Stress, Social Status, and Psychological Distress · Stress, Social Status, and Psychological Distress RONALD C. KESSLER University of Wisconsin—Madison Journal of Health and Social

Stress, Social Status, and Psychological Distress

RONALD C. KESSLERUniversity of Wisconsin—Madison

Journal of Health and Social Behavior 1979, Vol. 20 (September):259-272

II has long been known that persons in socially disadvantaged statuses are particularly likely toexhibit symptoms of psychological distress. Why this is so remains the subject of controversy.This paper sheds some light on the controversy by analyzing survey data on stress, socialstatus, and psychological distress. Two components of the status-distress relationship areisolated empirically: (I) a differential exposure to stress component, interpretable as one sort ofsocial causation influence; and (2) a component due to the differential impact of comparablestresses on people in different categories of various status dimensions. The analysis shows thatdifferential impact is the more important determinant in relationships between social class, sex,and marital status and self-reported distress. Only in the comparison of whites to nonwhites isdifferential exposure the key determinant. The implications of these findings for our under-standing of the social status-psychological distress relationship are discussed.

A central problem to emerge from the studyof sociodemographic correlates of mental dis-order has been to explain why persons in cer-tain disadvantaged social statuses have thehigh rates of emotional distress they do. Theearliest work on this problem concemed itselfwith social class (Faris and Dunham, 1939;Hollingshead and Redlich, 1958; Langner andMichael, 1963). But over the last two decades agreat deal of research has also been directedtoward the overrepresentation of certain emo-tional problems among the unmarried (Bach-rach, 1975), women (Weissman and Klerman,1977), and nonwhites (Dohrenwend andDohrenwend, 1970).'

Two general sorts of explanations have beenoffered for the relationship between disadvan-taged social status and emotional distress. (1)The first says that persons who either havechronic emotional problems, or are prone todevelop such problems, are channeled into theranks of the unmarried and into the lowerclass, the latter channeling possibly taking

Work on this paper was partially supported byNIMH Grants No. 5 T32 MH14641-02 and 1 R03MH32490-01. I wish to thank Jerome K. Myers andhis associates for generously allowing me to use theirdata in this paper. Their research was supported byPHS Contract No. 43-67-743 and Research GrantNo. MHI5522 from the National Institute of MentalHealth, Department of Health, Education and Wel-fare. I wish to thank Cris Laskowski for providingcomputer assistance and Ron Diamond for dis-cussions. The comments of the anonymous review-ers were extremely helpful.

place over a succession of generations andbeing characterized by an inability to move up,as well as by a tendency to drift down, the classhierarchy. This explanation has been associ-ated with the "drift" hypothesis in the litera-ture on social class and mental illness, and withthe "social selection" hypothesis in the litera-ture on marital status and mental illness. How-ever, it is better described generically as a"constitutional frailty" explanation,^ becauseit emphasizes intrapsychic propensities to dis-tress rather than environmental influences. (2)A second set of explanations, on the otherhand, has emphasized the importance of lifeconditions as determinants of emotional disor-der. Although not limited to one hypothesis,these explanations have been dominated by theview that exposure to stressful experiencesleads to emotional distress and that the distri-bution of stressful experiences in society ac-counts for differential rates of emotional prob-lems. Together these two explanations suggestthat socially disadvantaged persons will beboth more highly exposed to stressful experi-ences and also more highly influenced bystressful experiences than socially advantagedpersons.

Documentation of the influence of differen-tial exposure, if it could be done, would dem-onstrate that the greater psychological distressof disadvantaged persons is due, at least inpart, to greater exposure to distress-provokingenvironmental experiences. In contrast, thereare a number of ways to interpret a differential

259

260 JOURNAL OF HEALTH AND SOCIAL BEHAVIOR

impact across two sets of individuals. One is interms of psychic frailty. That is, because oftheir relative inability to manage stressfulsituations—their lack of social competence—some people might exhibit a great deal moreemotional distress than others who are morerobust. The sociocultural environment can bethe direct source of differential impact in anumber of ways. One of these is by determin-ing the subjective meanings of objectiveevents. Life situations are not inherentlystressful, of course. They must be interpreted.These interpretations, in turn, are partly de-pendent on the context of one's life, the rel-evancies and circumstances to which oneselectively attends. To the extent that socialstatus shapes interpretations of life experi-ences, the subjective stressfulness, and thusthe impact, of these experiences will vary withstatus.

There are also a number of ways in which thesocial environment can modify the impacts ofstress events as they are subjectively experi-enced. One of these is by influencing the cop-ing strategies used by persons subjected tostress. Just as objective occurrences are inter-preted through subjective filters, so one's op-tions are detennined by subjective appraisalsof availability and efficacy. One's position inthe social structure also defines one's access tovarious resources that can be useful in dealingwith stress. The stress-buffering effects of so-cial support systems have been widely dis-cussed (Cobb, 1976; Dean and Lin, 1977;Kaplan and Cassel, 1977; Mechanic, 1974), ashave the more obvious advantages of politicaland financial power.

Empirically it is very difficult to disentanglethese various modifying influences. However,a researcher can estimate the relative impor-tance of differential impact in explaining therelative functioning of two or more sets of indi-viduals without teasing out all the various in-fluences that go to make up this differentialimpact. In the analysis reported below, this isexactly what is done. A set of stress measuresare used to predict self-reported emotionaldistress among a sample of persons inter-viewed in a major epidemiological survey. Iexamine the relative importance of differentialexposure to these stresses and of differentialimpact of comparable stresses in accountingfor observed relationships between various

disadvantaged social statuses and an indicatorof emotional functioning. As shall be seen, thisanalysis sheds light on why persons in lessadvantaged social statuses are in relativelypoor emotional health.

METHOD

Data

The data used here come from the Myers etal. (1972, 1974, 1975) New Haven survey. Thesample consists of 720 persons interviewedboth in 1967 and 1969 from a larger randomsample of adults in the New Haven mentalhealth catchment area.

The interviews included a series of questionsabout exposure to potentially stressful life ex-periences that might lead to psychological dis-tress. I have focused on five summary mea-sures for purposes of this analysis. Three ofthese are life-event scales, created from a listof 56 events in the interview. All of these scalespertain to life events occurring during the twoyears between the two data points of the panel.The other two stress measures pertain to on-going life situations that can be stressful.' Thefirst is a measure of financial status, defined asa ratio of income to number of family memberssupported by this income. The second is ameasure of physical illness, defined as aweighted sum of self-reported symptoms."

There has been considerable debate in thelife-events literature about the best way toconceptualize global event indexes (B. P.Dohrenwend, 1974; Dohrenwend et al., 1978;Vinokur and Selzer, 1975). To provide as broadan analysis as possible, results are shown herefor each of two separate event conceptualiza-tions. The first uses raw counts of undesirable,desirable, and total life events as the three in-dexes of exposure.' This unweighted codingscheme is one that has been used in much ofthe work reported in the life-events literature.The second employs weighted counts of desir-able and undesirable events and a dummy vari-able to describe those who have experiencedno events at all over the course of the panel.The weights used here are those presented inHolmes and Masuda (1974) for the life-changeunits (LCU) scale. The weighted codingscheme is a modification of the Myers et al.(t974) Desirability-Change Life EventsIndex.*'

STRESS, SOCIAL STATUS, AND PSYCHOLOGICAL DISTRESS 261

The measure of emotional distress used hereis Macmillan's (1957) symptom checklist scaleas modified by Gurin et al. (1960). The scale ismade up of 20 statements about bodily feelingsthat indicate moods of depression and anxiety.Over the course of the two years, the scale hasa stability coefficient of from .50 (females) to.56 (married) in the subsamples examined here,which can be taken as a lower bound on thereliability of the scale.^

Although the symptoms included in thisscale are taken from those presented bypatients in treatment, it should be recognizedthat they do not tap a person's need for treat-ment. Nor do they allow one, in a communitysample, to define psychiatric "cases." How-ever, the scale does permit respondents to beordered on a continuum of reported distress.Scores on the scale range from 20 (all symp-toms are reported as occurring "often") to 80(all symptoms are reported as occurring"never"). The sample-wide mean was 72.8.Only 16% of the sample reported extreme dis-tress, defined, following Myers et al. (1974), asa score of 66 or lower.

Analysis Strategy

I have recently presented an analysis strat-egy for studying the relative importance ofdifferential exposure to stress and differentialimpact of stress (Kessler, 1979). This strategyis used below to compare men with women,whites with nonwhites, the married with theunmarried, and persons in the lower class withpersons in the middle and upper classes.

The strategy is based on a model of thecausal nexus between stress and emotionalfunctioning that assumes emotional distress (P)is influenced by exposure to stressors (S) act-ing on persons who vary in the extent to whichstress influences their emotional functioning(V). I assume that this impact varies dependingon the intrapsychic defenses and social re-sources used by the individual to interpret andmanage the stresses to which he or she is ex-posed. This model can be expressed as a pre-diction equation for a particular set of stressormeasures, where V is expressed as a metricregression coefficient. The comparison strat-egy consists of, first, estimating models of thissort in subsamples and, next, systematicallycomparing observed differences in emotional

distress across these subsamples in terms ofthe estimation equations obtained in the sub-samples.

If we take as our model

P = 2V(S) + a , (1)

then we can systematically compare averagesin the distress score between two subsamples(g,h) in terms of the following decomposition(Winsborough and Dickinson, 1971).

(2)

(3)

If we take as our model

In [p/(l-p)] = V(S) + a ,

then we can systematically compare odds ofextreme distress (treated as a dummy depen-dent variable, where Pi is the predicted prob-ability that person i is extremely distressed andp = 2 Pi/n) in terms of the following decom-position (Kessler, 1979):

(4)

In[p/(l-p)]h x eX eX e

These decompositions perfectly divide aver-age differences in distress into components dueuniquely to each of four conceptually distinctinfluences.* In both equations (2) and (4), thefirst component is a differential exposure ef-fect; the expected result of giving members ofstatus category g the exposures found to char-acterize members of status category h. Thesecond component in each of these equations isa differential impact effect; the expected resultof giving members of status category g the im-pact coefficients of persons in status categoryh. The third component is an empirical esti-mate of the interaction between the impact andexposure components. The last component is aresidual, representing the difference in distressbetween persons in the two contrast categoriesstandardized on the infiuences modeled in theprediction equations.

The exposure and impact components are ofthe most substantive interest, for they tell us


the relative importance of differential stressesand differential influence of comparable stress-es on symptom differences across the twostatus subsamples. The interaction term repre-sents the extent to which the symptom dif-ferences explained by the prediction equationscannot be attributed uniquely either to sub-sample differences in exposure or to dif-ferences in impact. When the interaction com-ponent is large relative to the exposure andimpact components, then it becomes imprac-tical to place a great deal of emphasis on therelative magnitudes of the latter two compo-nents.

In the analyses below, equations of the form(1), (3) are estimated separately for a series ofstatus category subsamples. These equationsare then decomposed by means of (2), (4) toevaluate the relative importance of differentialexposure and differential impact on the ob-served variations in self-reported distress be-tween the contrasted subsamples.

RESULTS

A number of critics of the life-events ap-proach to studying psychological distress havenoted that distressed persons frequently ex-pose themselves to stress (Brown, 1974;Dohrenwend et al., 1978; Rabkin and Struen-ing, 1976). Therefore it cannot be automaticallyassumed that relationships observed betweenexposure and emotional disorder are due solelyto the causal impact of stress on distress. Thecausal influence might work the other wayaround. One way to guard against this sort ofconfounding is to include in one's event in-dexes only those events that are relatively un-affected by the prior emotional functioning ofthe focal respondents (Brown, 1974). How-ever, this approach forces the analyst to dis-card the life stresses that discriminate mostclearly between persons in socially advantagedand disadvantaged status categories—suchevents as job layoff, arrest, eviction from one'shome, and marital dissolution. Fortunately, thefact that this analysis is concerned with paneldata makes it possible to adjust for the con-founding of causal direction in another way, bycontrolling Time 1 distress in the decomposi-tions. In this way I can include events thatmight be partially confounded by selection ef-fects in the indexes and partial out these effects

by means of statistical controls. Therefore, ineach of the decompositions reported below Isequentially include and delete the initial dis-tress scale as a way of assessing the influenceof selection on my results.

Decomposition of mean differences in thedistress scores of two contrast subsamples letsus glimpse the forces that make "typical" per-sons in the subsamples differ from each otherin emotional functioning. Decompositions ofdifferences in the likelihood of extreme distressgive us a somewhat different view, one of"extremes" rather than typical persons. It isnot always the case that stress influences typi-cal and extreme pattems in the same way. Inseveral cases below we find that the typicalperson in subsample A is less highly influencedby a stress than the typical person in subsam-ple B, but that this stress leads to a higherproportion of extreme distress in subsample Athan in B. The two patterns together describeone subsample (A) in which most people areunaffected by stress but in which those who areinfluenced respond in an extreme fashion, andanother subsample (B) in which most peopleare affected by stress but in which few of thesepeople respond in an extreme fashion.

Tables 1 and 2 present R̂ coefficients for the36 OLS regression and 36 individual-level logitequations used to estimate (1), (3) respectively,for the decompositions to be discussed below.Comparison of the coefficients in Tables 1 and2 shows that the estimation equations morecompletely predict the distress scale than the"extreme" distress dummy variable. This re-sult is to be expected, since the sizable skew inthe extreme distress dichotomy has the effectof attenuating correlations between predictedand observed scores.

The first two coefficient columns in eachtable are taken from equations estimated withthe unweighted predictors. The third andfourth coefficient columns are taken fromweighted-predictor equations. Comparison ofthe coefficients in these columns shows thatthe unweighted predictor set has a higher re-lationship to the distress scale than theweighted set in 35 of the 36 pairs of coeffi-cients. I do not want to enter into the con-troversy about the "best" way to concep-tualize life-event scales here. However, inexamining the results of the decompositionsreported below, one should bear in mind that


TABLE 1. Coefficients of Multiple Determination in the Subsample OLS Regression

MenWomen

Classes I-IIIClasses IV-VClass V

WhiteNot white

MarriedSep/div

(N)*

(318)(402)

(173)(547)(223)

(627)(90)

(541)(42;

Unweighted

.214

.222

.101

.238

.334

.185

.387

.149

.474

R

Unweightedwith G,

.417

.379

.368

.400

.454

.386

.473

.371

.579

2

Weighted

.196

.127

.069

.168

.192

.136

.238

.097

.332

Weightedwith G,

.402

.327

.331

.361

.378

.355

.349

.340

.440

Note: All coefficients are significant at p < .05. Both unweighted and weighted predictor sets includeincome per family member and the scale of physical health complaints as well as the three life-events indexes.G, refers to the Time 1 Gurin symptom scale.

*Missing values on the predictors were assigned mean scores. Complete data are available for thecriterion.

the distress scores tneasured iti this particular beeti unable to move beyond the position ofsample are better predicted by the unweighted their birth, are socially incompetent and highlypredictor set.' influenced by stress. Lower class people ap-

pear to be at a disadvantage with respect toSocial Class learned coping responses. In their Chicago

study, Pearlin and Schooler (1978) found thatThe literature on social class and mental ill- income and education were both positively

ness has emphasized the importance of the correlated to the use of coping strategies and todistribution of stress across classes as the key the possession of intrapsychic coping re-determinant of class differences in emotional sources found to be effective in managingdisorder (Markush and Favero, 1974; Myers et stress. There is also considerable indirect evi-al., 1974). However, there is good theoretical dence (Dohrenwend and Dohrenwend, 1970)reason to believe that there should also be class that lower class people have low access to so-differences in the impacts of stress. The social cial supports. These, in turn, are now believedselection hypothesis, for one, argues that lower to be extremely important resources in helpingclass persons who have drifted down from a person cope with life problems (Liem andhigher status families of origin, or who have Liem, 1978:148-51).

TABLE 2. CoefTidents of Multiple Determination in the Subsample Logits

MenWomen

Classes I-IIIClasses IV-VClass V

WhiteNot white

MarriedSep/div

(N)*

(280)(344)

(148)(476)(189)

(541)(81)

(478)(39)

Unweighted

.195

.146

.082

.176

.240

.120

.298

.088

.654

R

Unweightedwith G,

.272

.297

.292

.299

.325

.241

.450

.269

.725

2

Weighted

.175

.099

.084

.133

.184

.083

.329

.055

.379

Weightedwith G,

.257

.265

.280

.265

.285

.214

.435

.245

.284

Note: All coefficients are significant at p < .05.*Listwise deletion of missing values was used.


Table 3 presents the results of 16 decom-positions in which differential impact and ex-posure are compared across social classcategories.'" The first panel (A) comparespeople in the middle and upper classes (I-III)with those in the lower and lowest classes(IV-V). The second panel (B) compares thosein classes I-III with the very lowest class (V).With only one exception the results in thesetwo panels are consistent with each other.

In terms of mean differences in self-reporteddistress, the data are quite clear in showingthat the lower classes are disadvantaged in tworespects: They are exposed to more stressfulexperiences than upper class persons, andcomparable events impact on their emotionalfunctioning more severely than on the func-tioning of upper class persons. For instance,line 6b shows that the 2.08-point difference inmean symptoms between persons in classesI-III and V is made up largely of (1) a 1.00-pointdifference due to the greater slopes of distresson the stress measures in the regression for thelower class subsample, and (2) a .68-point dif-ference due to the higher mean levels of stress

in the lower class subsample. The interactionand residual components are comparativelymuch less important, contributing a combineddifference of only .35 points. With only oneexception, the eight contrasts of mean dif-ferences show that the differential impact ofcomparable stressors is more important thandifferential exposure to stress. The exceptionis the weighted prediction of classes I-III vs.classes IV-V in which the Time 1 distress scalewas included as a predictor. Here the exposurecomponent (-.83) is sizably larger than theimpact component (-.39)."

These patterns are consistent with the datareported by Dohrenwend (1973), who foundthat the poor in her urban sample were exposedto more stresses than persons in the middleclass and also that the relationship betweenexposure and distress was higher in the lowerthan in the middle class. However, my resultsare at odds with the findings of Myers et al.(1974) who, analyzing the New Haven data,reported that "the relationship between socialclass and psychiatric symptoms is a reflectionof social class differences in the distribution of

TABLE 3.

Predictors

Decomposition of Class Differences

Impact

A. Classes

in Self-Reported Distress

Exposure Interaction

I-III versus Classes IV-V

Residual Total

Decomposition of Observed Mean Differences (I-Itt-tV-V)la Unweightedlb Unweighted with G,2a Weighted2b Weighted with G,

Decomposition of Predicted3a Unweighted3b Unweighted with G,4a Weighted4b Weighted with G,

Decomposition of Observed5a Unweighted5b Unweighted with G,6a Weighted6b Weighted with G,

-1.87-1.11-3.53- .39

Relative Odds*.52.72

1.47.76

-1.44- .90

.01- .83

(I-ItlllV-V)- .66-.46-.68-.52

B. Classes I-III versus Class V

Mean Differences [I-III-V)-2.97-2.37-2.54-1.00

Decomposition of Predicted Relative Odds*7a Unweighted7b Unweighted with G,8a Weighted8b Weighted with G,

.55

.651.531.79

-2.53-1.17-1.99- .68

(I-IIt/V)-1.36- .82- .81-1.17

.47

.09-.74

.24

.35

.12

.62

.64

1.21.11

1.91-.19

.91

.32

.731.17

1.44.52

2.89- .37

- .38-1.05-1.68-1.53

2.211.34.59

- .16

- .55-1.24-2.23-2.66

-1.37-1.37-1.37-1.37

.84

.52

.77

.52

-2.08-2.08-2.08-2.08

.63

.32

.46

.41

*Logit decompositions are expressed in additive form. The "Total" column is the ratio of mean relativeodds for the contrast categories. For instance, in row 3a, the predicted probabilities of extreme distress forpersons in classes I-III and IV-V at the means of the independent variable are . 113 and .131 respectively. Thisyields a relative odds [.113/(1-.I13)]/[.131/(1-.131)] of .84. This total is equal to the product of theexponentiations of the four components of the decomposition.


Desirability-Change Life Events in the com-munity" (Myers et al., 1974:199-200). Theiranalysis failed to consider impact differentials,thus masking the greater importance of impactthan of exposure in the data.

The second set of decompositions in Table 3focuses on extreme distress. With extremedistress defined as a scale score of 66 or lower,13% of those in classes I-III, 17% of those inclass IV, and 21% of those in class V reportextreme distress at Time 2 of the panel (I-IIIvs. IV-V: t = 1.3, NS; I-III vs. V: t = 2.07, p <.05. The results here are unlike those for meansymptoms in one striking respect. Persons inclasses IV and V are less likely than those inclasses I-III to become extremely distressedwhen exposed to comparable stresses. Despitethis advantage, the poor report more extremedistress. The much greater exposure of thepoor to stressful experiences cancels out theirrelative advantage in effectively bufferingstress impacts.

This result is quite consistent, occurring inall eight decompositions of the set. The impactcomponents are always positive, indicatinglower impacts among the lower classes. Thereis no systematic magnification or diminution ofthe effect when Time 1 distress is controlled.The effect is more evident, though, whenclasses I-III are compared only with class Vthan when compared with IV and V together.

This intriguing pattem will not be pursuedvery deeply here. As we shall see below,though, it appears again in the comparison ofwhites and nonwhites. What we see is thattypical lower class persons respond more tostress than do typical upper class persons. Thisis to be expected, given the lower levels ofsocial support, the less effective coping strat-egies, the secondary stresses set off for lowerclass persons when serious life crises come up,and the myriad other ways in which the poorare disadvantaged relative to the rich in termsof managing stress. However, we find propor-tionally fewer among the poor who respond tostress with really extreme distress than amongthose who are more advantaged economicallyand socially.

One interpretation of this pattern worthy offuture work is that lower class persons areaccustomed to stressful experiences and, as aresult, seldom have extreme emotional re-sponses to stress. That is, objective stressors

might not have the same subjective meaningsfor lower and middle class people. Another isthat lower class people, even though experi-encing the same feelings of-helplessness andhopelessness as middle class people, havelearned to adopt a more fiexible style of emo-tionality. We do not know enough about thesociology of emotions to assess this possibilitysystematically. But it is general clinical lorethat people vary between very rigid and veryflexible styles of emotionality. Flexible peopletend to react quickly and with great emotion tostress, but seem to recover quickly, too. Rigidpeople, on the other hand, tend not to bend,only to break. As a result, when they do be-come emotional they tend to do so uncontrol-lably. It is quite possible that the flexible andrigid styles characterize lower and upper classpersons respectively; such a possibility wouldaccount for the anomalous pattems of stressimpact observed in these data.'^

Race

Of all the contrasts examined in this paper,there is least support in the data, and in theliterature at large (Dohrenwend and Dohren-wend, 1969:31; Warheit et al., 1973, 1975), fora relationship between race and psychologicaldistress. With "nonwhites" defined as those inthe sample who describe themselves as either"Negro" or "Puerto Rican," no significantdifference exists in the average distress scoresof whites and nonwhites (73.0 vs. 71.8, t = 1.4,NS)." However, nonwhites are twice as likelyas whites to report extreme distress (28% vs.14%, t = 2.72, p < .05).

Table 4 represents the results of eight de-compositions between whites and nonwhites inthe sample. I will have little to say about theseresults, for they closely parallel those de-scribed for the social class contrasts in Table 3with respect to the reversal of signs for theimpact components between the decomposi-tion of averages and the decomposition of ex-treme scores. I believe the same considerationsapply here as in the earlier discussion of thoseresults.

There is one respect, though, in which theresults for whites and nonwhites contrast withthe results reported above for social class;namely, that the greater distress among non-whites than among whites is far more a func-

1.07.67

1.07.47

.941.00

.76

.25

- .14- .23

.48-1.35

-1.05-1.07- .68- .60

-1.22-1.22-1.22-1.22

.53

.57

.51

.58


TABLE 4. Decomposition of White-Nonwhite (Negro and Puerto Rican) Differences in Self-Reported Distress

Predictors Impact Exposure Interaction Residual Total

A. Decomposition of Observed Mean Differences (White-Nonwhite)la Unweighted .15 -2.30lb Unweighted with G, -.10 -1.562a Weighted -.67 -2.112b Weighted with G, .86 -1.23

B. Decomposition of Predicted Relative Odds* (WhitelNonwhite)3a Unweighted .71 -1.243b Unweighted with G, .69 -1.194a Weighted .42 -1.164b Weighted with Gi .36 -1.05

Note: Entries have been rounded to two decimal places from computations based on five decimal places.As a result, totals may not correspond exactly to those predicted by adding or multiplying the components.

*See note to Table 3.

tion of differential exposure to stress than was male and female mental illness (Gove, 1972;found to be the case in the greater distress of Gove and Tudor, 1973; Pearlin, 1975; Radloff,lower class persons. As we see in panel A of 1975; Weissman and Klerman, 1977). The em-the table, the average difference in self- phasis in this literature has been on the uniquereported distress between whites and non- experiences of women in our society aswhites is much more a function of the many housewives and mothers and on the stressesstressful experiences to which nonwhites are associated with these experiences. It has beenexposed than of any impact differential be- argued that exposure to these stresses ac-tween whites and nonwhites. Indeed, in two of counts for the preponderance of women amongthe four contrasts the impact coefficient is the mentally ill (Gove and Geerken, 1977).positive, which means that comparable stress- Less has been written about how men andes have more impact on whites than non- women might differ in their vulnerability towhites. Panel B shows that nonwhites have an stress. However, both biological argumentsimpact advantage over whites when considered about female constitutional frailty (Weissmanin terms of extreme distress. However, in none and Klerman, 1977) and socioculturalof the four decompositions is this advantage arguments about learned helplessness (Kaplan,great enough to overcome the greater exposure 1977) and ineffective coping (Pearlin andof nonwhites to stress. Schooler, 1978) suggest that stress may have a

more severe impact on women than on men.o Table 5 presents eight decompositions of

male-female differences in distress. As in otherOver the last decade a great deal has been data reported from sample surveys (Dohren-

written about sex roles and their meanings for wend and Dohrenwend, 1976; Gove and

TABLE 5. Decomposition of Sex Differences in Self-Reported Distress

Predictors Impact Exposure Interaction Residual Total

A. Decomposition of Observed Mean Differences (Men-Women)la Unweighted -1.55 -.31lb Unweighted with G, - .60 -.312a Weighted -3.19 -.192b Weighted with G, -2.58 -.11

B. Decomposition of Predicted Relative Odds* (MenlWomen)3a Unweighted - .92 -.093b Unweighted with Gi - . 3 9 -.114a Weighted -1.15 -.144b Weighted with G, -1.26 -^05



.00

.13

.24

.27

-.09-.02-.07-.07

.24-.841.52.78

.21-.29

.40

.53

-1.62-1.62-1.62-1.62

.41

.44

.38

.43


Tudor, 1973), the data in this study show thatwomen are, on the average, more distressedthan men (72.1 vs. 73.7, t = 2.8, p < .05). Inaddition, women are significantly more likelythan men to report symptoms of extreme dis-tress (19.9% vs. 11.6%, t = 3.1, p < .05). Thedecompositions in both panels A and B showthat differential impact is by far the dominantinfluence on these observed differences.'*

While there is nothing in the data to help usknow why this impact differential exists, amore complete analysis of these data presentedelsewhere" suggests strongly that the lessadequate coping strategies of women, com-pared to those of men, and structural condi-tions that magnify the subjective stressfulnessof objective problems for women account forthe greater impact of comparable stress experi-ences on women than on men.

Marital Status

One of the most consistently documentedfmdings in the epidemiological literature is thatthe married are in better mental health than thenonmarried (Bachrach, 1975). A number ofreasons for this advantage have been hypoth-esized. First, the married are better bufferedpsychologically than the unmarried by virtue ofthe strong emotional attachments that developbetween them and their spouses and children.Next, the married are less exposed to stressfullife situations than the unmarried because oftheir greater financial power, political power,and social contacts. Finally, it has been hy-pothesized (Martin, 1976) that constitutionallyfragile persons are selected out of marriage.This is especially true, according to some ac-

counts, for men, since they must be assertiveand socially competent in order to attract aspouse. Coupled with the assumed disadvan-tages of the unmarried with respect to copingresources, this hypothesis suggests stronglythat the impact of comparable stressors will begreater among the unmarried than among themarried.

In the New Haven data, significant dif-ferences in average and extreme distressscores exist between the married and thosewho are separated or divorced (averages of73.05 vs. 69.19, t = 2.55, p < .05; extremescores of 15% vs. 38%, t = 2.99, p < .05), butnot between the married and either singles orthe widowed. Therefore, the decompositionspresented in Table 6 contrast the married onlywith the separated/divorced. In these resultswe see consistent confirmation of the view thatthe unmarried are disadvantaged both in beingmore exposed to stress and in being morehighly influenced by stress than the married.The relative importance of these two compo-nents varies from one decomposition to an-other, but the results are consistent in theirsigns in all eight decompositions. On balance,it would seem that differential impact is moreimportant than differential exposure in thecomparison of averages and that the two areequally important in the comparison of ex-treme distress. These findings, that both com-ponents have a role to play in the greater dis-tress of the unmarried, are entirely consistentwith recent results of Pearlin and Johnson(1977).

It is not clear from the data why theseparated/divorced are more highly influencedthan the married by comparable stressors.

TABLE 6. Decomposition of Married.-Separated/Divorced Differences in Self-Reported Distress

Predictors Impact Exposure Interaction Residual TotalA. Decomposition of Observed Mean Differences (Married—Separated-Divorced)

la Unweighted -4.47 -3.17 .94lb Unweighted with G, -4.71 -2.33 .892a Weighted -7.65 - .27 -1.512b Weighted with G, -5.52 - . 0 8 -1.02

B. Decomposition of Predicted Relative Odds* (MarriedlSeparated-Divorced)3a Unweighted -2.56 -1.79 1.243b Unweighted with G, -2.56 -3.48 3.184a Weighted -1.09 - .80 .384b Weighted with G, -3.33 -1.43 1.19



2.832.305.582.75

1.671.56

-.101.57

-3.87-3.87-3.87-3.87

.24

.27

.19

.14


Unfortunately, the sample is too small to com-pare the impact coefficients across pairwisecontrasts of the married with the separated, thedivorced, the widowed, and the single. It ispossible that a detailed comparison of this sortwould be able to shed further light on the de-terminants of the impact differential, for, asGove (1972) has pointed out, any effect thatselects constitutionally fragile people out ofmarriage will characterize singles and possiblythe separated/divorced, but not the widowed.However, the sort of social support disadvan-tages that accompany being single probablyexist for the widowed as well as for singles andthe separated/divorced. Therefore, the socialselection argument would predict that the im-pact differentials found in Table 6 would fail toappear in contrasts of the married with thewidowed. But the argument that differentialsocial resources account for the greater dis-tress of the unmarried would predict that thevery same impact component would appear inthe married vs. widowed contrast as wasfound here for the married vs. separated/divorced contrast.

DISCUSSION

The most general conclusion to be drawnfrom the above results is that differential im-pact is an important component in most con-trasts between socially advantaged and disad-vantaged status categories. Over the last de-cade there has been a tendency for researchersto focus too narrowly on the hypothesis that"persons in low social status are dispropor-tionately exposed to stressful life events, andthat this exposure can explain the highersymptom levels of those with low socialstatus" (Markush and Favero, 1974:184). Thisdifferential exposure hypothesis is central toGove's (1972) sex-role theory of mental illnessand to the work of the Dohrenwends (1969) andMyers et al. (1974) on social class and mentalillness, to name a few of the most prominentexamples. The results of my decompositionanalysis show that differential exposure is onlypart of the story for the contrasts examined inthis study. For sex differences, in fact, there isno evidence that differential exposure playsany meaningful role in the greater distress ofwomen.

As I mentioned above, it is extremely dif-

ficult to disentangle the effects of constitu-tional makeup, social resources, and copingstyles on the strength with which stress im-pacts on emotional functioning. Any or all ofthese influences could modify the impact ofstress. However, the existence of negligibleimpact components or impact components thatshow low-status persons to be advantagedallow us to make certain inferences about theforces at work in certain contrasts.

The most obvious point to be made is thatcontrasts dominated by differential exposurecomponents can usually be assumed to be un-influenced by constitutional frailty. In thesimplest sense this is so because such effectswould appear in substantial impact compo-nents. However, a more complex sort of rea-soning, suggested by Dohrenwend (1966), addsextra weight to this claim. Let us take whitesand nonwhites in Table 5. Here differences inmean symptom scores are found to be dueoverwhelmingly to the greater stresses towhich nonwhites are exposed. There is nomeaningful impact component in two of thedecompositions, and the signs of the compo-nents vary in the remaining decompositions. Itmight be, then, that whites and nonwhitesshare the same sorts of coping strategies andresources, and also share the same aggregatecharacteristics in terms of constitutionalstrengths and frailties. However, to the extentthat differences do exist it is almost certainlythe case that nonwhites are disadvantagedsocioculturally relative to whites (Dohrenwendand Dohrenwend, 1969:49-56). Therefore, ifthere are any variations in intrapsychicstrengths, whites must be disadvantaged rela-tive to nonwhites. The argument here is thatthe insubstantial impact component found inthe white/nonwhite decomposition can beplausibly interpreted as evidence either ofcompletely absent impact effects—includingany due to constitutional makeup—or ofcounterbalancing impact effects of more thanone sort. In the latter interpretation the onlyplausible line of thought suggests that non-whites have a stronger constitutional makeupthan whites.

A similar sort of argument might be made forthe relationship between social class and psy-chological distress. At least this would seem tobe the case for the contrasts of extreme dis-tress, where the impact coefficients all show


lower class persons to be less highly influencedby stress than their more socially advantagedcounterparts in the middle and upper classes.However, the impact components in the de-compositions of mean symptom scores are allpositive, which means that impact is higher inthe lower class. Much more analysis of thisdiscrepancy between the two types of impactcomponents will be necessary before anythingcan be said about the importance of constitu-tional differences in the relationship betweensocial class and distress.

For the remaining two contrasts, betweenmen and women and between the married andthe separated/divorced, important impactcomponents appear in the results. Indeed, forboth contrasts it would seem that differentialimpact is more important than differential ex-posure. Here we come to the grey area in ourunderstanding of sociodemographic variationsin emotional distress. We know that womenand the unmarried suffer because of relativeinabilities to buffer themselves from the emo-tional distress accompanying life stresses. Butwhy?

Without direct measures of some of the fac-tors that might plausibly lead to the differentialinfluence of stress, it is extremely difficult toanswer this question. Although there are anumber of theories about the genetic,neurochemical, and neurobiological factorsthat might go into constitutional differences,there is no obvious way to index such dif-ferences in survey data. However, it is possibleto measure coping styles (Pearlin andSchooler, 1978), social support (Cobb, 1976),and many other sociocultural factors that canbuffer the impact of stress on distress. By usingmeasures of this sort to interpret impact com-ponents, we might be able to tease out thecontributions of the various plausible determi-nants of differential impact described above.

I have not been able to do this in the presentpaper. However, I feel that future work mustmove in this direction if we are to disentanglethe thorny problems of causation that have solong baffled those of us who study thesociodemographic correlates of mental illness.

NOTES

1. Some emotional problems are less characteristicof socially disadvantaged than of advantagedpersons. For instance, women are less likely to

be alcoholics than men (Weissman and Klerman,1977), and the poor are less likely to beobsessive-compulsive than the rich or upwardlymobile (Langner and Michael, 1963). There issome clinical speculation, in fact, that certainclusters of problems are functionally equivalent,but are structurally more likely to appear in cer-tain persons than in others. Hysteria in women isthought to be an equivalent of sociopathy inmen; depression in women to be an equivalent ofalcoholism in men. Were it possible to formalizeand substantiate these equivalences it might wellbe that we would conclude, not, as now, thatdisadvantaged persons are more distressed, butonly that they are distressed differently thantheir more advantaged counterparts. At thepresent time, though, it is not clear how to makethese equivalences formally. The best strategywould seem to be to conduct separate analysesof multiple domains of distress, coupled withsystematic assessments of the extent to whichcausal processes found to trigger one type ofproblem in one particular category of peopleparallel the triggering of other types of problemsin other categories of people.

2. There are several different versions of the "so-cial selection" argument. One of these suggeststhat early psychological disability differentiallypredisposes an individual to fail in economic af-fairs and in the marriage market. To the extentthat the disability is permanent, then, suchpeople will tend to inflate aggregate measures ofpsychological disability and distress in thesestatus categories. However, since this argumentmakes no reference to impaired coping abilities,it is not obvious that the people who " drift'' intothese disadvantaged social statuses will be morehighly influenced than other people by stressesthat they come to experience as members ofthese status categories.

A broader selection argument, on the otherhand, suggests that predispositions toward dis-tress intemal to the individual, determined, inpart, by the early childhood environment, and,in part, by genetic, neurochemical, andneurobiological factors unrelated to the adultenvironment, are more common among the poorthan the well-to-do, the unmarried than the mar-ried, women than men, and nonwhites thanwhites. For the poor and for the unmarried suchdisadvantages in social competence must be dueto some sort of social selection process—sincethese are both acquired statuses. For women andnonwhites, though, arguments about compe-tence usually are based either on factors intrinsicto body chemistry (for women) or on interac-tions between prenatal environmental assaultand genetic predispositions (for nonwhites).

3. At the same time, these life situations canprovide resources for dealing with other types ofstress. For instance, the lack of fmancial re-sources has been hypothesized to be a centraldeterminant of the coping disadvantages ofblacks and persons in the lower class (Dohren-


wend and Dohrenwend, 1970). Similarly; physi-cal health can be a resource on which one candraw in coping with certain events that requirephysical intervention. It is generally the case, infact, that ongoing life situations or char-acteristics that can be thought of as stresses can,when considered from another perspective, beconsidered resources. It is possible, then, thatcomplex interactions exist in the predictionequations considered in this paper. For instance,physical health problems might influence theemotional functioning of a person who lacks fl-nancial resources more severely than one whopossesses flnancial resources. Interactions ofthis type, though, will not be examined in theanalyses that follow.

4. Since the scale of emotional distress used in thisanalysis consists largely of psychophysiologicalsymptoms, great care was taken in constructingthe physical illness index. The interviewschedule contained a list of some 44 physicalhealth complaints. Some of these were clearlypsychosomatic. To create the scale of physicalillness, I selected a subset of 16 complaints forwhich a psychosomatic component is probablyof only minor importance. I arbitrarily gave dou-ble or triple weights to some of those that Iconsidered on commonsensical grounds to bethe most serious.

5. Of the 56 events in the interview schedule, 28 aredeflned as undesirable, 14 as desirable, and 14 asambiguous. The ambiguous-events subsczile hasbeen omitted here in order to avoid collinearitybetween the total-events scale and theundesirable- and desirable-events subscales. Theoverall results of the decomposition are unaf-fected by the decision to exclude theambiguous-events scale rather than one of theothers. However, this decision would becomemore important if one were interested inevaluating the signiflcance of specific slope dif-ferences (see Note 10), in which case it would bepreferable to include all the component scalesand exclude the scale of total events.

6. The Myers Desirability-Change Index is drawnup in two steps. First, weighted desirable- andundesirable-event indexes are created that areidentical to those I use in this analysis. Next, thetwo indexes are combined, the desirable-eventindex being given a positive score and theundesirable-event index a negative score. Re-spondents who report no events are given ascore higher than the highest empirical scoreobtained among those who report at least oneevent.

The use of three indexes in the procedure fol-lowed in this paper allows the weighting of thedesirable and undesirable indexes (implicitlygiven equal weight in the Myers procedure) aswell as their signs (constrained to be + , - for thedesirable and undesirable indexes in the Myersprocedure) to be estimated empirically. In addi-tion, the use of a dummy variable for those whoreport no events allows empirical estimation of

their position on the index rather than the con-strained assignment of them to the highest score.Given the fact that my three indexes relax threeassumptions of the Myers index, it is not sur-prising to flnd, as I did in preliminary compari-sons, that my three indexes always have higherR2 values in subsample regressions than does theMyers Desirability-Change Index.

7. Since the stress measures correlate signiflcantlywith the Time 2 scale, controlling for the Time 1score, we know that some real change has takenplace in symptoms over time. On the basis of theregression results, this change component can beestimated to be a minimum of 20% of the ob-served Time 2 symptom scale variance. On thebasis of this information, we can adjust the esti-mate of the scale's reliability upward to aminimum of .6.

8. The decomposition in (2) perfectly decomposesobserved mean differences in distress. The de-composition in (4) perfectly decomposes theratio of the mean predicted log odds of extremedistress. Although the mean of the predictedprobabilities does equal the observed probabilityof extreme distress in the sample, the mean ofthe predicted log odds does not equal the log ofthe observed mean odds. Consequently, the"Total" columns in Tables 3-6 vary for pre-dicted relative odds, even though they do notvary for observed mean differences.

9. One possible reason for this is that the event listsare shorter in the weighted scales than in theunweighted scales. Following Myers, only thoseevents that were included in the Holmes andRahe (1967) events list were made part of theweighted event scales. In the unweighted scales,28 undesirable, 14 desirable, and 56 total eventswere included. In the weighted scales, only 16undesirable and 10 desirable events were in-cluded.

10. The theory necessary to develop signiflcancetests for decomposition components has only re-cently been worked out by Michael Sobel at theUniversity of Wisconsin. No machine-readableprocedures for making the tests yet exist.Therefore, no signiflcance tests are presented forthe decompositions in Tables 3-6. However, it ispossible to evaluate the signiflcance of meandifferences across contrast subsamples with t-tests and to evaluate the signiflcance of slopedifferences either with t-tests or pooled regres-sion models containing interaction terms. Anexample of this sort of analysis is given in Kess-ler (1979:Note 4). However, since the purposehere is to describe the major pattems in the data,rather than to make inferences to some broaderpopulations, this type of signiflcance-testingprocedure has not been employed.

11. One reviewer suggested that this exception de-serves attention because it occurs in one of themost realistic speciflcations. However, it shouldbe remembered that the weighted scales containfewer items than the unweighted scales (seeNote 9), making inconsistent results based on


models using these scales somewhat less persua-sive than they otherwise might be. In addition,substantive interpretation of the result in row 2bwould imply that the same reversal should befound in row 6b. That this reversal does notappear calls a substantive interpretation furtherinto question.

12. It is important to distinguish this clinical notionof emotional flexibility from Kohn's (1969) no-tion of intellectual flexibility. A reviewer of anearlier draft commented that there seemed to bea contradiction between my speculation thatlower class persons might be emotionally flexi-ble and Kohn's finding that lower class personsare intellectually inflexible. No such contradic-tion exists, though. Kohn's flnding refers to theidea that persons in the lower class "experienceconditions of life that foster conceptions of so-cial reality so limited and so rigid as to impairtheir ability to deal resourcefully with the prob-lematic and stressful" (Kohn, 1977:xliv). Theclinical notion of emotional flexibility, on theother hand, refers to the idea that people can bequick or slow to hold back emotional reactionsfrom themselves and others. It is perfectly con-ceivable that lower class persons could be bothintellectually inflexible and emotionally flexible.

13. The extremely small numbers of respondentswho were assigned codes as "Japanese,""Chinese," "Other," or "Unknown" were ex-cluded from this contrast. Only 90 respondentswere assigned codes as either "Negro" or"Puerto Rican." These were combined, sinceneither of these two response categories con-tained sufflcient respondents to warrant separateanalysis. It should be noted, though, that theDohrenwends (1969) have reported certain dif-ferences between blacks and Puerto Ricans onscales like the Gurin scale, some of which mightreflect differences in response styles. The resultsobtained for the combined subsample in thispaper, then, should be interpreted with caution.

14. The stresses that have been discussed in thesex-role literature as key determinants of dis-tress among women are thought to be sex-rolestresses, ones that are relatively unique towomen. The impact of these stresses is capturedin the residual components in Table 5. As we seethere, though, women have positive residuals insix of the eight decompositions, which suggeststhat they are less distressed than men net theinfluences modeled explicitly in the predictionequations. Were unique sex-role stresses actu-ally at work, we would expect to have foundnegative residuals.

15. Data available from the author on request.

REFERENCES

Bachrach, Leona L.1975 Marital Status and Mental Disorder: An

Analytical Review. DHEW Publication No.(ADM) 75-217. \yashington, D.C.: U.S.Government Printing Office.

Brown, G. W.1974 "Meaning, measurement, and stress of life

events." Pp. 217-43 in Barbara S. Dohren-wend and Bruce P. Dohrenwend (eds.).Stressful Life Events: Their Nature andEffects. New York: Wiley-Interscience.

Cobb, S.1976 "Social support as a moderator of life

stress." Psychosomatic Medicine 38:300-314.

Dean, A., and N. Lin1977 "The stress-buffering role of social support:

Problems and prospects for systematic in-vestigation." Paper presented at the annualmeeting of the American SociologicalAssociation, Chicago.

Dohrenwend, B. P.1966 "Social status and psychological disorder:

An issue of substance and an issue ofmethod." American Sociological Review31:14-34.

1974 "Problems in defming and sampling therelevant population of stressful lifeevents . " Pp. 275-310 in Barbara S.Dohrenwend and Bruce P. Dohrenwend(eds.). Stressful Life Events: Their Natureand Effects. New York: Wiley.

Dohrenwend, Bruce P., and Barbara S. Dohrenwend1969 Social Status and Psychological Disorder.

New York: Wiley.1976 "Sex differences and psychiatric disor-

ders." American Journal of Sociology81:1447-54.

Dohrenwend, B. S.1973 "Social status and stressful life events."

Journal of Personality and Social Psychol-ogy 9:203-14.

Dohrenwend, B. S., and B. P. Dohrenwend1970 "Class and race as status-related sources of

stress." Pp. 111-40 in Sol Levine andNorman A. Scotch (eds.). Social Stress.Chicago: Aldine.

Dohrenwend, B. S., L. Krasnoff, A. R. Askenasy,and B. P. Dohrenwend

1978 "Exemplification of a method for scalinglife events: The PERI Life Events Scale."Journal of Health and Social Behavior19:205-29.

Faris, Robert, and H. Warren Dunham1939 Mental Disorders in Urban Areas. New

York: Hafner.Gove, W. R.

1972 "The relationship between sex roles, mari-tal status and mental illness." Social Forces51:34-44.

Gove, W. R., and M. R. Geerken1977 ' 'The effect of children and employment on

the mental health of married men andwomen." Social Forces 56:66-76.

Gove, W. R., and J. F. Tudor1973 "Adult sex roles and mental illness."

American Journal of Sociology 98:812-35.Gurin, Gerald, Joseph Veroff, and Sheila Feld

1960 Americans View Their Mental Health. NewYork: Basic Books.


Hollingshead, August B., and Fredrick C. Redlich1958 Social Class and Mental Illness. New York:

Wiley.Holmes, T. H., and M. Masuda

1974 "Life change and illness susceptibility."Pp. 45-72 in Barbara S. Dohrenwend andBruce P. Dohrenwend (eds.). Stressful LifeEvents: Their Nature and Effects. NewYork: Wiley.

Holmes, T., and R. H. Rahe1967 "The social readjustment rating scale."

Journal of Psychosomatic Research11:213-18.

Kaplan, B. H., and J. C. Cassel1977 "Social support and health." Medical Care

15:47-58.Kaplan, H. B.

1977 "Gender and depression: A sociologicalanalysis of a conditional relationship." Pp.81-113 in William E. Fann, lsmet Karacan,Alex D. Pokorny, and Robert L. Williams(eds.). Phenomenology and Treatment ofDepression. New York: Spectrum Publica-tions.

Kessler, R. C.1979 "A strategy for studying differential vul-

nerability to the psychological conse-quences of stress." Journal of Health andSocial Behavior 20:100-108.

Kohn, Melvin L.1969 Class and Conformity: A Study in Values.

Homewood, 111.: Dorsey.1977 "Reassessment, 1977." Pp. xxv-lx in Class

and Conformity, 2nd ed. Chicago: Univer-sity of Chicago Press.

Langner, Thomas S., and Stanley T. Michael1963 Life Stress and Mental Health. New York:

Free Press.Liem, R., and J. Liem

1978 "Social class and mental illness reconsid-ered: The role of economic stress and socialsupport." Journal of Health and Social Be-havior 19:139-56.

Macmillan, A. M.1957 "The Health Opinion Survey: Technique

for estimating prevalence of psychoneuroticand related types of disorder in com-m u n i t i e s . " Psychological Repor ts3:325-29.

Markush, R. E., and R. V. Favero1974 "Epidemiologic assessment of stressful life

even t s , depressed mood, andpsychophysiological symptoms—A pre-liminary report." Pp. 171-90 in Barbara S.Dohrenwend and Bruce P. Dohrenwend(eds.). Stressful Life Events: Their Natureand Effects. New York: Wiley.

Martin, W. T.1976 "Status integration, social stress, and

mental illness: Accounting for maritalstatus variations in mental hospitalizationrates." Journal of Health and Social Be-havior 17:280-94.

Mechanic, D.1974 "Social structure and personal adaptation:

Some neglected dimensions." Pp. 32-44 inGeorge V. Coelho, David A. Hamburg, andJohn E. Adams (eds.). Coping and Adapta-tion. New York: Basic Books.

Myers, J. K., J. J. Lindenthai, and M. P. Pepper1974 "Social class, life events, and psychiatric

symptoms: A longitudinal study." Pp.191-206 in Barbara S. Dohrenwend andBruce P. Dohrenwend (eds.). Stressful LifeEvents: Their Nature and Effects. NewYork: Wiley.

1975 "Life events, social integration and psychi-atric symptomatology." Joumal of Healthand Social Behavior 16:421-27.

Myers, J. K., J. J. Lindenthai, M. P. Pepper, and D.R. Ostrander

1972 "Life events and mental status: A lon-gitudinal study." Journal of Health and So-cial Behavior 13:398-406.

Pearlin, L. I.1975 "Sex roles and depression." Pp. 191-207 in

Nancy Datan and Leon Ginsberg (eds.).Proceedings of Fourth Life-Span Devel-opmental Psychology Conference: Norma-tive Life Crises. New York: Academic.

Pearlin, L. I., and J. S. Johnson1977 "Marital status, life-strains and depres-

sion." American Sociological Review42:104-15.

Pearlin, L. I., and C. Schooler1978 "The structure of coping." Journal of

Health and Social Behavior 19:2-21.Rabkin, J. G., and E. L. Stmening

1976 "Life events, stress and illness." Science194:1013-20.

Radloff, Lenore1975 "Sex differences in depression: The effects

of occupation and marital status." SexRoles 1:249-65.

Vinokur, A., and M. L. Selzer1975 "Desirable versus undesirable life events:

Their relationship to stress and mental dis-tress." Joumal of Personality and SocialPsychology 32:329-37.

Warheit, G. I., C. E. Holzer III, and S. A. Arey1975 "Race and mental illness: An epidemiologic

update." Journal of Health and Social Be-havior 16:243-56.

Warheit, G. I., C. E. Holzer III, and J. J. Schwab1973 "An analysis of class and racial differences

in depressive symptomatology: A commu-nity study." Journal of Health and SocialBehavior 14:291-99.

Weissman, M. M., and G. L. Klerman1977 ' ' Sex differences in the epidemiology of de-

pression." Archives of General Psychiatry34:98-111.

Winsborough, H. H., and P. Dickinson1971 "Components of Negro-White income dif-

ferences." Proceedings of the SocialStatistics Section. Washington: AmericanStatistical Association.

Stress, Social Status, and Psychological Distress · Stress, Social Status, and Psychological Distress RONALD C. KESSLER University of Wisconsin—Madison Journal of Health and Social

Documents