composite scales that combine 1983). questions about various aspects of support into Monroe (1983) did find a prospective buff- a single score. Andrews et al. (1978) used a 5- ering effect in the case of physical but not de- item index termed crisis support, including pressive symptomatology. In contrast to most questions such as "In an emergency do you other studies, there were no main effects of have friends/neighbors who would look after stress on depression and, correspondingly, no (text continued on page 344) tn Z :I: Vl Z § n :I: tn 0 N ..,. .., » Vl Z O s:: » :c ~ 0 I.H :I: Vl » ~ t"" t"" 342 SHELDON COHEN AND THOMAS ASHBY WILLS '" do for of of the who only data, work effect Remarks predictors, ~us measure married criterion stress not women measures home analysis dichotomi2Cd outside 1continu_ i Dichotomoos Categorical Main Physic.al effect Bulfering who the work *****for not only married home women do outside Psychological No Psy c Yes None None Pnysical support· of elfect only both onl:y CF Main for for for married men sample Psychological y., Yes Yes Yes * of scale scale (cut overnI) score) variable Dependent GHQ (CES-D) symptom well-being (PERI) symptoms (GWB) (GHQ), Depression Depressive Depressive p! Depression of to (b) of of call & crisis talK close & who support index measure help, mo. private of support friends, of fricnds, support index 2 index composite can for instrumental No. from neighbors. relatives crisis relatiy~ you matters friends 6-item past helpdUIing neiahbors, on index about from esteem SUl>'POrt, relatives crisis confidant & support Support S·item (a) 5-item S-item kiss &. (b) index. scale life on life life mea.o;ure : 2 3 12-item based events scale strain marital, work-related financial, events, strain perceived checldist events m events Stress IS-item 63-item (a) 63-item
IIndexes Design Support Cross-scctionaJ Cross-sectional Cross-sectional Cross-sectional or
& S3 220 L & ~ (1982) 863 1,000 (,026 6 (1981) N ~ & 8 8 S " Hewson. Tennant, Vaillant (1978) N= N~ Stone Mechanic children N (1983) parentsrn Isil CF, Study/sample
Community, ~ Community, Community, Specialized: Glohal Functional Table Andrews, Aneshensel Frydman Cleary ~ ~ ~ ~ ~ t'l1 t'l1 Z Z on >- c:: Cl '"d '"d 5!l 0 >-I :: £be Scale. found 1 may effect marital Remarks i ambiguous. measure measure status utilization support for WelJ_~ing fit is Variable Single-item Main effects General = main & GWB such of for Physical 3l illness, depression cholesterol, «ijje " l symptoms 11 Yes No effect il Interview. meaning the Bufferin g effec t Buffering Research that lreverse t e!foe! _%g;ca/ Psychologica l Physica No No No No, assumed is it Epidemiology noted, Psychiatric for = Physical cholesterol otherwise No Yes support' PERI iof I Unless Scale. effect n men 1Main reported. for 1 rorrelation) (positive Depression only iS !*i not Psychological No Yes y" zNo of arc Study means 13- of of groups. scale count scale health illness depre;sive scale depressive diary) index variable because cell support Epidemioiogic Dependent 5 from physicaJ physica1 item (Langner), (symptom symptoms., symptoms symptoms., serum depressive cho1esterol symptomatology symptoms (CES-D) for I8-item Depressive 26-itern 11 P20-item between Center ambiguous = ii .=is S of of and of help ogiS for of are difference with outside -» to areas friends, on measure CES-D scale no i spouse, of 111! 6 home; friends is 111; inde:.: interaction nearby help relatives; the from friends. supportiveness perceived relationships relatives over discussions called confidant frequency activity an which i llhelpfulness i fibrosis. 13-item there Rating Close SupJXH1 ! £Frequency significant cystic stress = of low CF life social life b!l '§§a .2 ^ "S events measure under EC? promptly j presence .S IJ-g reemployed loss checklist vs. checklist events cbecklist i*i even" III the leukemia. 27-item Unemployed 6O-item 50-item Stresll £ jjleffect; in « = L bulfer effect (1- 13 (6- pure main a J H/ear. Is is interval) ofthc Design t Questionnaire. effect 2-year analysis) (cross- sectional buffering year month, interval) Longitudinal Longitudinal Longitudinal UCross-sectional Health meaning the interaction S«General 5 & the faced = (continued) family collar studies 100 517 428 job (1983) 965 6 Neff, case (1979) ~ Moore ?!i GHQ (1978) physician Roy N Norman, practice), workers termination), with l'ol= (blUe N"" Streiner, Newbrough, & (from N- (1982) many this Community, Community Specialized Community, Study/sampl e Study/sample In Nole. Warheit -In b McFarlane, Gore £ Husaini, Table 344 SHELDON COHEN AND THOMAS ASHBY WILLS your family for a week?" and "If everything quasi-functional measure asked about the went badly, how many people could you turn number of close friends and relatives the re- to for comfort and support?" This seems close spondent had who "you can feel at ease with, to being a structural measure, but we have can talk with about private matters, and can classified it as functional because it asks about call on for help." The other index was con- undefined help and support that could be in- structed from six Likert-type items asking how terpreted by respondents as functional support. often during the past 2 months someone had In this study, there was a main effect for this provided them with either of two types of measure on psychological symptomatology. functional support (emotional and instru- Although frequencies of impairment were mental). Two stress indexes were used, one consistent with the buffering hypothesis, there based on a list of 12 social loss events, and the was no significant interaction. As previously other on 3 measures of "strain" (perceived noted, the use of dichotomized predictors and stress). Log-linear analysis based on dichoto- criterion variable work against the demonstra- mized and trichotomized variables indicated tion of an interaction effect in this study. The main effects for both stress and support mea- same measure was used by Frydman (1981) sures, but no interactions. There was a strong with two samples of parents. Here, a significant relation between perceived strain and the sup- main effect was found for parents of children port measures, although not between the social with cystic fibrosis; although the patterning of loss events and support measures. Here, the means was consistent with a buffering effect, confounding of stress and support, and the re- the interaction was not significant. For parents duction of continuous variables to categorical of children with leukemia, no significant effects scores, severely weaken the sensitivity of this of any kind were noted. Again, the results sug- study for detecting interaction effects. gest that the support needs of the two parent Two longitudinal studies using complex groups may have been quite different. functional indexes are those reported by Gore In Cleary and Mechanic's (1983) study, the (1978) and McFarlane, Norman, Streiner, and support measure was based on five items rep- Roy (1983). Gore (1978) analyzed data on un- resenting a composite of crisis support (e.g., employed male factory workers originally col- "If you had a serious problem, is there some- lected by Cobb and Kasl (1977), who obtained one you would be willing to wake up in the information on support and health outcomes middle of the night to talk with?") and confi- prior to and subsequent to unemployment. dant support (e.g., "Is there someone with Support was indexed by a 13-item scale cov- whom you can discuss almost any problem?"). ering perceived supportiveness from wife, The complexity of the scale is indicated by a friends, and relatives, frequency of activity low internal consistency coefficient (alpha = outside the home, and an index of confidant .43). Analyses conducted separately for three relationships. In this study stress was inferred subsamples (employed, married men; em- from the imminence and actual experience of ployed, married women; and married women unemployment, and subsequent stress levels who did not work outside the home) showed were inferred from continued unemployment a weak main effect of support that was of com- versus reemployment. Gore (1978) found that parable absolute magnitude across the three for subjects in the low-support category (lowest groups but was statistically significant for only one third of support scores), serum cholesterol, the group of employed, married men. A buff- physical illness symptoms, and depression ering interaction was found for the last group, tended to remain high over the 2-year follow- the only group with a significant main effect up interval, whereas for unemployed but sup- for life events. The complexity of the composite ported men levels of symptomatology tended support scale is a noteworthy aspect of this to decline or remain stable, although means study, and it may be that some component of for depression suggest a main effect for support this scale (possibly confidant support) was rather than a buffering effect. There was no more salient for one subgroup than for the statistical test for interaction and, because the other subgroups in the community sample. numbers of subjects in some of the post hoc Two complex-global support indexes were groups are rather small, this study provides used by Aneshensel and Stone (1982). One only suggestive support for a buffering model. SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 345 Subsequent analyses of these data have used also suggest group differences in support needs, slightly different groupings of the data. House although no buffering was indicated for this (1981, pp. 64-70) concluded that buffering ef- measure. Results from a study of unemployed fects were observable for some symptoms such workers (analyzed differently by Gore, 1978; as depression and rheumatoid arthritis. Kasl House, 1981; Kasl & Cobb, 1982) using a and Cobb (1982) noted that different results compound index of confidant and instrumen- occurred for subjects in urban versus rural set- tal support are suggestive of buffering for both tings, and that there was differential patterning psychological and physical symptomatology. In of results by phase of study for variables such this case, however, the complexity of the field as serum cholesterol, diastolic blood pressure, study setting, the richness of the data base, and serum uric acid, and anxiety/tension. These the variety of approaches to analysis resulted authors noted that the complex interactions in findings that do not provide a consistent between support, environmental setting, and picture of buffering. A measure indexing fre- phase of study made causal interpretations dif- quency of recent support use (Husaini et al., ficult. 1982) produced a Stress X Support interaction In a longitudinal study, McFarlane et al. that was inconsistent with the buffering hy- (1983) assessed the average rated helpfulness pothesis. This result (which contrasts with of persons with whom the respondent had dis- those from the confidant measures used by cussions about life stress in any of six areas Husaini) further illustrates the conceptual dif- (work, economic, family, personal, social, or ficulty with past support measures. Single-item societal stress). There was no main effect for global measures (Warheit, 1979) appear to suf- this measure or any interaction for either psy- fer from low statistical reliability, and several chological or physical symptomatology. In failures to find buffering effects with com- contrast, the investigators noted that marital pound functional measures (Aneshensel & status did have a main effect on depressive Stone, 1982; Andrews et al., 1978) may be at- symptomatology (although they did not per- tributable to statistical considerations, because form a buffering analysis for this measure). The these investigators used a dichotomous crite- failure of McFarlane et al.'s (1983) support in- rion and dichotomized all predictor variables. dex to show even a main effect, in contrast to This evidence as a whole seems to exemplify the marital status measure, suggests a defi- the principle that buffering effects are detected ciency in the construct validity of the measure. primarily when functional support measures In sum, consistent with our prediction, re- provide at least a rough match with the needs sults on buffering effects from complex mea- elicited by particular types of life stress. sures that include various indexes of functional Complex-global indexes of functional sup- and structural support are inconsistent. Sig- port: Studies of perceived occupational stress. nificant buffering effects are not typically As noted earlier, a series of studies using com- found, although in some studies the pattern of plex-global functional measures are different results is consistent with buffering. We attrib- enough in character from those described pre- ute this to the fact that the measures include viously to deserve separate attention. These a diversity of indexes that in many cases have studies differ on two dimensions: First, they no clear relation to the stressors to be buffered. address the role of support processes in buff- Cleary and Mechanic (1983) found a buffering ering a particular stressor—occupational effect for a compound index of confidant and stress; second, they all use measures of per- crisis support in only one of three subgroups, ceived stress as opposed to objective measures possibly attributable to the fact that there was (e.g., the occurrence of life events). no main effect of stress for the other two Studies on buffering of perceived occupa- subgroups and possibly also to the matching tional stress are summarized in Table 7. The of stress and support needs in that group (the term perceived stress as used in this literature group in question was composed of women, refers to variables such as role conflict, work who seem to be more responsive to confidant overload, poor communication between work- support). Results from Frydman (1981), where ers and supervisors, unclear organizational a complex crisis support scale produced a main goals, and job future uncertainty. There is effect in one subsample but not in another, some ambiguity in this literature because of r '" :E » Vl ~ r tn >- Z Z F i:: 0 0 ..., l: l: l: Vl Vl tn >- Z 0 0 0 I"') 346 SHELDON COHEl: N AND THOMAS ASHBY WILLS for effect Remarks main ambiguous stress is No effects b a, main for for such Physical VI5 Yes Non, of effect meaning of yes & de- yes the BuH"ering on on c a 5 g S a % ' b con that for for f t 1 » pressioll irritation; fex- depression anxiety; for symptoms Psychological Yes Non, Yes assumed is it all for noted, Physical y" Yes None support- of otherwise for effect only Unless all b Main b) for for a, (stronger PsychologicaJ Yo. Yes Ye~ reported. not are to (4- (10 visits means variable irritation anxiety checklist} somatic health depression (no. physical or self-esteem items), items). items), physical illness infirmary) symptomatology, symptom complaints items) index, (3 (ri (4 because cell Dependent Depressive 4-item Indexes ambiguous co- is & (b) open home fr:i~nds (b) of co- (c) (c) super- measure reJations work Cd) support support leader, (b) (a) of (a) (a) assessing terms relatives 1 » interaction visor, workers, workers, instrumental oooperativene&s, from & esteem from friendliness, SJXluse, perceived with communication) (in supervisor, group co-workers Support Perceived Index Quality significant of .• (e.g Hie!. stress work job job (role load, over- goal con measure job conflict) future of presence Stress measure work rok conlhct, overload, stress load, ambiguity) (role work dissatisfaction ambiguity) job tli the Stress 7~item Perceived Index in effect main the Occupational Design of Cross-sectional Cross-sectional Cross-sectional meaning the '" plant. = N Jones o(Perceived House, Navy 23 workers N studies & 2.800" 7 Wells wOlken; - large ~ 1 § French & IV in (1978) from occupational & (1980) 3.125 men, groups, (l1}78) 636 many Rocco, Rocco " Uourly Study/sample Male enlisted In Studies House Table La La • SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 347 potential redundancy between the stress mea- each of three sources: supervisor, co-workers, sure and the support measure. For example, and wife/family/friends. Dependent variables workers with an objectively high level of job included both job-related and health-related stress and a high level of work-related support outcomes. Regression analyses indicated sig- may report less perceived stress than similarly nificant buffering effects for home, co-worker, stressed workers with lower levels of support. and supervisor support. Overall, the results This potential confounding is mitigated when showed work-related sources of support to be studies include measures of support from more important than home support probably nonwork sources (e.g., spouses). because the stress measures used in this study The study by House and Wells (1978) in- were highly specific to the work setting. cluded seven indicators of occupational stress In summary, methodological characteristics and a social support measure that assessed the in studies of occupational stress suggest that male workers' perceptions of support from four the studies vary considerably in their sensitivity different sources; supervisors, wives, co-work- as tests of the buffering hypothesis. The find- ers, and friends and relatives. Limitations of ings of significant interactions for psychological space do not allow a detailed presentation of symptoms by House and Wells (1978) and La the results, because tests of main effects and Rocco et al. (1980) provide confirmation for interactions were performed for many sources a buffering model, and the latter study seems of stress and support. Overall the data are con- to provide an exemplary test because of the sistent with a buffering model. Although sig- use of continuous predictor and outcome vari- nificant Support X Occupational Stress inter- ables and its careful consideration of the ex- actions did not occur in all cases, there were periment-wide error rate. The negative finding many more significant interactions than would by La Rocco and Jones (1978) appears to de- be expected by chance. The buffering effects rive from insensitive measures, which failed to were primarily due to the effects of supervisor show a main effect for stress or support on support on physical symptoms and wife sup- relevant outcomes. The studies differed some- port on depressive symptoms. what as to whether work-based support is more La Rocco and Jones (1978) studied enlisted relevant than family-based support. This Navy men (mean age = 24 years) aboard 20 probably derives from differences in the stress ships, obtaining separate measures of support measures, and we suggest that more general from work group leaders and co-workers. A indexes of stress would show effects for both diverse set of dependent variables included job home and work support, whereas measures of satisfaction, self-esteem, and a physical illness job-specific stressors would show effects pri- score based on number of visits to the ship's marily for work-based supports (cf. S. Cohen infirmary over an 8-month period. There were & McKay, 1984). Overall, however, the liter- main effects of an occupational stress measure ature on occupational stress does provide con- for job-related outcomes (e.g., job satisfaction) siderable support for the buffering model. but not for self-esteem or physical illness. Similarly, the occupational support measures Discussion showed main effects for only job-related, not health-related, outcomes. No significant inter- The purpose of this article is to determine actions were found. In this study the lack of whether the association between social support main effects of stress for self-esteem and illness and well-being is more attributable to an over- variables suggests that the design was not sen- all beneficial effect of support (main effect sitive for detecting buffering effects for health- model) or to a process of support protecting related outcomes. persons from potentially pathogenic effects of La Rocco, House, and French (1980) ana- stressful events (buffering model). Our review lyzed data from a pool of over 6,000 male re- concludes that there is evidence consistent with spondents from 23 occupational groups in a both models. Evidence for a buffering model number of different organizations. A func- is found when the social support measure as- tional support score was based on the sum of sesses interpersonal resources that are respon- four-item scales assessing the availability of es- sive to the needs elicited by stressful events. teem support and instrumental support from Evidence for a main effect model is found 348 SHELDON COHEN AND THOMAS ASHBY WILLS when the support measure assesses a person's fects of support on well-being in studies using degree of integration in a large community so- multiple-item structural support indexes. cial network. Both conceptualizations of social These same studies provide little evidence of support are evidently correct in some respects, buffering interactions. Evidently, embedded- but each represents a different process through ness in a social network is beneficial to well- which social support may affect well-being. being but not necessarily helpful in the face of stress. As noted earlier, these results may be Evidence for Buffering and Main attributable either to a more general effect of Effect Models social networks on feelings of stability, pre- dictability, and self-worth or to the effect of Evidence for the buffering model. The extreme isolation for those with very few social studies reviewed in this article provide consis- connections. tent evidence for the buffering effects of social Hence, a central conclusion of this article support when certain conditions are present. is that social integration influences well-being First, the study must meet the minimal meth- in ways that do not necessarily involve im- odological and statistical criteria described proved means of coping with stressful events. earlier in this article. Second, the support in- There is little evidence for the view that strument must measure perceived availability imbeddedness in a social network is related of a support function or functions. Studies us- to well-being primarily because it defines a ing instruments measuring the structure of so- potential for coping action (e.g., Gore, cial networks, and those measuring the degree 1985; Wheaton, 1982). There are two sources to which support functions were provided in of support for our argument: (a) The correla- the past, have not found buffering effects. As tion between social integration and functional we discussed earlier, structural measures do support availability measures is low, and (b) not assess supportive functions that are re- although a single confidant is sufficient for sponsive to stressful events, and measures of stress buffering, a large range of social contacts support received in the past confound the is not. are availability of support and the recent need for A number of studies using complex func- that support. Third, the support functions as- tional indexes (Table 5) indicated main effects sessed must be ones that enhance broadly use- of support without buffering interactions. Un- ful coping abilities. This condition reflects the fortunately, these studies are characterized by global nature of the cumulative life stress in- a range of methodological and statistical prob- struments used in this literature. Studies using lems that severely reduce the probability of support instruments that tap the broadly useful detecting interactions. A general problem is esteem and informational support functions that the highly compound nature of many of have been consistently successful in showing these measures reduces the stress-support evidence of a buffering process. linkage that is theoretically necessary for The literature provides little direct evidence showing buffering effects. for our proposal that persons experiencing a Pure buffering or buffering plus main effect? specific stressor would be best protected by As noted earlier, the main effect and buffering supportive functions that provide stressor-spe- hypotheses are not mutually exclusive; buff- cific coping resources (S. Cohen & McKay, ering interactions could occur with support 1984). Evidence discriminating the stress- being associated with symptomatology even support matching hypothesis from one that within the low-stress group. In view of consis- suggests that esteem and/or informational tent evidence for a beneficial main effect of support alone are the sole sources of stress various social support measures in the predic- buffering is not provided by the existing lit- tion of symptomatology, one might expect erature. Instead, these hypotheses must be buffering effects and main effects to occur to- compared in studies assessing the buffering gether. The evidence, however, does not sup- adequacy of a range of support resources in port such an assertion. Those studies that (a) response to specific stressful events. obtained a significant buffering interaction and Evidence for the main effect model. There (b) provided enough data to estimate if there is consistent evidence for beneficial main ef- was an association between support and SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 349 symptomatology under low stress are virtually for buffering effects have primarily used mea- unanimous in showing a pure buffering effect sures that assess the perceived availability of (S. Cohen & Hoberman, 1983; Fleming et al., functional support. Why is perception impor- 1982;Frydman, 1981; Henderson, 1981;Hen- tant? If one assumes that the buffering qualities derson et al., 1980; Miller &Ingham, 1979; of social support are to some extent cognitively Paykel et al., 1980; Wilcox, 1981). This con- mediated (i.e., support operates by affecting sistency is even more impressive in that the one's interpretation of the stressor or knowl- definition of what constitutes the low-stress edge of coping resources), then a measure of group in these studies ranges from persons who perception of the availability of support would have not experienced any major event to those be a sensitive indicator of buffering effects (cf. with a stress score falling below the 50th per- S. Cohen & McKay, 1984). We expect that centile (i.e., a median split dividing the sample available functional support would be drawn into relatively low- and high-stress groups). The on when necessary in times of stress, but as consistent finding of a pure buffering effect yet there has been no direct examination of suggests that certain support resources act only this process (see Gottlieb, in press; Wilcox & in the presence of an elevated stress level; for Vernberg, 1985). Although it is likely that there example, having access to persons to talk to is correspondence between perceptions of about one's problem promotes well-being in support availability and the actual social the face of stress but not necessarily under transactions that occur in response to stress, nonstressful conditions. These data are con- the ability of measures of support perception sistent with the proposal that specific support to show buffering effects is theoretically inter- functions are responsive to stressful events, esting. whereas social network integration operates to One might postulate that perceived avail- maintain feelings of stability and well-being ability of support would work in the face of irrespective of stress level. We conclude that acute stressors, but not in the face of ongoing social integration and functional support rep- chronic strains. After all, in the case of chronic resent different processes through which social strains, one would need eventually to engage resources may influence well-being. in a support transaction that would be either successful or not. The data, however, clearly Other Issues Clarified in this Review indicate the buffering effectiveness of perceived support measures in studies with stress mea- Buffering effects do not result from an arli- sures that tap chronic strain (e.g., Fleming et factual confounding of stress and social support al., 1982; House & Wells, 1978; Kessler & Es- measurement. One of the goals of this article sex, 1982; Linn &McGranahan, 1980; Miller was to examine the possibility that buffering & Ingham, 1979; Pearlin et al., 1981; see Kes- effects are artifactual, deriving from a con- sler & McLeod, 1985, review). Only Anes- founding of social exit life events and social hensel and Stone (1982) failed to find buffering support measures (cf. Berkman, 1982; Gore, of chronic strain, and we have previously dis- 1981; Thoits, 1982b). Contrary to the predic- cussed the methodological problems with this tion generated by the artifact hypothesis, stud- study. Hence, the perception of support avail- ies in which there was no correlation between ability continues to operate in chronically the stress and support measures all show clear stressing conditions and/or provides a good evidence for the buffering model. Conversely, indirect measure of the effective support people studies in which there was obvious, explicit are actually receiving. confounding of stress and social exits (Anes- Quality of available support is important. hensel & Stone, 1982; Dean & Ensel, 1982; Although there is considerable evidence for the Warheit, 1979) fail to show buffering effects. effectiveness of support availability measures, Hence, although a social exit confound may Henderson and colleagues (Henderson, 1981; present a problem for interpretation of certain Henderson et al., 1980) reported that the per- individual studies, it does not provide a tenable ceived adequacy of available support is more explanation for the literature as a whole. important than availability per se. It is likely Perception of available functional support that there is no real discrepancy between Hen- operates as a buffer. Studies finding evidence derson's work and the remaining literature. 350 SHELDON COHEN AND THOMAS ASHBY WILLS Most support measures used in this literature problems, and people, whereas men derive assess perceived availability of adequate sup- satisfaction from companionship activities and port. The distinction made by Henderson instrumental task accomplishment (Caldwell comes to light only when responses regarding & Peplau, 1982; Wills, Weiss, & Patterson, less adequate available support are elicited (and 1974). Hence, the content of supportive inter- distinguished from adequate support) as in his actions may be different for men and women. interview technique. The appraisal model dis- Sex differences may also be an important cussed previously would also argue that only factor in the main effect relation between social support that is perceived as adequate would network integration and well-being. Although influence the appraisal process and as a con- none of the reviewed studies using complex- sequence operate as a buffer. structural measures examined sex differences, There are other scattered data on support there is evidence from other work (e.g., Berk- satisfaction as well, but they differ in terms of man & Syme, 1979; House et al., 1982) that exactly what is being evaluated. For example, women may profit less than men (or not at all) Husaini et al. (1982) found a buffering effect from social integration as measured in these for marital satisfaction. In other words, those studies. with a potentially supportive person who were The available evidence also shows clearly satisfied with the support they receive were that support needs vary by type of stressor. protected. Barrera (1981) failed to find a buff- Although the work reviewed in this article was ering effect with a scale assessing satisfaction generally based on measures of cumulative life with opportunities to receive support, and events, all studies that included more than one McFarlane et al. (19S3) failed to find an effect subpopulation (Frydman, 1981; Henderson et with a scale that appears to assess the past al., 1980) or more than one stressor (Kessler helpfulness of actual supporters. Clearly fur- & Essex, 1982; Linn & McGranahan, 1980) ther work in this area needs to distinguish be- reported stress buffering results that vary tween satisfaction with how much support one across subgroups and stressors. These results has, with the perceived quality of available imply that specific stress-support linkages may support (holding amount constant), and with be obscured by the use of global measures of the quality of past support. life stress. There are individual and group differences Finally, although several studies included in support needs. Several studies indicate that social class measures, the role of social class support functions that are effective buffers for in support effects remains unclear. Although women may not be effective for males and vice there is evidence that lower SES persons score versa. Both Husaini et al. and Henderson found lower on structural support measures (Bell et buffering effects of confidant support for al., 1982; Warheit, 1979), SES does not seem women but not for men, and Henderson found to be important in discriminating between buffering effects of his measures of "acquain- persons who are or are not affected by support tanceship, friendship, reassurance of worth, (Bell et al., 1982; Turner & Noh, 1983; War- and reliable alliance" (Henderson et al., 1981, heit, 1979). However, none of these studies p. 38) for men but not for women. These dif- used support measures that are clearly func- ferences may be attributable to differences in tional in nature; hence, the data on the role of the types of stressors experienced by men and SES in support effects is insufficient at this women (cf. Billings & Moos, 1981), but also time. may be attributable to sex differences in needs There is little evidence for a negative effect elicited by the same stressors or to sex differ- of social networks on symptomatology. Some ences in styles of either socializing or coping. commentators have suggested that social net- Unfortunately, there is little direct evidence works can be sources of stress and conflict and regarding the reason for these differences. may thus increase symptomatology as opposed Suggestive evidence is, however, provided by to decreasing it (e.g., Fiori, Becker, & Coppel, the literature on gender differences in the con- 1983; Hall & Wellman, 1985; Rook, 1984). tent of interpersonal interactions. These studies Although this may be true for individual cases, suggest that women derive satisfaction from it does not seem to be an important phenom- talking with intimate friends about feelings, enon in general. We have noted that compound SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 351 structural measures, indexing a wide range of vided by esteem and informational support social connections, consistently show a signif- enhance the buffering process. Future research, icant, beneficial main effect. Evidence for sin- testing the effectiveness of specific support gle-item structural measures is mixed; some resources in response to specific stressors, measures show positive effects and others are would help to clarify the operative mecha- unrelated to health outcomes. However, there nisms. For example, one could ask whether a is no evidence of a negative effect of a single- threat to self-esteem, like failure on an im- item structural measure. Only measures of portant exam, is best buffered by esteem sup- network conflict, a structural aspect of a social port or whether a loss of income (when there network that would not be considered sup- is no self-attribution of failure) is best buffered portive, have been found to have negative ef- by material aid. fects on symptomatology (Barrera, 1981; Future work needs to be based on clear Henderson, 1981). Functional measures are theoretical models of mediating processes in similarly consistent in indicating a beneficial support-well-being relationships. With regard effect of support (excepting past-use measures to stress buffering, we noted earlier that sup- which are confounded with stress). It may be port may reduce stress by altering appraisal of that there are some costs associated with re- stressors, by changing coping patterns, or by ceiving support in particular instances, espe- affecting self-perceptions. For example, sup- cially when it is asked for or when the receiver port may serve a buffering function through feels obligated to the giver as a result of the direct effects on self-esteem, enhancement of transaction (see, e.g., Greenberg & Westcott, self-efficacy (leading to increased persistence 1983; Rosen, 1983; Wills, 1983). It is also at coping efforts), or direct changes in problem- likely, however, that most effective support is solving behaviors. Currently, we know little given and taken in the context of daily social about which of these possibilities is most rel- intercourse without being asked for, and with- evant for buffering effects or about the relation out supporters feeling that they are giving between social support and other cognitive and something, or supportees feeling that they are behavioral coping measures (cf. Gore, 1985). receiving something (cf. Pearlin & Schooler, Similarly, embeddedness in a social network 1978). The availability of persons for this in- may enhance well-being by facilitating the de- formal give-and-take may be what is being velopment of feelings of predictability and sta- measured by the perceived-support scales in bility, by maintaining positive affective states, this literature. or by providing status support through social recognition of self-worth. The embeddedness Directions for Further Research question is especially interesting because the measures used in the existing literature have Where do we go from here? Below we list a not provided any information about the psy- number of unanswered questions whose res- chological mechanisms underlying the bene- olution we think would add significantly to our ficial effects of social integration. Future stud- understanding of the role of social support in ies could focus on how social integration is health and health-related behaviors. related to changes in social skills, comfort with How does social support work? It is clear or desire to use network resources, affective from the present review that embeddedness in states, and feelings of competence, self-esteem, a social network and social resources that are control, and predictability. responsive to stressful events have beneficial Another crucial theoretical issue is whether effects on well-being. The important questions buffering or main effects of support derive pri- now have to do with how these two types of marily from one (or a few) close relationships. social support act to prevent the development The possibility that one relationship is suffi- of symptomatology. In the case of the buffering cient is implied by the highly consistent find- model, the first issue is whether buffering de- ings of buffering effects with confidant mea- pends on a match between the needs elicited sures, which in many cases are based on single by particular stressful events and social re- relationships, and by the fact that measures sources perceived to be available. It is clear attempting to index the existence of close or that the broadly useful coping resources pro- friendly relationships typically show buffering 352 SHELDON COHEN AND THOMAS ASHBY WILLS effects. Future research could be designed to & Wellman, 1985). Apart from structural provide estimates of the variance in well-being considerations to some extent perceptions accounted for by each of the few most sup- must be based on actual social exchanges and portive relationships in a person's life. supportive transactions, either personally ex- Is social support related to serious health perienced or observed; but there is little direct outcomes? At present, there is a large gap in evidence of the types of exchanges that con- knowledge of the relationship between support tribute to support perceptions. A reasonable and serious physical health outcomes. The ev- direction for further research is to use knowl- idence for buffering and main effects of support edge from social exchange theory, coping the- is well established in the case of self-reported ory, and formulations of interpersonal rela- psychological distress, mixed in the case of self- tionships (Clark, 1983; Fisher, Nadler, & reported physical symptomatology, and vir- DePaulo, 1983; Gottlieb, in press; Wilcox & tually nonexistent for intermediate-level health Vernberg, 1985; Wills, 1983, 1985) to elucidate outcomes such as disease onset, objectively which aspects of the social environment are measured changes in physiological function- perceived as supportive, how and where sup- ing, or clinically diagnosed physical problems. portive transactions occur, and how support- Although there is clear evidence of a link be- seeking and support-receiving experiences are tween social support and mortality (Berkman involved in the general process of coping with & Syme, 1979; Blazer, 1982; House et al., stress. 1982), there is little understanding of the pro- What are the issues in developing social cesses or intermediate stages in this relation. support measures adequate to the tasks pre- From the limited data base, we are reluctant viously outlined? Answers to many of the to make sweeping generalizations about the conceptual questions previously raised depend relation between social support and physical on the development of support instruments illness. We think the evidence suggests that so- that provide precise measures of theoretically cial support does more than simply affect defined support functions (S. Cohen & Syme, symptom reporting, but more research is 1985a; House & Kahn, 1985). In the case of needed before we can understand how the the buffering model, these instruments should support processes discussed in this article are provide relatively independent measures of related to physical health. In theory, social functions that are responsive to needs elicited support could be related to physical health be- by stressful events, such as esteem support and cause it (a) affects exposure to the disease instrumental support (cf. S. Cohen et al., agent, (b) influences susceptibility versus re- 1985). Comparisons of support as perceived sistance to infection by the disease agent, (c) by different observers or measured by different affects self-care or medical help seeking once methods and comparison of these perceptions disease has been contracted, or (d) modifies with actual supportive behaviors (see measures the severity of disease or the disease course (cf. in Reis, Wheeler, Kernis, Spiegel, & Nezlek, Jemmott & Locke, 1984; Kasl, Evans, & Nei- 1985) could provide interesting new perspec- derman, 1979). Studies assessing the effects of tives on the conceptualization of social support support on several or all of these mechanisms assessment. Similarly, a combination of qual- would help clarify which aspects of health are itative and quantitative research on social net- being affected and how. works (cf. Ingersoll, 1982; Wilcox, 1980) may How are perceptions of support formed and provide a valuable perspective on the dynamics maintained? As discussed earlier, measures of how networks operate and respond to par- of perceived support availability are notably ticular life stressors. successful in showing evidence for buffering. It would also be valuable to have measure- But where do perceptions of adequate support ment procedures that allow distinctions be- originate? It is clear from the existing literature tween sources of support. At present there are that functional support availability is only only a handful of studies that compare the ef- modestly correlated with structural measures. fectiveness of work and family support (e.g., It is possible, however, that more sophisticated Holahan & Moos, 1981; House & Wells, 1978; measures of network structure will result in Kobasa & Puccetti, 1983; La Rocco et al., better prediction of perceived support (see Hall 1980). Research examining the relative effec- SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 353 tiveness of work- and family-based support need to change persons' perceptions of avail- would be particularly important for testing able support. Altering perceptions is a difficult specific predictions about buffering for differ- task even in a relatively well-controlled labo- ent types of stressors. ratory setting. Moreover, as noted, support fs social support the causal factor in the perceptions may be mediated by stable per- studies supporting main effect and buffering sonality traits. Despite the increased complex- models? As noted in this review, several lon- ity of conducting well-controlled research in gitudinal studies have shown either main ef- real-world settings (see Rook & Dooley, in fects of support, or buffering effects, using press), we feel that the yield from such research prospective analyses where symptomatology at would be highly valuable to the field and that Time 2 was predicted from stress and/or social extensive investment of time and effort in well- support at Time 1, with control for Time 1 designed longitudinal intervention work is symptomatology. In view of these findings, justified. several alternative causal interpretations of this literature are considered less likely. For ex- Conclusion ample, it is unlikely that these effects are at- tributable to high stress lowering support levels In conclusion, we note that studies com- or high symptomatology reducing support. paring the main effect and buffering models However, there is still the possibility that social have opened an important area of psycholog- support is a "proxy" for some causal vari- ical research. With the accumulated knowledge able(s) with which support is highly correlated. from a decade of work, there is no longer a Stable personality variables such as social need to ask which model is correct. Both competence are plausible candidates (cf. S. models contribute to understanding the rela- Cohen & Syme, 1985a; Heller, 1979; Heller & tion between social support and health. New Swindle, 1983; Henderson et al., 1981). That research in this area will have important im- is, it may be that socially competent people are plications for the understanding of stress and coping, the determinants of psychological ad- more capable of developing strong support networks and of staying healthy by effectively justment and physical health, and the social coping with stressful events or by performing structure of communities. Such knowledge will serve to strengthen the supportive aspects of effective health-enhancing behaviors. Hence, informal helping networks and may provide a effects that we have attributed to support may be partially or wholly attributable to person- basis for a new partnership between lay helping ality traits such as competence and sociability resources and professional helpers. This work, that are highly correlated with support. Studies we think, will contribute in many ways to the using longitudinal prospective designs that in- well-being of individuals, families, and the clude measures of variables such as IQ, social larger society. competence, sociability, extraversion, neurot- icism, and social class (and other demographic References variables) would be crucial in ruling out spe- Andrews, G., Tennant, C, Hewson, D. M., & Vaillant, cific rival explanations for social support ef- G. E. (1978). Life event stress, social support, coping fects. style, and risk of psychological impairment. Journal of Nervous and Mental Disease, 166, 307-316. Although the body of evidence on social Aneshensel, C. S., & Frerichs, R. R. (1982). Stress, support, support reviewed in this article is fairly com- and depression: A longitudinal causal model. Journal of pelling, it is nonexperimental; only a true ex- Community Psychology, 10, 363-376. perimental study can provide the kind of strong Aneshensel, C. S., & Stone, J. D. (1982). Stress and depres- evidence that will allow definitive causal in- sion: A test of the buffering model of social support. Archives of General Psychiatry, 39, 1392-1396. ferences. Intervention studies in which subjects Antonucci, T. C.. & Depner, C. E. (1982). Social support are randomly assigned to conditions and fol- and informal helping relationships. In T. A. Wills (Ed.), lowed over time would allow such inferences Basic processes in helping relationships (pp. 233-254). (cf. Gottlieb, 1985). It is easy to suggest inter- New York: Academic Press. Barrera, M., Jr. (1981). Social support in the adjustment vention work, but this work is demanding in of pregnant adolescents: Assessment issues. In B. H. practice. For example, in the case of buffering Gottlieb (Ed.), Social networks and social support (pp. stressful events, an effective intervention may 69-96). Beverly Hills, CA: Sage. 354 SHELDON COHEN AND THOMAS ASHBY WILLS Ban-era, M., Jr., & Ainlay, S. L. (1983). The structure of Cobb, S. (1976). Social support as a moderator of life stress. social support: A conceptual and empirical analysis. Psychomatic Medicine, 38, 300-314. Journal of Community Psychology, 11, 133-143. Cobb, S., & Kasl, S. V. (1977). Termination: The conse- Barrera, M., Jr., Sandier, I. N., & Ramsay, T. B. (1981). quences of job loss(DHEW Publication 77-224). Wash- Preliminary development of a scale of social support: ington, DC: U.S. Government Printing Office. Studies on college students. American Journal of Com- Cohen, P., Struening, E. L., Muhlin, G. L., Genevie, L. E., munity Psychology, 9, 435-447. Kaplan, S. R., & Peck, H. B. (1982). Community stress- Baum, A., Singer, J. E., & Baum. C. S. (1981). Stress and ors, mediating conditions, and well being in urban the environment. Journal of Social Issues, 37, 4-35. neighborhoods. Journal of Community Psychology, 10. Bell, R. A., LeRoy, J. B., & Stephenson, J. J. (1982). Eval- 377-391. uating the mediating effects of social supports upon life Cohen, S., & Hoberman, H. (1983). Positive events and events and depressive symptoms. Journal of Community social supports as buffers of life change stress. Journal Psychology, 10, 325-340. of Applied Social Psychology, IS, 99-125. Berkman, L. F. (1982). Social networks and analysis and Cohen, S., & McKay, G. (1984). Social support, stress and coronary heart disease. Advances in Cardiology, 29, 37- the buffering hypothesis: A theoretical analysis. In A. 49. Baum, J. E. Singer, & S. E. Taylor (Eds.), Handbook of Berkman, L. E, & Syme, S. L. (1979). Social networks, psychology and health (Vol. 4, pp. 253-267). Hillsdale, host resistance, and mortality: A nine-year follow-up NJ: Erlbaum. study of Alameda County residents. American Journal Cohen, S., Mermelstein, R., Kamarck, T., & Hoberman, of Epidemiology, 109, 186-204. H. (1985). Measuring the functional components of so- Billings, A. G., & Moos, R. H. (1981). The role of coping cial support. In I. G. Sarason & B. Sarason (Eds.), Social responses and social resources in attenuating the stress support: Theory, research and applications (pp. 73-94). of life events. Journal of Behavioral Medicine, 4, 139- The Hague, The Netherlands: Martinus Nijhoff. 157. Cohen, S., & Syme, S. L. (1985a). Issues in the study and Billings, A. G., & Moos, R. H. (1982). Stressful life events application of social support. In S. Cohen & S. L. Syme and symptoms: A longitudinal model. Health Psychology, (Eds.), Social support and health (pp. 3-22). New York: 7,99-117. Academic Press. Blazer, D. G. (1982). Social support and mortality in an Cohen, S., & Syme, S. L. (Eds.). (1985b). Social support elderly community population. American Journal of and health. New York: Academic Press. Epidemiology, 115, 684-694. Dawes, R. M. (1969). Interaction effects in the presence Broadhead, W. E., Kaplan, B. H., James, S. A., Wagner, of asymmetrical transfer. Psychological Bulletin, 71. 55- E. H., Schoenbach, V. J., Crimson, R., Heyden, S., Tib- 57. blin, G., & Gehlbach, S. H. (1983). The epidemiologic Dean, A., & Ensel, W. M. (1982). Modelling social support, evidence for a relationship between social support and life events, competence, and depression in the context health. American Journal of Epidemiology, 117, 521- of age and sex. Journal of Community Psychology, 10. 537. 392-408. Brown, G. W., Bhrolchain, M. N., & Harris, T. (1975). Dean, A., & Lin, N. (1977). The stress-buffering role of Social class and psychiatric disturbance among women social support. Journal of Nervous and Mental Disease, in an urban population. Sociology, 9, 225-254. ItiS, 403-417. Brown, G. W., & Harris, T. (1978). Social origins of deAraujo, G., van Arsdel, P. P., Holmes, T. H., & Dudley, depression: A study of psychiatric disorders in women. D. L. (1973). Life change, coping ability and chronic London: Tavistock. intrinsic asthma. Journal of Psychosomatic Research, Caldwell, M. A., & Peplau, L. A. (1982). Sex differences 17. 359-363. in same sex friendship. Sex Roles, 8, 721 -732. DePaulo, B.. Nadler, A., & Fisher, J. D. (1983). New di- Caplan, G. (1974). Support systems and community mental rections in helping: Vol. 2. Help-seeking. New \brk: Ac- health. New York: Behavioral Publications. ademic Press. Dohrenwend, B. P., Dohrenwend, B. S., Gould, M. S., Link, Caplan, R. D. (1979). Social support, person-environment B., Neugebauer, R., & Wunsch-Hitzig, R. (1980). Mental fit, and coping. In L. F. Furman & J. Gordis (Eds.), illness in the United States: Epidemiological estimates. Mental health and the economy (pp. 89-131). Kala- New York: Praeger. mazoo, MI: Upjohn Institute for Employment Research. Dohrenwend, B. P., Shrout, P. E., Ergi, G., & Mendelsohn, Cassel, J. C. (1976). The contribution of the social envi- F. S. (1980). Nonspecific psychological distress and other ronment to host resistance. American Journal of Epi- dimensions of psychopathology: Measures for use in the demiology, 104, 107-123. general population. Archives of General Psychology, 37. Clark, M. S. (1983). Some implications of close social 1229-1236. bonding for social seeking. In B. M. DePaulo, A. Nadler, Dooley, D. (1985). Causal inferences in the study of social & J. D, Fisher (Eds.), New directions in helping: Help- support. In S. Cohen & S. L. Syme (Eds.), Issues in the seeking (Vol. 2, pp. 205-229). New York: Academic study and application of social support (pp. 109-128). Press. New York: Academic Press. Cleary, P. D. & Kessler, R. C. (1982). The estimation and Eaton, W. W. (1978). Life events, social supports, and psy- interpretation of modifier effects. Journal of Health and chiatric symptoms: A reanalysis of the New Haven data. Social Behavior, 23, 159-168. Journal of Health and Social Behavior, 19, 230-234. Cleary, P. D., & Mechanic, D. (1983). Sex differences in Fiori, J., Becker, J, & Coppel, D. B. (1983). Social network psychological distress among married women. Journal interactions: A buffer or a stress? American Journal of of Health and Social Behavior, 24, 111-121. Community Psychology, 11, 423-439. SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 355 Fisher, J. D., Nadler, A., & DePaulo, B. M. (1983). New Henderson, S. (1981). Social relationships, adversity and directions in helping: Vol. 1. Recipient reactions to aid. neurosis: An analysis of prospective observations. British New York: Academic Press. Journal of Psychiatry, 138. 391-398. Fleming, R., Baum, A., Gisriel, M. M., & Gatchel, R. J. Henderson, S., Byrne, D. G., & Duncan-Jones, P. (1981). (1982). Mediating influences of social support on stress Neurosis and the social environment. New York: Aca- at three mile island. Journal of Human Stress, 8, 14- demic Press. 22. Henderson, S., Byrne, D. G., Duncan-Jones, P., Scott, R., Frydman, M. I. (1981). Social support, life events and psy- & Adcock, S. (1980). Social relationships, adversity and chiatric symptoms: A study of direct, conditional and neurosis: A study of associations in a general population interaction effects. Social Psychiatry, 16, 69-78. sample. British Journal of Psychiatry, 136, 354-583. Garber, J., & Seligman, M. E. P. (1980). Human helpless- Holahan, C. J., & Moos, R. H. (1981). Social support and ness: Theory and applications. New York: Academic psychological distress: A longitudinal analysis. Journal Press. of Abnormal Psychology, 90, 365-370. Garrity, T. F., Somes, G. W., & Marx, M. B. (1978). Factors Holmes, T, & Rahe, R. (1967). The social readjustment influencing self-assessment of health. Social Science and rating scale. Journal of Psychosomatic Research, 11, Medicine. 12. 77-81. 213-218. Gore, S. (1978). The effect of social support in moderating House, J. S. (1981). Work stress and social support. Read- the health consequences of unemployment. Journal of ing, MA: Addison-Wesley. Health and Social Behavior, 19, 157-165. House, J. S., & Kahn, R. L. (1985). Measures and concepts Gore, S. (1981). Stress-buffering functions of social sup- of social support. In S. Cohen & S. L. Syme (Eds.), Social ports: An appraisal and clarification of research models. support and health (pp. 83-108). New York: Academic In B. S. Dohrenwend & B. P. Dohrenwend (Eds.), Press. Stressful life events and their contexts. New York: Prodist. House, J. S., Robbins, C., & Metzner, H. L. (1982). The Gore, S. (1985). Social support and styles of coping with association of social relationships and activities with stress. In S. Cohen & S. L. Syme (Eds.), Social support mortality: Prospective evidence from the Tecumseh and health (pp. 263-280). New York: Academic Press. Community Health Study. American Journal of Epi- Gottlieb, B. H. (1978). The development and application demiology, 116, 123-140. of a classification scheme of informal helping behaviors. House, J. S., & Wells, J. A. (1978). Occupational stress, Canadian Journal of Science, 10(2), 105-115. social support and health. In A. McLean, G. Black, & Gottlieb, B. H. (Ed.). (1981). Social networks and social M. Colligan (Eds.), Reducing occupational stress: Pro- support. Beverly Hills, CA: Sage. ceedings of a conference (HEW Publication No. 78-140, Gottlieb, B. H. (1983). Social support strategies. Beverly pp. 8-29). Washington, DC: U.S. Government Printing Hills, CA: Sage. Office. Gottlieb, B. H. (1985). Social support and community Husaini, B. A., Neff, J. A., Newbrough, J. R., & Moore, mental health. In S. Cohen & S. L. Syme (Eds.), Social M. C. (1982). The stress-buffering role of social support support and health (pp. 303-326). New York: Academic and personal confidence among the rural married. Jour- Press. nal of Community Psychology, 10, 409-426. Gottlieb, B. H. (in press). Social support and the study of Ingersoll, B. (1982, August). Combining methodological personal relations. Journal of Social and Personal Re- approaches in social support. Research paper presented lations. at the annual meeting of the American Psychological Greenberg, M. S., & Westcott, D. R. (1983). Indebtedness Association, Washington, DC. as a mediator of reactions to aid. In J. D. Fisher, A. Jemmott, J. B., Ill, & Locke, S. E. (1984). Psychosocial Nadler, & B. M. DePaulo (Eds.), New directions in help- factors, immunologic mediation, and human suscepti- ing: Vol. 1. Reactions to aid (pp. 85-112). New York: bility to infectious diseases: How much do we know? Academic Press. Psychological Bulletin, 95, 78-108. Habif, V. L., & Lahey, B. B. (1980). Assessment of the life Kaplan, B. H., Cassel, J. C., & Gore, S. (1977). Social stress-depression relationship: The use of social support support and health. Medical Care, 15, 47-58. as a moderator variable. Journal of Behavioral Assess- Kasl, S. V., & Cobb, S. (1982). Variability of stress effects ment. 2, 167-173. among men experiencing job loss. In L. Goldberger & Hall, A., & Wellman, B. (1985). Social networks and social S. Breznitz (Eds.), Handbook of stress (pp. 445-465). support. In S. Cohen & S. L. Syme (Eds.), Social support New York: Free Press. and health (pp. 23-42). New York: Academic Press. Hammer, M. (1981). "Core" and "extended" social net- Kasl, S. V., Evans, A. S., & Neiderman, J. C. (1979). Psy- works in relation to health and illness. Social Science chosocial risk factors in the development of infectious and Medicine, 17, 405-411. mononucleosis. Psychosomatic Medicine, 41, 441-466. Heller, K. (1979). The effects of social support: Prevention Kessler, R. C., & Essex, M. (1982). Marital status and and treatment implications. In A. P. Goldstein & F. H. depression: The role of coping resources. Social Forces, Kanfer (Eds.), Maximizing treatment gains: Transfer 61, 484-507. enhancement in psychotherapy (pp. 353-382). New York: Kessler, R. C., & McLeod, J. D. (1985). Social support Academic Press. and mental health in community samples. In S. Cohen Heller, K.., & Swindle, R. W. (1983). Social networks, per- & S. L. Syme (Eds.), Social support and health (pp. 219- ceived social support and coping with stress. In R. D. 240). New York: Academic Press. Felner, L. A. Jason, J. Moritsugu, & S. S. Farber (Eds.), Kobasa, S. C., & Puccetti, M. C. (1983). Personality and Preventive psychology, research and practice in com- social resources in stress resistance. Journal of Person- munity intervention (pp. 87-103). New \brk: Pergamon. ality and Social Psychology, 45, 839-850. 356 SHELDON COHEN AND THOMAS ASHBY WILLS Krantz, D. S., Grunberg, N. E., & Baum, A. (1985). Health Nadler, A., Fisher, J. D., & DePaulo, B. M. (1984). New psychology. Annual Review of Psvchology, 36, 349-383. directions in helping: Vol. 3. Applied perspective in help- La Rocco, J. M., House, J. S., & French, J. R. P., Jr. (1980). seeking and receiving. New York: Academic Press. Social support, occupational stress, and health. Journal Norbeck, J. S., & Tilden, V. P. (1983). Life stress, social of Health and Social Behavior, 21, 202-218. support and emotional disequilibrium in complications La Rocco, J. M., & Jones, A. P. (1978). Co-worker and of pregnancy: A prospective, multi-variant study. Journal leader support as moderators of stress-strain relation- of Health and Social Behavior, 24(1), 30-45. ships in work situations. Journal of Applied Psychology, Nuckolls, K. B., Cassel, J., & Kaplan, B. H. (1972). Psy- 63, 629-634. chosocial assets, life crisis and the prognosis of preg- Lazarus, R. S. (1966). Psychological stress and the coping nancy. American Journal of Epidemiology, 95, 431 -441. process. New York: McGraw-Hill. Paykel, E. S., Emms, E. M., Fletcher, J., & Rassaby, E. S. Lazarus, R. S., & Launier, R. (1978). Stress-related trans- (1980). Life events and social support in puerperal actions between persons and environment. In L. A. Per- depression. British Journal of Psychology, 136, 339-346. vin & M. Lewis (Eds.), Perspectives in interactional psy- Pearlin, L. I., & Lieberman, M. (1979). Social sources of chology (pp. 287-327). New York: Plenum. emotional distress. In R. Simmons (Ed.), Research in Leavy, R. L. (1983). Social support and psychological dis- community and mental health: Vol. 1 (pp. 217-248). order: A review. Journal of Community Psychology, 11, Greenwich, CT: JAI. 3-21. Pearlin, L. L, Menaghan, E. G., Lieberman, M. A., & Mul- Lei, H., & Skinner, H. A. (1980). A psychometric study lan, J. T. (1981). The stress process. Journal of Health of life events and social readjustment. Journal of Psy- and Social Behavior, 22, 337-356. chosomatic Research, 24, 57-65. Pearlin, L. I., & Schooler, C. (1978). The structure of con- Levinger, G., & Huesmann, L. R. (1980). An "incremental ing. Journal of Health and Social Behavior, 19, 2-21. exchange" perspective on the pair relationship. In K. J. Pinneau, S. R., Jr. (1975). Effects oj social support onpsy- Jurgen, M. S. Greenberg, & R. H. Willis (Eds.), Social chological and physiological strains. Unpublished doc- exchange: Advances in theory and research (pp. 165- toral dissertation. University of Michigan. 188). New York: Plenum. Reis, H. T. (1984). Social interaction and well-being. In S. Lin, N., Simeone, R. S., Ensel, W. M., & Kuo, W. (1979). Duck (Ed.), Personal relationships: Vol. 5. Repairing Social support, stressful life events, and illness: A model personal relationships (pp. 21-45). London: Academic and an empirical test. Journal of Health and Social Be- Press. havior, 20, 108-119. Reis, H. T., Wheeler, L., Kernis, M. H., Spiegel, N., & Linn, J. G., & McGranahan, D. A. (1980). Personal dis- Nezlek, J. (1985). On specificity in the impact of social ruptions, social integration, subjective well-being, and participation in physical and psychological health. Jour- predisposition toward the use of counseling services. nal of Personality and Social Psychology, 48, 456-471. American Journal of Community Psychology, S, 87-100. Rook, K. S. (1984). The negative side of social interaction: McFarlane, A. H., Norman, G. R., Streiner, D. L., & Roy, Impact on psychological well-being. Journal of Person- R. G. (1983). The process of social stress: Stable, recip- ality and Social Psychology, 46, 1097-1108. rocal and mediating relationships. Journal of Health and Rook, K. S., & Dooley, D. (in press). Applying social sup- Social Behavior, 24, 160-173. port research: Theoretical problems and future direc- Miller, P. M., & Ingham, J. G. (1979). Reflections on the tions. Journal of Social Issues. life events to illness link with some preliminary findings. Rosen, S. (1983). Perceived inadequacy and help-seeking. In I. G. Sarason & C. D. Spielberger (Eds.), Stress and In B. M. DePaulo, A. Nadler, & J. D. Fisher (Eds.), New anxiety (ftp. 313-336). New York: Hemisphere. directions in helping: Vol. 2. Help-seeking (pp. 73-107). Mitchell, R. E., Billings, A. G., & Moos, R. H. (1982). New York: Academic Press. Social support and well-being: Implications for preven- Ross, C. E., & Mirowski, J. (1979). A comparison of life tion programs. Journal of Primary Prevention, 3, 77- event weighting schemes: Change, undesirability, and ef- 98. fect proportional indices. Journal of Health and Social Monroe, S. M. (1982). Life events and disorder: Event- Behavior, 20, 166-177. symptom associations and the course of disorder. Journal Sandier, I. N. (1980). Social support resources, stress and of Abnormal Psychology, 91, 14-24. maladjustment of poor children. American Journal of Monroe, S. M. (1983). Social support and disorder: Toward Community Psychology, 8, 41-52. an untangling of cause and effect. American Journal of Sandier, I. N.', & Lakey, B. (1982). Locus of control as a Community Psychology, 11, 81-96. stress moderator: The role of control perceptions and Monroe, S. M., Imhoff, D. E, Wise, B. D., & Harris, J. E. social support. American Journal of Community Psy- (1983). Prediction of psychological symptoms under chology, 10, 65-80. high-risk psychosocial circumstances: Life events, social Sarason, I. G., Johnson, J. H., & Siegel, J. M. (1978). As- support, and symptom specificity. Journal of Abnormal sessing the impact of life changes: Development of the Psychology, 92, 338-350. Life Experience Survey. Journal of Consulting and Clin- Moos, R. H., & Mitchell, R. E. (1982). Social network ical Psychology. 46. 932-946. resources and adaptation: A conceptual framework. In Sarason, I. G.. Levine, H. M., Basham, R. B., & Sarason, T. A. Wills (Ed.), Basic processes in helping relationships B. R. (1983). Assessing social support: The social support (pp. 213-232). New York: Academic Press. questionnaire. Journal of Personality and Social Psy- Myers, J. K., Lindenthal, J. J., & Pepper, M. P. (1975). chology, 44, 127-139. Life events, social integration and psychiatric symp- Sarason, I. G., & Sarason, B. R. (Eds.). (1985). Social sup- tomatology. Journal of Health and Social Behavior, 16, port: Theory, research and applications. The Hague, The 421-427. Netherlands: Martinus Nijhof. SOCIAL SUPPORT AND THE BUFFERING HYPOTHESIS 357 Schaefer, C, Coyne, J. C, & Lazarus, R. S. (1981). The pressive symptomatology. American Journal of Psychia- health-related functions of social support. Journal of try, ]}(,, 502-507. Behavioral Medicine, 4. 381-406. Wetnington, E. (1982, August). Can social support functions Sells, S. B. (1970). On the nature of stress. In J. E. McGrath be differentiated: A multivariate model. Paper presented (Ed.), Social and psychological factors in stress (pp. 134- at the annual meeting of the American Psychological 139). New York: Holt, Rinehart &. Winston. Association, Washington, DC. Silver, R. L., & Wortman, C. B, (1980). Coping with un- Wheaton, B. (1982). A comparison of the moderating ef- desirable life events. In J. Garber & M. E. P. Seligman fects of personal coping resources in the impact of ex- (Eds.), Human helplessness (pp. 279-340). New York: posure to stress in two groups. Journal of Community Academic Press. Psychology, №.293-311. Surtees, P. G. (1980). Social support, residual adversity Wilcox, B. L. (1980). Social support in adjusting to marital and depressive outcome. Social Psychiatry, 15, 71-80. disruption: A network analysis. In B. H. Gottlieb (Ed.), Tessler, R., & Mechanic, D. (1978). Psychological distress Social networks and social support (pp. 97-116). Beverly and perceived health status. Journal of Health and Social Hills, CA: Sage. Behavior, 19, 254-262. Wilcox, B. L. (1981). Social support, life stress, and psy- Thoits, P. A. (1982a). Conceptual, methodological, and chological adjustment: A test of the buffering hypothesis. theoretical problems in studying social support as a American Journal of Community Psychology, 9, 371- buffer against life stress. Journal of Health and Social 386. Behavior, 23, 145-159. Wilcox, B. L., & Vernberg, E. (1985). Conceptual and Thoits, P. A. (1982b). Life stress, social support, and psy- theoretical dilemmas facing social support research. In chological vulnerability: Epidemiological considerations. I. G. Sarason & B. R. Sarason (Eds.), Social support: Journal of Community Psychology, 10, 341-362. Theory, research and application (pp. 3-20). The Hague, Thoits, P, A. (1983). Multiple identities and psychological The Netherlands: Martinus Nijhof. well-being: A reformulation and test of the social iso- Williams, A. W., Ware, J. E., Jr., & Donald, C. A. (1981). lation hypothesis. American Sociological Review, 48, A model of mental health, life events, and social supports 174-187. applicable to general populations. Journal of Health and Thoits, P. A. (1985). Social support processes and psycho- Social Behavior, 22, 324-336. logical well-being: Theoretical possibilities. In I. G. Sar- Wills, T. A. (1983). Social comparison in coping and help- ason & B. Sarason (Eds.), Social support: Theory, re- seeking. In B. M. DePaulo, A. Nadler, & J. D. Fisher search and applications (pp. 51-72). The Hague, The (Eds.), New directions in helping: Vol. 2. Helpseeking Netherlands: Martinus Nijhof. (pp. 109-141). New York: Academic Press. Turner, R. J. (1981). Social support as a contingency in Wills, T. A. (1985). Supportive functions of interpersonal psychological well-being. Journal of Health and Social relationships. In S. Cohen & S. L. Syme (Eds.), Social Behavior, 22, 357-367. support and health (pp. 61-82). New York: Academic Turner, R.). (1983). Direct, indirect and moderating effects Press. of social support upon psychological distress and asso- Wills, T. A., & Langner, T. S. (1980). Socioeconomic status ciated conditions. In H. B. Kaplan (Ed.), Psychosocial and stress. In I. L. Kutash & L. B. Schlesinger (Eds.), stress: Trends in theory and research (pp. 105-155). New Handbook on stress and anxiety (pp. 159-173). San York: Academic Press. Francisco: Jossey-Bass. Turner, R. J., & Noh, S. (1983). Class and psychological Wills, T. A., Weiss. R. L., & Patterson, G. R. (1974). Be- vulnerability among women: The significance of social havioral analysis of the determinants of marital satis- support and personal control. Journal of Health and So- faction. Journal of Consulting and Clinical Psychology, cial Behavior, 24, 2-15. 42.802-811. Wallston, B. S., Alagna, S. W., DeVellis, B. M., & DeVellis, R. F. (1983). Social support and physical health. Health Psychology. 4, 367-391. Received July 18, 1984 Warheit, G. J. (1979). Life events, coping, stress, and de- Revision received February 19, 1985