13
Journal of Personality and Social Psychology 1993, Vol. 64, No. 5, 794-806 Copyright J 993 by the American Psychological Association, Inc. 0022-3514/93/S3.00 Gender and the Relationship Between Job Experiences and Psychological Distress: A Study of Dual-Earner Couples Rosalind C. Barnett, Nancy L. Marshall, Stephen W Raudenbush, and Robert T. Brennan This article examines the association between job role quality and psychological distress in a sample of 300 full-time employed dual-earner couples, controlling for such individual level vari- ables as age, education, occupational prestige, and marital quality and for such couple level vari- ables as length of marriage, parental status, and household income. The magnitude of this effect is compared for men and for women. Results indicate that job role quality is significantly negatively associated with psychological distress for women as well as for men and that the magnitude of the effect depends little, if at all, on gender, casting doubt on the widely held view that job experiences more significantly influence men's mental health states than women's. The results are discussed in the context of differentiating between sex differences and gender differences. As more and more people are living in dual-earner families, concern is being voiced about the stresses inherent in this life- style and their consequences for the mental health of men and women (Kessler & McRae, 1982; Ross, Mirowsky, & Huber, 1983). Although it is widely believed that workplace stress has differential impact on the distress of men and women, surpris- ingly little attention has been paid to gender differences in the relationship between job quality and psychological distress lev- els. If the workplace more closely regulates the psychological fate of men than of women, the association between job role quality and psychological distress ought to be stronger for men than for women. This study examined the linkage between job quality and psychological distress levels for men and women in a random sample of 300 dual-earner couples. Research concerning the job stress-psychological distress re- lationship in dual-earner couples has been severely limited by unwarranted assumptions and methodologies that were less credible and precise than those now available. In this section, we first discuss the assumptions underlying the belief that gender differences influence the relationship between job expe- riences and psychological distress. Then, we review previous research findings and analytic strategies used within psychol- ogy and sociology for handling data from couples. Finally, we propose a new strategy for analyzing gender effects in a sample of dual-earner couples. The relationship between job role quality and psychological distress has received considerable research attention (Barnett, Marshall, & Singer, 1992; Frankenhaeuser, Lundberg, & Ches- ney, 1991; Karasek & Theorell, 1990; Repetti, Matthews, & Waldron, 1989). For men and women, poor job quality has been associated with psychological distress, both cross-section- ally (Baruch & Barnett, 1986; Barnett & Marshall, 1991; Bar- nett, Marshall, & Pleck, 1992) and longitudinally (Barnett, Marshall, & Singer, 1992; House, Strecher, Metzner, & Robbins, 1986). 1 Prevailing assumptions suggest that this relationship will be stronger among men than among women. More specifically, it is generally believed that the job role is crucial to men's mental health, whereas family roles are secondary. In contrast, family roles are construed as crucial to women's mental health and the job role as secondary (Barnett & Baruch, 1987). As an illustra- tion of the power of these assumptions, Gurin, Veroff, and Feld (1960), in a classic study of mental health in the United States, questioned male respondents about their work and home lives and female respondents about their home lives only. This strat- egy was justified on the grounds that work was emotionally relevant only to men. 2 These assumptions persist (Simon, 1992; Thoits, 1992) de- spite the fact that they have been severely challenged on empiri- cal grounds (Rodin & Ickovics, 1990; Wethington & Kessler, 1989; for a recent review see Barnett, 1993). Here, we challenge Rosalind C. Barnett and Nancy L. Marshall, Center for Research on Women, Wellesley College; Stephen W Raudenbush, Department of Counseling, Educational Psychology and Special Education, Michi- gan State University; Robert T. Brennan, Graduate School of Educa- tion, Harvard University. Data for this study were collected with funding from the National Institute of Mental Health (MH 43222). We thank our dedicated inter- viewing team: Carol Anello, Joyce Buni, Lillian Coltin, Connie Festo, Carla Fink, Lorraine McMullin, Pam Miller, Jennifer Rochow, Rosa- lind Sandier, and Kathryn Wheeler. Correspondence concerning this article should be addressed to Ro- salind C. Barnett, Center for Research on Women, Wellesley College, Wellesley, Massachusetts 02181. ' Although there is no universal consensus, the combination of high work demands (overload) and low control has been related to high strain and ultimately to health problems (Karasek & Theorell, 1990). This model has received some empirical support (Karasek et al., 1982), more so for men than for women (but see LaCroix & Haynes, 1987). Alternatively, the combination of high overload and low opportunity to help others on the job has been associated with several mental and physical health outcomes among women in the helping professions (Barnett & Marshall, 1991). 2 In the follow-up study, The Inner American, the authors question both men and women about work and family (Veroff, Douvan, & Kulka, 1981). 794

Document (66)

Embed Size (px)

DESCRIPTION

Document (66)

Citation preview

Journal of Personality and Social Psychology1993, Vol. 64, No. 5, 794-806

Copyright J 993 by the American Psychological Association, Inc.0022-3514/93/S3.00

Gender and the Relationship Between Job Experiences and PsychologicalDistress: A Study of Dual-Earner Couples

Rosalind C. Barnett, Nancy L. Marshall, Stephen W Raudenbush, and Robert T. Brennan

This article examines the association between job role quality and psychological distress in asample of 300 full-time employed dual-earner couples, controlling for such individual level vari-ables as age, education, occupational prestige, and marital quality and for such couple level vari-ables as length of marriage, parental status, and household income. The magnitude of this effect iscompared for men and for women. Results indicate that job role quality is significantly negativelyassociated with psychological distress for women as well as for men and that the magnitude of theeffect depends little, if at all, on gender, casting doubt on the widely held view that job experiencesmore significantly influence men's mental health states than women's. The results are discussed inthe context of differentiating between sex differences and gender differences.

As more and more people are living in dual-earner families,concern is being voiced about the stresses inherent in this life-style and their consequences for the mental health of men andwomen (Kessler & McRae, 1982; Ross, Mirowsky, & Huber,1983). Although it is widely believed that workplace stress hasdifferential impact on the distress of men and women, surpris-ingly little attention has been paid to gender differences in therelationship between job quality and psychological distress lev-els. If the workplace more closely regulates the psychologicalfate of men than of women, the association between job rolequality and psychological distress ought to be stronger for menthan for women. This study examined the linkage between jobquality and psychological distress levels for men and women ina random sample of 300 dual-earner couples.

Research concerning the job stress-psychological distress re-lationship in dual-earner couples has been severely limited byunwarranted assumptions and methodologies that were lesscredible and precise than those now available. In this section,we first discuss the assumptions underlying the belief thatgender differences influence the relationship between job expe-riences and psychological distress. Then, we review previousresearch findings and analytic strategies used within psychol-ogy and sociology for handling data from couples. Finally, wepropose a new strategy for analyzing gender effects in a sampleof dual-earner couples.

The relationship between job role quality and psychologicaldistress has received considerable research attention (Barnett,Marshall, & Singer, 1992; Frankenhaeuser, Lundberg, & Ches-ney, 1991; Karasek & Theorell, 1990; Repetti, Matthews, &Waldron, 1989). For men and women, poor job quality hasbeen associated with psychological distress, both cross-section-ally (Baruch & Barnett, 1986; Barnett & Marshall, 1991; Bar-nett, Marshall, & Pleck, 1992) and longitudinally (Barnett,Marshall, & Singer, 1992; House, Strecher, Metzner, & Robbins,1986).1

Prevailing assumptions suggest that this relationship will bestronger among men than among women. More specifically, itis generally believed that the job role is crucial to men's mentalhealth, whereas family roles are secondary. In contrast, familyroles are construed as crucial to women's mental health and thejob role as secondary (Barnett & Baruch, 1987). As an illustra-tion of the power of these assumptions, Gurin, Veroff, and Feld(1960), in a classic study of mental health in the United States,questioned male respondents about their work and home livesand female respondents about their home lives only. This strat-egy was justified on the grounds that work was emotionallyrelevant only to men.2

These assumptions persist (Simon, 1992; Thoits, 1992) de-spite the fact that they have been severely challenged on empiri-cal grounds (Rodin & Ickovics, 1990; Wethington & Kessler,1989; for a recent review see Barnett, 1993). Here, we challenge

Rosalind C. Barnett and Nancy L. Marshall, Center for Research onWomen, Wellesley College; Stephen W Raudenbush, Department ofCounseling, Educational Psychology and Special Education, Michi-gan State University; Robert T. Brennan, Graduate School of Educa-tion, Harvard University.

Data for this study were collected with funding from the NationalInstitute of Mental Health (MH 43222). We thank our dedicated inter-viewing team: Carol Anello, Joyce Buni, Lillian Coltin, Connie Festo,Carla Fink, Lorraine McMullin, Pam Miller, Jennifer Rochow, Rosa-lind Sandier, and Kathryn Wheeler.

Correspondence concerning this article should be addressed to Ro-salind C. Barnett, Center for Research on Women, Wellesley College,Wellesley, Massachusetts 02181.

' Although there is no universal consensus, the combination of highwork demands (overload) and low control has been related to highstrain and ultimately to health problems (Karasek & Theorell, 1990).This model has received some empirical support (Karasek et al., 1982),more so for men than for women (but see LaCroix & Haynes, 1987).Alternatively, the combination of high overload and low opportunity tohelp others on the job has been associated with several mental andphysical health outcomes among women in the helping professions(Barnett & Marshall, 1991).

2 In the follow-up study, The Inner American, the authors questionboth men and women about work and family (Veroff, Douvan, &Kulka, 1981).

794

GENDER, JOB EXPERIENCES, AND DISTRESS 795

these assumptions on the grounds that they confound the ef-fects of sex with those of gender. Sex differences refer to innatedifferences between males and females, differences attribut-able to hormonal, morphological, or other sex-linked biologicalvariations. True sex differences, by definition, are relativelyinvariant over time and across cultures. More exactly, invariantsex differences are attributable to Invariant Sex X HistoricalContext interactions, because biological sex per se typicallyconnotes a tendency that is manifest given the appropriate cul-tural stimulus (Bern, 1993).

Because there is general agreement that, from birth on, thesocialization experiences of male and female babies differ, it isvery hard for social scientists to ferret out sex differences. Yetclaims about essential "maleness" and "femaleness" are oftenmade or inferred. For example, in his seminal writing on thefamily, Parsons (Parsons & Bales, 1955) attributed the inevita-ble and functional asymmetry in work and family roles to thebiological fact that women bear children. In his words,

The fundamental explanation of the allocation of the roles be-tween the biological sexes lies in the fact that the bearing and earlynursing of children establish a strong presumptive primacy of therelation of the mother to the small child and this in turn estab-lishes a presumption that the man, who is exempted from thesebiological functions, should specialize in the alternative instru-mental direction, (p. 355)

Also reflecting his belief in the inevitability of these distinc-tions, he went on to say

The broad structural outlines of the American nuclear family, aswe have delineated it, are not 'fortuitous' in the sense of beingbound to a particular highly specific social situation, but are ofgeneric significance with respect to the structure and functions ofthe family in all societies, (p. 355)

This highly influential formulation of the sex-linked relativesalience of work and family roles for men and women, and byextension, their relative influence on psychological distress,has had a major influence on research. By not addressing theeffect of historical period on hypotheses concerning the effectof gender on relationships between social role experiences anddistress, researchers treat these effects as if they were time in-variant, that is, sex not gender differences. Indeed gender dif-ferences in social roles is central to many studies of the socialrole-distress relationship; however, results are discussed pri-marily in the context of "sex [not gender] differences in depres-sion" (Cleary & Mechanic, 1983, p. 115).

Gender, in contrast, is a construct consisting of biological,psychological, and sociological meanings. Following Ortnerand Whitehead (1981) and Bern (1993), we define gender asincluding, but not limited to, biological sex, differential sociali-zation experiences, individually held expectations for sociallydefined, gender-specific roles, and sex role attitudes. In ourview, gender is primarily the product of social and culturalprocesses. By definition, gender varies over time and acrosscultures. Confusion arises when differences between men andwomen, differences reflecting a particular culture and time,are treated as sex differences, that is, as natural differences and,therefore, invariant over time. Such is true for many stronglyheld views in our culture about male-female differences in therelationships between role experiences and mental health.

Thus, these differences are treated as if they persist indepen-dent of time or place.

Within the social sciences, these assumptions took holdfirmly in the 1950s (Parsons, 1949; Parsons & Bales, 1955), atime when men and women operated in rigidly segregatedworlds. These assumptions, however valid they may have beenat that time, have taken on a normative and prescriptive ur-gency that still strongly affects our lives (Spence, Deaux, &Helmreich, 1985). Although in the 1950s men were engaged inboth job and family roles, their sense of self was tied to theirsuccess on the job. Indeed, achievement at the workplace wasconstrued as a necessary precondition for healthy male develop-ment. Moreover, success on the job established men's success inthe family; being a good husband and father was defined interms of being a good provider. Even though significant num-bers of women were in the labor force at that time (Crosby,1991; Faludi, 1991; Skolnick, 1991), the only socially sanctionedlife pattern for women centered around the home; being a wifeand mother was tantamount to being a psychologically healthywoman. Accordingly, it was believed that "the work place andits events, in our society, more closely regulate the psychologicalfate of men than of women" (Pearlin, 1975, p. 202).

Given the minimal overlap in men's and women's social roles,it was impossible to disentangle the effects of sex from those ofsocial roles in developing any theory of sex differences. "Vetdifferences in such "instrumental" characteristics as aggressive-ness, competitiveness, and dominance and such "expressive"characteristics as emotionality, nurturance, and sensitivity wereattributed to sex (not gender) differences. Clearly, it is only ifand when men and women have similar experiences and livesimilar lives that one can hope to detect sex differences. Thisobservation is consistent with the growing consensus (Deaux,1984; Thoits, 1992) that, at any point in time, the main effects ofsex on a host of behavioral variables "are less pervasive thanmany thought" and that situational variables are critical ineliciting or suppressing differences (Deaux, 1984, p. 108). Thisline of reasoning suggests that the behavioral repertoires ofmen and women are basically the same. However, context andhistorical period differences influence men's and women's se-lection of behaviors from this repertoire. Job and family rolesreflect, therefore, not innate differences in repertoires but dif-ferences in choices. In other words, "enacted differences mayderive from similar potentials" (Deaux, 1984, p. 114).

Having concluded that the observed behavioral and psycho-logical differences were "sex" differences (Parsons & Bales,19 5 5), it is not surprising that they were assumed to persist evenas women's lives became more similar to men's. As more andmore married women entered the labor force and "woman asworker" became the norm, these assumptions colored thinkingabout the relationship between employment and psychologicaldistress. For women, employment was viewed as the added-onrole; by virtue of taking on the employee role, married women(especially married women with children) were by definitionunder increased stress and exposed to high and potentiallyharmful levels of role stress and role conflict. Indeed somestudies reported higher levels of distress among employed com-pared with nonemployed women (Hochschild & Machung,1989).

In the 1990s, men and women are increasingly living in

796 BARNETT, MARSHALL, RAUDENBUSH, AND BRENNAN

worlds that are far more similar than different: Mothers areemployed full-time year around; men and women are increas-ingly working at similar jobs; and husbands and fathers areexpected to participate in the family, in addition to bringinghome the bacon. As the roles in which men and women operatebecome more similar, the assumed "sex" differences in the re-lationship between work role and family role quality on the onehand and psychological distress on the other may prove to haveactually been a time-linked gender difference. For example, asmen live in less traditional family forms, their mental healthmay be influenced as much by the quality of their family rolesas by the quality of their job role (Barnett, Marshall, & Pleck,1992; Pleck, 1985). Also, as mothers engage in full-time em-ployment, their mental health may be influenced as much bythe quality of their job role as by the quality of their family roles(Barnett & Marshall, 1991). This convergence in social roleoccupancy offers the interesting possibility of disaggregatinggender by holding constant one of its major sex-linked socialcomponents, namely, the social context of men's and women'sdaily lives. After controlling for social role occupancy, we exam-ine whether gender affects the relationship between job rolequality and distress.

In this article, we estimate the relationship between job rolequality and psychological distress for men and for women in asample of 300 dual-earner couples where both partners are em-ployed full-time and where, within each couple, parental status,number and ages of children, and household income are all thesame. Thus, after controlling for these important elements ofgender, we estimate the magnitude of the residual gender dif-ference in the relationship between social role stress and psycho-logical distress. However, even after controlling for theseaspects of men's and women's lives, it is not possible to talkabout true sex differences because so many other gender-linkedvariables, such as early socialization experiences, expectations,and sex role attitudes, cannot be controlled.

Because of major differences in men's and women's laborforce involvement, it has not been possible until recently tostudy this relationship. Most important, because of the sex-se-gregated nature of the workplace, men and women have heldvery different jobs. If all women worked in types of jobs neverheld by men it would be difficult to interpret a significant inter-action effect between job role quality and psychological dis-tress.

In the past it has been difficult to examine whether the linkbetween job quality and distress is similar for men and womenbecause not enough women were working at jobs that weresimilar to the jobs held by men. In addition, employed womenhave been more likely than men to work part-time, part of theyear. Finally, marital status has been more closely related towomen's employment status than to men's. Each of these differ-ences could reasonably contribute to the differential rates ofpsychological distress reported by men and women. The afore-mentioned problems are addressed in this sample because themen and women are all employed full-time, year around, theyare all married, and there is a substantial overlap in the types ofjobs and incomes of the men and women (see sample descrip-tion below). In addition, we collected information on occupa-tional prestige and household income and controlled for thesevariables in the analyses.

Literature Review

Although several studies have examined the job stress-men-tal health relationship, the applicability of their findings todual-earner couples is compromised by methodological limita-tions. For example, Cleary and Mechanic (1983) and Pearlin(1975) reported gender differences in the relationship betweenjob satisfaction and psychological distress; more specifically,they found that job satisfaction was more closely related todepression among men than among women. However, in bothstudies some of the working women were working part-time,some full-time, yet there was no assessment of the number ofhours worked, occupational prestige, or salary. In contrast, themen were all working full-time at probably higher prestige jobsand for better salaries, although these data are not provided.Moreover, the employment status of the men's wives was notstated; some unspecified fraction of the wives were employedand undoubtedly differed in terms of the number of hours theyworked, their salaries, and so forth. Although such differencesin the social context of men's and women's lives could account,at least in part, for the reported "sex differences" in the associa-tion between job satisfaction and distress, their effects havereceived scant attention.

Moreover, dual-earner couples, especially those in whichboth members work full-time, may differ from other couples inwhich one spouse is employed less regularly. Indeed, Gore andMangione (1983) found that when role occupancy was identi-cal, differences between men and women in distress were notseen (p. 310). In a sample of married men and women, allemployed at least 20 hr per week, Bielby and Bielby (1989)found that when women had work statuses and experiencessimilar to men's, they identified as strongly with and were ascommitted to the work role as were the men. Also, when menand women were both employed in occupations of similar pres-tige, the sex differential in distress was lowest (Golding, 1988).In a parallel fashion, most studies find that women reporthigher levels of physical health symptoms than do men. How-ever, this "sex" difference disappears (and even reverses) whenrole burdens are controlled (Verbrugge, 1986). These findingssuggest that the more similar men's and women's role patterns,the more similar their mental and physical health reports.Thus, what have often been mistaken for sex differences maybe time-bound gender differences having little to do with sexper se.

Past research on gender differences has suffered from diffi-culties in design and analysis. When studies compare employedmarried men with employed married women, most of whomare not married to each other (Cleary & Mechanic, 1983), fail-ure to control couple level variables such as length of marriage,stage of the life cycle, and household income may bias the analy-sis. The major issue is the issue of confounding. Suppose thathousehold income predicts distress. In a study not based oncouples, the household income of employed married men,some of whom are married to women who are not employed,would likely differ from the household income of employedmarried women. Unless household income was controlled inthe analysis, such a study would yield a biased estimate of thegender difference in the link between job quality and distress.A careful investigator would attempt to identify and control for

GENDER, JOB EXPERIENCES, AND DISTRESS 797

such household level covariates, but it would be difficult toargue that all such variables had been identified and well mea-sured even in the most careful study. An advantage of a matchedsample is that all such household variables—even those that areunanticipated—are effectively controlled and cannot bias theestimate of gender differences. Moreover, the matched sampleallows a more powerful test of gender differences in the linkbetween job quality and distress than does a nonmatched sam-ple, because in the matched sample analysis all household levelvariation is removed from the error term of the test.

Moreover, even when researchers have obtained matchedsamples, limitations in analytic approaches cast doubt on con-clusions regarding gender differences in the link between jobrole characteristics and outcomes. Some studies have used ana-lytic approaches that do not take full advantage of the matchedcharacter of the sample (Aneshensel, Frerichs, & Clark, 1981;Biernat & Wortman, 1991; Greenberger & O'Neil, 1990; Kessler& McRae, 1982; Rosenfeld, 1980). For example, correlated /tests and profile analysis as used by Biernat and Wortman(1991) appropriately capitalize on the within-couple depen-dence of the observations in testing for main effects of genderbut do not facilitate study of sex differences in the effect of jobrole characteristics on outcomes. When studying such GenderX Job Role interactions, researchers have computed separateregressions for men and women and then compared the regres-sion coefficients. This approach ignores the dependent charac-ter of the observations.

The following examples highlight some of the problems ofinterpretation resulting from the aforementioned limitations.First, household level predictors are confounded with individ-ual level predictors, resulting in a substantial danger of bias. Toillustrate, in couples, the economic status of one spouse is con-founded with the economic status of the other. Suppose that theman's income is the key determinant of household income. As-sume further that household income negatively predicts dis-tress (Belle, 1982). Suppose that household income is in fact thesole determinant of distress. Now suppose we take a sample ofmarried adults and estimate regressions separately for men andwomen. Then the association between men's income and dis-tress will be stronger than the association between women'sincome and distress, because men's income is the key determi-nant of household income, and household income is the soledeterminant of distress. This difference in regression coeffi-cients would not provide any evidence of a gender difference.Rather, the difference would be purely an artifact of the unspec-ified effect of household income. Without drawing samples ofdual-earner couples and including both individual level andhousehold level covariates, the accuracy and relevance of re-search findings are suspect. These problems are addressed inthis study by having a matched-pairs design and controlling forcovariates measured at the couple and individual levels.

Second, an analysis that fails to take into account the rela-tionships between men and women will be inefficient and,therefore, weak in statistical power in assessing gender differ-ences because all unexplained variation between households(couples) will appear in the error term of such an analysis. Thestronger the correlation between male and female partners onthe outcome, the more pronounced will be the loss of effi-ciency and power. In contrast, an appropriate analysis on the

basis of a matched-pairs design removes all of the variabilitybetween couples from the error term, increasing the statisticalpower associated with gender differences.

Finally, failure to incorporate effects of the marital relation-ship precludes assessing the relational character of individualdistress levels: The distress level of one spouse is likely affectedby the distress level of the other. Most adults function in dyads,and there are significant correlations between spouse pairs insuch stress-related variables as marital dissatisfaction, moneymatters, and worries about sexual problems (Haynes, Eaker, &Feinleib, 1983). Indeed, some researchers claim that as much as60% of the variance in individual marital satisfaction can beaccounted for by emotional linkage, that is, interactive pro-cesses whereby one spouse's emotions are influenced by theemotions of the other (Hatfield, Cacioppo, & Rapson, 1992;Levenson & Gottman, 1983; Sullins, 1991). In addition, withrespect to mental health states, men married to employedwomen compared with men married to homemakers were athigher risk for negative mental health outcomes (Kessler &McRae, 1982). Finally, with respect to stress-related physicalhealth outcomes, men married to higher educated, employedwives were at higher risk for coronary disease than their coun-terparts who were married to nonemployed women (Haynes etal., 1983). Such within-couple effects are lost when analyses areconducted separately on men and women.

The design of the present study addresses the aforemen-tioned limitations. First, the men and women are all employedfull-time, and there is substantial overlap in the types of jobsand incomes of the men and women in the sample; we collectedinformation on job type and salary, which are controlled in theanalyses. Second, we use a matched-pairs design, which in-creases precision of estimation and power in hypothesis testing.Third, our data analytic technique capitalizes on the matched-pairs design in studying both main effects and interactions.

Statistical Overview

Researchers studying gender differences commonly estimateseparate regression equations for men and women, with eachregression predicting an outcome such as psychological dis-tress. If the intercepts of the two models are different, the con-clusion is that a main effect of gender is present. If the regres-sion coefficients (the slopes) are different, the conclusion is thatthe predictor in question interacts with gender. This analyticapproach, however, is suboptimal when information is avail-able on the relationships that exist between pairs of subjects inthe sample.

An alternative approach would be a multivariate regressionin which the couple is viewed as the unit of analysis. Associatedwith each couple would be a pair of outcomes, one for thewoman and one for the man. Such an analysis would use theinformation about the positive association between male andfemale responses within couples to reduce the standard error oftests of gender differences in the magnitude of regression coeffi-cients. However, standard multivariate regression methods (cf.Anderson, 1984) require the same set of predictors for the fe-male and male outcomes. In our study, it was necessary to usethe woman's job role quality to predict her distress and to usethe man's job role quality to predict his distress. Under these

798 BARNETT, MARSHALL, RAUDENBUSH, AND BRENNAN

circumstances, ordinary least squares regression as imple-mented by standard multivariate approaches is inappropriate,and generalized least squares regression is required to achieveefficient estimation (Seber, 1977).

In multivariate regression based on ordinary least squares,the correlation between outcomes has no effect on estimates ofthe regression coefficients, but it does affect the standarderrors. In generalized least squares regression, the correlationinfluences both the coefficient estimates and the standarderrors. When every outcome is regressed on the same set ofpredictors, generalized least squares reduces to ordinary leastsquares. However, when the predictor set differs for differentoutcomes, generalized least squares is required for efficient es-timation. Reformulation of a standard multivariate model interms of a generalized least squares model is discussed in detailby Raudenbush, Becker, and Kalaian (1988). The analytic ap-proach selected here uses generalized least squares regressionto achieve efficient estimation.

Another advantage of the analytic approach adopted here isthat it takes into account the errors of measurement of thedependent variable, psychological distress. These measure-ment errors would otherwise have two undesirable effects oninference. First, the correlation between male and female dis-tress would be attenuated by measurement error and would,therefore, understate the relational character of distress withincouples. Second, the proportion of variability accounted for bythe model would be underestimated. Without partitioning vari-ability into components attributable to true score variance andmeasurement error variance, one might mistakenly concludethe model to have poor explanatory power when, in fact, asubstantial proportion of variation might be attributable toerrors of measurement that are not explainable. As a by-pro-duct, the analysis yielded estimates of the internal consistencyof the outcome separately for men and women.

In particular, we adapted the two-level hierarchical linearmodel as described by Raudenbush and Bryk (1986) to makefull use of the matched-pairs design. Although this methodol-ogy has commonly been used to analyze data collected on indi-viduals nested within groups, such as classrooms and schools,as well as to model individual change over time (Bryk & Rau-denbush, 1987), the methodology can also be adapted to incor-porate within-subject variability attributable to measurementerror (Goldstein, 1987; Raudenbush, Rowan, & Kang, 1991).

To apply the hierarchical analytic approach, a Level 1 modelrepresents each observed psychological distress score as a func-tion of a true score plus a measurement error. At Level 2, eachpair of true scores associated with a couple is viewed as varyingrandomly over the population of couples.

Level 1 Model

Psychological distress was measured by 24 items, with eachitem measured on a Likert scale (see Measures section). Fromthese 24 items we constructed two parallel scales. First, wematched items on their sample standard deviations. Then,within each pair of items, we randomly assigned each item toone of two parallel scales. This procedure guaranteed that thescales would be truly parallel, each having similar standarddeviations and reliability. Thus, for each couple there were four

outcome scores, two for the man and two for the woman. Themeasurement model postulates simply that each outcome scorefor each member of a couple is the sum of a true score plus ameasurement error, where the measurement errors are un-correlated with constant variance. The model may be written

where Yv is the subscale score / for couple j on the psychologicaldistress outcome, with / = 1 , . . . , 4 subscale scores per coupleand j = 1, . . . , 300 couples. Thus, there are four subscale scoresper couple, two for the woman and two for the man. By conven-tion, the first two subscale scores, Yv and Y2j are those asso-ciated with the female partner in the couple and the secondtwo, Y3J and Y4j are those associated with the male partner. Thepredictor Xuj is an indicator for women equal to 1 if Ytj is mea-sured on a woman and 0 if Yv is measured on a man; X2ij is anindicator for men equal to 1 if Yo is measured on a man and 0 ifYy is measured on a woman. With these definitions in mind, itfollows that fitj is the true psychological distress score for thewoman in couple j , and fi2j is the true score for the man, andthat ev and e2j are the errors of measurement of the female truescore, whereas e3J and e4J are errors of measurement of the maletrue score. We assume these measurement errors to be indepen-dent, normally distributed with constant variance, a2? In es-sence, the Level 1 model is a classical measurement model inwhich each of two observed scores per subject is equated to thesum of a true score plus error.

Level 2 Model

The Level 2 model specifies the female and male true scoresas outcomes to be predicted by a set of explanatory variablesbelieved possibly related to psychological distress. The modelmay be written

and

l\i ~ y io

720

(2)

(3)

where 710 and y20 are the intercepts for women and men, respec-tively, WXgj is the gth predictor for female distress, W2qj is the <jthpredictor for male distress, and the residuals uy and u2J are as-sumed bivariate normally distributed over couples with vari-ances TU and T22, respectively, and covariance T12. The covari-ance captures the dependence of pairs of residuals computedfrom the same subject.

We formulate a model in which female job role quality pre-dicts female distress and male job role quality predicts maledistress. The model allows control for covariates measured atthe indiviudal and couple levels. For example, household in-

3 The random assignment of items to the two parallel subscales en-sures the validity of the assumption that they have equal variance, andan exploratory analysis confirmed the efficacy of randomization inthis regard. The assumption that men and women share the same mea-surement error variance was supported by an analysis revealing thatseparate estimates of this variance for women and men were nearlyidentical.

GENDER, JOB EXPERIENCES, AND DISTRESS 799

come can predict female distress, and household income alsocan predict male distress.

The model allows testing of the significance of differencesbetween female and male intercepts and between female andmale regression coefficients. In the terms of Equation 2, we cantest the hypothesis

Ho: yig = 72,- (4)

When the two coefficients are found not to be significantlydifferent for men and women, these coefficients can be con-strained to be equal. The test of the significance of such a con-strained coefficient is more powerful than are either of theseparate tests for the gender-specific coefficients.

Of particular interest is the Level 2 model in which no pre-dictors are specified:

and

•• = 7 i o + " u

V = 720 + Uy

(5)

(6)

In this model, the only regression parameters are the male in-tercept and the female intercept, actually definable as the maleand female true score means. This simple model enables (a) atest of the overall gender difference in distress (equivalent inessence to a matched-pairs t test), (b) estimation of the male andfemale true score variance in distress, (c) estimation of the reli-ability (internal consistency) of the measurement of distress forboth women and men and (d) estimation of the correlationbetween male and female true scores. A detailed technical dis-cussion of this approach appears in Raudenbush et al. (1991).The estimation of male and female true score variance serves asa baseline for evaluating the explanatory power of subsequentmodels in that it defines the magnitude of potentially explain-able variance in the outcomes. Later models can be assessed bycalculating the proportion of this baseline variance (technicallythe unconditional variance) accounted for by the model.

Using these techniques, we test the following two hypothe-ses:

1. Job role quality will be associated with psychological dis-tress, net of the effect of marital role quality and parental sta-tus.

2. The association between job role quality and distresswill not be different for women and men.

More specifically, we first specify a model in which eachpredictor is allowed to have a different effect for women andmen. We then test the difference between female and maleregression coefficients for each predictor. If no gender differ-ence in a pair of regression coefficients is found, that coeffi-cient is constrained to be equal in the next analysis. In thiscontext, we distinguish between two types of control variables;the first are variables that have a single value for each couple(household income, number of years together, and parental sta-tus), and the second are variables for which each member of thecouple has a unique value (occupational prestige, educationallevel, age, and marital role quality). To minimize the chance ofType I error attendant on undertaking separate tests for signifi-cant differences in each of the pairs of predictors, we conductedomnibus tests of the hypotheses that for each group of control

variables there is no difference in the magnitude of the regres-sion coefficients for men and women. Given the retention ofeach of these omnibus hypotheses, we reestimated the modelwith the coefficients for the control variables constrained to beequal for men and women. The test for gender difference in jobrole quality was undertaken individually after all other slopeshad been constrained to be the same. This approach maxi-mizes the power in testing for gender differences in the job rolequality effect because the standard error of that test is mini-mized when the other covariates are constrained to have thesame effect for women and men. Hence, the two-stage hypothe-sis-testing strategy maximizes the opportunity to find a genderdifference in the relation between job role quality and distress,if such a gender difference exists.

Method

Sample

The data for these analyses come from the first wave of a three-wavedata collection (over 2 years) of a random sample of 300 dual-earnercouples residing in eastern Massachusetts, in which both spouses wereemployed full-time and the men were between the ages of 25-40. Thesample was stratified on parental status; at the first wave, 60% of thesample were parents, 40% were not. The sample was drawn from thetown lists of all residents of two Boston area towns. These towns wereselected because they were socioeconomically diverse and included alarge proportion of working women. (In one town, 70.1% of womenages 20-54 were employed in 1980, according to the U.S. Census: 1980.In the other, 75.2% of women ages 20-40 were employed in 1980,according to the U.S. Census: 1980.) The participation rate among theeligible couples who we were able to contact was 68% (see Barnett,Marshall, & Pleck, 1992, for a complete description of the samplingprocedures).

The population of these towns was overwhelmingly White, as wasthe sample we obtained. Thus, we were unable to examine race differ-ences. The actual racial composition of the sample was 97% White, 1%Hispanic, 1% Black, and 1% Native American and other. To have ob-tained an analyzable sample of Black or Hispanic couples would haverequired a sampling design beyond the scope of the project.

On average the men in the sample were 34.99 years (SD = 4.29),whereas the women were 34.21 (SD = 4.83). The men and women, onaverage, had completed 16 years of schooling, that is, they completed acollege degree (M = 16.40, SD = 2.34 and M= 16.20, SD = 2.09, formen and women, respectively). However, there was a wide range ofeducational attainment: Among the men, 27% had not completed 4years of college, whereas 40% had some graduate education; among thewomen, 26% had not completed 4 years of college compared with 36%who had some postcollege education. More than 90% of the samplewere in first marriages; 9.3% of the women and 7.2% of the men hadbeen married previously. Among the couples who had children, theaverage family size was 1.76 children (SD = 1.04).

The majority of both men and women were employed in managerialor professional occupations (67.7% of the men and 71.1% of thewomen). Approximately one quarter of the women (24.9%) comparedwith less than 20% of the men (17.7%) were employed in technical,sales, or administrative occupations. Finally, more men than women(14.6% compared with 4.0%) were employed in either service, precisionproduction, crafts, or repair occupations or as operators or laborers.

Procedure

Subjects were interviewed in their homes or offices by trained inter-viewers. The interviews were conducted between the fall of 1989 and

800 BARNETT, MARSHALL, RAUDENBUSH, AND BRENNAN

the spring of 1990. They took about 1.5 hrand covered many aspects ofthe men's and women's lives, including the rewards and concerns intheir job and marital roles, as well as measures of psychological dis-tress. Before the interview the subjects received a packet of forms to befilled out and returned to the interviewer. Each couple received $25 forparticipating.

Measures

Psychological distress. Psychological distress was measured by twoparallel scales, which are essentially frequency-of-symptoms mea-sures. Subjects indicated on a 5-point scale (from 0 = not at all to 4 =extremely) how often in the past week they were bothered by each of 10symptoms of anxiety and 14 symptoms of depression (Derogatis, 1975).The anxiety and depression items were first combined into one 24-itemmeasure of psychological distress on the basis of (a) a coefficient alphaof ,90 for the combined items and (b) a correlation of .80 between theanxiety and depression scales. This measure was then divided intoparallel scales by matching items on their sample standard deviations.From each pair, the items were randomly assigned to each of the twoscales, yielding two parallel scales with similar distributions. Thecharacteristic skewness of the distributions was addressed throughnatural log transformation of each of the scales. In the interest of inter-pretability, the scales were adjusted to have a mean of 50 and a stan-dard deviation of 10. The scales are well correlated with each other (r =.834). The strength of this relationship is consistent with those re-ported for similar scales (Derogatis, 1983) and substantially greaterthan the correlation between independent measures and depressionand anxiety.

Role quality. On the basis of previous research (Baruch & Barnett,1983; Barnett, Marshall, & Pleck, 1992) and pilot studies, we identi-fied the rewarding and distressing aspects of men's and women's joband marital roles. On the basis of response frequency, we identified 33job reward and 28 job concern items, as well as 26 marital reward and26 marital concern items.

Subjects used a 4-point scale (from 1 = not at all to 4 = extremely) toindicate to what extent, if at all, each of the items was currently reward-ing or of concern. For example, with respect to the job role, subjectswere asked how rewarding was "having a variety of tasks" and howmuch of a concern was "lack of job security." For the marital role,subjects were asked how rewarding was "enjoying the same activities"and how much of a concern was "your partner being critical of you."For each role, reward items were weighted by 1, the concern items by- 1 , and the weighted mean constituted the scale score. (The itemsincluded in the reward and concern scales for the job and marital rolesare reproduced in the Appendix.)

To establish the reliabilities of these scores, Cronbach alphas andtest-retest reliability coefficients were computed. Cronbach alphaswere computed separately for the men and women: For the job rewardscale, alphas were .89 for men and .90 for women; for the job concernscale, the score was .87 for men and women; for marital rewards, .93 formen and women; and for marital concerns, .89 and .90, respectively.Test-retest reliability coefficients were calculated for approximately10% of the sample (n = 64; 32 men and 32 women) who were reinter-viewed within 3 months of their initial interview. For the rewards in thejob role, the test-retest reliability coefficients were .68 and .84, for themen and women, respectively; for the concerns in the job role, theywere ,684 and .77, respectively; for marital rewards, they were .91 and.84, respectively; and for marital concerns, they were .95 and .81, re-spectively.

Occupational prestige. We used the Bose Index (Bose, 1985) to codeoccupational prestige. This index uses the 1980 Census Three-DigitOccupation Code and assigns prestige values separately for eachgender.

Education This individual level variable was the number of yearsof education that each respondent completed.

Household income. This couple level variable reflects the absolutesalaries of both partners plus their unearned income, if any, divided bythe number of children plus two. Because the distribution of this vari-able was highly skewed, we used the natural log of household income,which we refer to as log per capita income.

Years together. This couple level variable was the absolute numberof years the couple had been married or partnered (i.e., living together).(Only nine couples [3% of the sample] were partnered, 97% weremarried. For simplicity's sake, we, therefore, refer to marital role qual-ity in describing the quality of men's and women's relationships withtheir partners.)

Parental status. A dummy variable was created for parental status(0 = no children and 1 = one or more children).

Results

Descriptive Characteristics of the Sample

Table 1 presents the means and standard deviations of therelevant individual level and couple level variables. As can beseen, on average, women experienced higher levels of rewardsin their jobs than did men (M= 3.01, SD = 0.42 and M= 2.88,SD = 0.39, for women and men, respectively). That is, womenindicated that their jobs were, on average, considerably reward-ing, whereas men indicated that their jobs were between some-what and considerably rewarding. With respect to job con-cerns, both men and women reported that, on average, theirjobs were between somewhat and considerably a concern (M=1.69, SD = 0.37, and M= 1.71, SD= 0.39, for women and men,respectively). Thus, women's job role quality scores were higherthan men's (M = 1.32, SD = 0.63 and M = 1.14, SD = 0.63 forwomen and men, respectively). Results of a paired t test indi-cated that this mean difference was significant, /(300) = 3.90,

Results of the Analysis Hierarchical Model

We present the results of the hierarchical analysis in threesections. First, we report the results of the baseline model. Sec-ond, we report the results of the explanatory model based ongender-specific regression coefficients. Third, we consider theresults of the constrained model, that is, results after gender-specific coefficients found nonsignificantly different were con-strained to be equal.

Baseline model (Table 2, Model 1). As mentioned earlier,the baseline model predicts female and male true scores ondistress with just separate female and male intercepts. No ex-planatory variables are included. Paralleling the descriptiveanalysis, the mean distress of women (M = 51.0) was found tobe significantly higher than that of men (M = 49.0), ;(299) =2.76, p = .006. Note that the estimated true score variances forwomen and men were found to be highly similar (84.4 for

4 A possible explanation for the relatively low test-retest reliabilitycoefficients for the job scores is that the reinterviews took place amidstgeneral concern about the Massachusetts economy and massivelayoffs in companies and industries in which many of the subjects,their spouses, or their friends were employed.

GENDER, JOB EXPERIENCES, AND DISTRESS 801

Table 1Means and Standard Deviations for Individual Level and Couple Level Variables

Variable

Individual levelAgeEducation"Occupational prestige1"Job role quality

Job rewardsJob concerns

Marital role qualityCouple level

Log per capita incomeNo. of years togetherParental status

Women

M

34.2116.2059.95

1.323.011.691.69

SD

4.832.09

14.710.630.420.370.78

Men

M

34.9916.4062.26

1.142.881.711.67

SD

4.292.34

16.830.620.390.390.68

Couple

M

10.218.210.60

SD

0.505.210.49

' Years of schooling completed. b Bose Index (Bose, 1985).

women and 80.8 for men). The estimated correlation betweentrue scores of women and men in the same couple was found tobe .22. Reliabilities of estimation were found to be quite high,.91 both for women and for men.

The variance component estimates are shown separately forwomen and men for each model in Table 3. The proportion ofthese variances explained by subsequent explanatory modelscan be interpreted in the same way as the squared multiplecorrelation except that in our analyses, the objective was toaccount for true score rather than observed score variance, thatis variance in fiv and f}y rather than variance in the observed Yvalues (see Equations 1 and 2).

Explanatory model with gender-specific regression coeffi-cients (Table 2, Model 2). In the next analysis, all explanatoryvariables (i.e., age, education, occupational prestige, householdincome, number of years together, parental status, marital rolequality, and job role quality) were included as predictors of thefemale and male true scores on distress. For every predictor,separate coefficients were estimated for women and for men.An examination of the results in Table 2 (Model 2) reveals thatthe magnitude of the coefficients for women and for menlooked fairly similar (compared with the size of the estimatedstandard errors). To test for gender differences in the magni-tude of these coefficients, we formulated an omnibus test of thehypothesis that, for each group of control variables, female andmale regression coefficients are identical.

Under this hypothesis, the estimated coefficients, listed inTable 2 for males and females, for the control variables that hada single value for each couple (log per capita income, number ofyears together, and parental status) differed only as a result ofchance. For log per capita income, the male and female coeffi-cients looked different (0.498 vs. 1.798), but the difference wassmall relative to the standard errors. Similarly, for years to-gether, the coefficients looked a bit different (-0.111 vs. 0.056),but the difference was small relative to the standard errors.Bryk, Raudenbush, Seltzer, and Congdon (1988) provided amultivariate test of the hypothesis that each pair of coefficientsis identical. When the null hypothesis is true, the test statistic isdistributed as a chi-square with three degrees of freedom. In

our case, the observed chi-square statistic was 1.47 (df= 3, ns),indicating that the null hypothesis cannot be rejected and thatthe coefficients for males and females are best viewed as identi-cal for these control variables.

The second group of control variables contained those thathad individual values for each member of a couple (age, occupa-tional prestige, educational level, and marital role quality). Thetest for gender differences in these coefficients yielded a chi-square value of 0.51 (df= 4, ns). Hence, the data revealed no hintof evidence that the effects of any of the control variables dif-fered for men and women. The model was, therefore, reesti-mated with female and male coefficients constrained to be thesame constant for every control variable. The job role qualitycoefficients, however, were allowed to differ for men andwomen. Constraining the effects of the control variables to beidentical for men and women maximizes the power of the testfor gender differences in the effect of job role quality.

Explanatory model: Coefficients constrained (Table 2, Models3 and 4). After constraining the control coefficients to be thesame for women and men, the model was reestimated (Model3), and the difference between female and male coefficients forjob role quality reestimated. The test of the difference betweenthe magnitude of the male and female coefficients produced achi-square of 0.47 (df= 1, ns). Hence, we must conclude that thedata provide no evidence to support the proposition that theeffect of job role quality on psychological distress is differentfor women and for men.

On the basis of this result, the model was reestimated with alleffects, including the effect of job role quality, constrained to bethe same for men and for women (Table 2, Model 4). Theseresults give tests of significance for the explanatory variablesthat are more powerful than would be the case if the nonsignifi-cantly different female and male regression coefficients hadremained in the model.

In summary, we found no evidence that any of the predictorvariables had different effects for men than for women, whichis why the magnitudes of the coefficients listed in Table 2(Model 4) are identical for men and women. More specifically,

802 BARNETT, MARSHALL, RAUDENBUSH, AND BRENNAN

Table 2Regression Coefficient Estimation

Predictor

FemalesInterceptJob role qualityMarital role qualityEducation8

Occupational prestigeAgeParental statusPer capita incomeb

Years togetherMales

InterceptJob role qualityMarital role qualityEducationOccupational prestigeAgeParental statusPer capita income5

Years together

Model 1

Coefficient

50.958

49.042

baseline

SE

0.556

0.545

Model 2 unconstrained

Coefficient

51.380-3.528-4.778

0.00030.0240.008

-2.0431.7980.056

48.499-4.332-4.625-0.519

0.1030.119

-1.7960.498

-0.111

SE

0.4890.8070.6570.2910.0400.1281.3781.2990.124

0.4830.7920.7300.2950.0390.1461.3401.3360.124

Model 3 control effectsconstrained

Coefficient

51.425-3.621-4.672a

-0.246"0.061"O.O58a

-1.997a

1.142"-0.024"

48.541-4.362-4.672"-0.246"

0.061"0.058"

-1.997"1.142"

-0.024"

SE

0.4860.7820.4890.2060.0280.0970.9710.9420.088

0.4790.7730.4890.2060.0280.0970.9710.9420.088

Model 4 all effectsconstrained

Coefficient

51.459-3.996"-4.661"-0.257a

0.061"0.059"

-1.957"1.156"

-0.025 '

48.574-3.996"-4.661"-0.257"

0.061"0.059"

-1.957"1.156a

-0.025"

SE

0.4830.5600.4880.2060.0280.0970.9690.9420.088

0.4770.5600.4880.2060.0280.0970.9690.9420.088

* Value of coefficients has been constrained to be identical for males and females. b Natural log.

we found that, for both women and men, job role quality wassignificantly negatively related to psychological distress (b =-4.00), t(59O) = -7.14, p < .001. Regarding the control vari-ables, which themselves are of some theoretical interest, mari-tal role quality negatively predicted distress (b = —4.66), £(590)= -9.55, p < .001. Parental status also had a negative effect ondistress (b=-\.96), £(590) = -2.02, p < .05, such that coupleswho did not have children reported more distress than coupleswho had children. The respondent's occupational prestige waspositively related to distress (b = 0.061), £(590) = 2.19, p < .05.The remaining control variables—age, years of education, percapita income, and years together—had no significant effectsin predicting psychological distress. A significant difference inthe level of psychological distress for men and women re-mained, £(291) = 4.28, p < .001, even after the aforementionedfactors were accounted for in the model.

Discussion and Conclusion

The main findings of this study of 300 eastern Massachu-setts, White, predominantly middle-class, dual-earner couples,in which the man was between 25-40 years of age, were thatsubjective experiences on the job are significantly associatedwith psychological distress for both men and women and thatthe magnitude of this effect depends little, if at all, on gender. Inthis sample of full-time employed, married men and women,we found support for the previously reported association be-tween experiences on the job and distress. The absence ofgender differences, however, flies in the face of firmly heldcultural beliefs that "the work place and its events . . . moreclosely regulate the psychological fate of men than of women"(Pearlin, 1975, p. 202). Positive and negative job experienceshave similar influences on men's and women's mental health.

Table 3Variance Component Estimation

Variance component

Female varianceMale variance

Proportion of variance explainedFemalesMales

Model 1baseline

84.4480.82

Coefficients for

Model 2unconstrained

60.9558.97

.28

.27

Model 3controleffects

constrained

60.5858.65

.28

.27

Model 4all effects

constrained

60.5358.58

.28

.28

GENDER, JOB EXPERIENCES, AND DISTRESS 803

These findings suggest that the previously assumed sex differ-ence in the relationship between job role stress and psychologi-cal distress may be an artifact of the segregated worlds in whichmen and women operated. When, as in this study, it is possibleto hold constant the social role patterns of men and women (acrucial aspect of gender), then there is no significant residualeffect of the remaining (in this case unspecified) gender con-struct. These findings suggest that as the role patterns of menand women become increasingly similar, we need to rethinkour ideas about sex versus gender differences. Indeed, many ofour beliefs about sex differences may turn out to be time-boundgender differences, more a reflection of the different social rolecontexts in which men and women operated in the past than ofinherent "sex" differences.

The familiar caution against affirming a null hypothesis ap-plies here. Failure to reject the null hypothesis of no genderdifference in the effect of job role quality does not imply thatthese effects are identical for men and women. Interpretationdepends on the power of the test. Our point estimates of theregression coefficients specifying the relationship between jobrole quality and distress were equivalent to standardized coeffi-cients of-0.30 for men and -0.25 for women (based on Model3 of Table 2). The combined estimate, based on constrainingthe coefficients to be identical, was —0.28. Now suppose thatthe two true standardized coefficients were quite different, say—0.40 for men and -0.10 for women. These would correspondto raw coefficients of—5.77 and —1.44. Given the standarderror of 1.08 observed for the difference between male andfemale job role quality coefficients, it can readily be seen thatthe study would have power greater than 0.99 to detect such atrue difference. If the difference were more modest, say —0.35for men and —0.15 for women, the power would still be sub-stantial at 0.82. A reasonable conclusion is that the two effectsare quite similar in magnitude; clearly, both effects are highlysignificantly negative.

It is possible that past results confirming the presence ofsubstantial gender differences may also have been promul-gated, however unwittingly, by methodological strategies thatinaccurately labeled as gender differences differences that wereactually attributable to unspecified couple level variables. Themajor problem is that most methodologies analyze male andfemale samples separately by using samples of persons who arenot partners or by ignoring the partnerships that exist. Ourfindings challenge that assumption. Using methodologies thattake advantage of the matched-pairs design of the sample andthat control for measured couple level variables and such otherconfounding variables as marital role quality and parental sta-tus, we found no evidence of gender difference in the relation-ship between job experiences and psychological distress. Wecaution that our results generalize to dual-earner families of thetype studied here and do not facilitate comparisons betweenmen and women drawn from single-earner households orhouseholds with single adults living alone.

Women in the sample reported higher levels of job role qual-ity than did their partners. This finding is consistent with otherreports indicating that although women's working conditionsare typically less favorable than men's, women experience theirjobs just as positively or even more positively than do men(Cleary & Mechanic, 1983; Crosby, 1984). This counterintuitive

finding is consistent with predictions from the relative depriva-tion literature, which posits that subjective experiences aregauged against one's internal expectations and not against ex-ternal criteria (Crosby, 1984). Women whose jobs are not as"good" as their husbands' might be expected to feel more satis-fied because their jobs are in line with or may even exceed theirexpectations. Men, in contrast, might feel less positive becausetheir jobs may not measure up to their expectations.

Even after controlling for individual and couple level vari-ables, women reported significantly higher levels of psychologi-cal distress than did men. This gender difference has beenreported in many studies. Researchers attempting to accountfor this difference typically focus on the predictors and modera-tors, speculating that compared with men, women either areexposed to more stressors (or are more vulnerable to stressors)or do not have as adequate strategies for coping with stressors.

In contrast, an intriguing recent explanation focuses on theoutcome variable. This explanation grows out of the study ofgender differences in affect intensity levels. Fujita, Diener, andSandvik (1991) suggested that men and women have similarmean levels of affect intensity, however, women's affective rangeis considerably broader. In other words, compared with men,women report higher levels of both negative and positive affectintensity. Presumably, women would also have reported higherlevels of positive affect had we included such measures in ouranalysis. Thus, rather than connoting a higher risk for mentalhealth problems, women's higher scores on measures of symp-toms of anxiety and depression might merely reflect one side ofthe affect intensity coin.

Finally, it is important to note that this sample of relativelywell-educated and full-time employed dual-earner couplesfrom eastern Massachusetts may not be representative of thelarger national population of dual-earner couples. Future re-search is needed to address these relationships among a morediverse sample.

References

Anderson, T. W (1984). An introduction to multivariate statistical analy-sis (2nd ed.). New York: Wiley.

Aneshensel, C. S., Frerichs, R. R, & Clark, V A. (1981). Family rolesand sex differences in depression. Journal of Health and Social Be-havior, 22, 379-393.

Barnett, R. C. (1993). Multiple roles, gender, and psychological dis-tress. In L. Goldberger & S. Bresnitz (Eds.), The handbook of stress(2nd ed., pp. 427-445). New York: Free Press.

Barnett, R. C, & Baruch, G. K. (1987). Determinants of fathers' partici-pation in family work. Journal of Marriage and the Family, 49, 29-40.

Barnett, R. C, & Marshall, N. L. (1991). The relationship betweenwomen's work and family roles and subjective well-being and psycho-logical distress. In M. Frankenhaeuser, U. Lundberg, & M. Chesney(Eds.), Women, work and health: Stress and opportunities (pp. 111-136). New York: Plenum Press.

Barnett, R. C, Marshall, N. L, & Pleck, J. H. (1992). Men's multipleroles and their relationship to men's psychological distress. Journalof Marriage and the Family, 54, 358-367.

Barnett, R. C, Marshall, N. L., & Singer, J. D. (1992). Job experiencesover time, multiple roles, and women's mental health: A longitu-dinal study. Journal of Personality and Social Psychology, 62, 634-644.

804 BARNETT, MARSHALL, RAUDENBUSH, AND BRENNAN

Baruch, G. K., & Barnett, R. C. (1983). Adult daughters' relationshipswith their mothers. Journal of Marriage and the Family, 45, 601-606.

Baruch, G. K., & Barnett, R. C. (1986). Role quality, multiple roleinvolvement, and psychological well-being in midlife women. Jour-nal of Personality and Social Psychology; 51, 578-581.

Belle, D. (1982). Lives in stress: Women and depression. Beverly Hills,CA: Sage.

Bern, S. L. (1993). The lenses of gender: Transforming the debate onsexual inequity. New Haven, CT: Yale University Press.

Bielby, W T, & Bielby, D. D. (1989). Family ties: Balancing commit-ments to work and family in dual earner households. American So-ciological Review, 54, 776-789.

Biernat, M., & Wortman, C. (1991). Sharing of home responsibilitiesbetween professionally employed women and their husbands. Jour-nal of Personality and Social Psychology, 60, 844-860.

Bose, C. E. (1985). Jobs and gender: A study of occupational prestige.New York: Praeger.

Bryk, A. S., & Raudenbush, S. W (1987). Application of hierarchicallinear models to assessing change. Psychological Bulletin, 101,147-158.

Bryk, A. S., Raudenbush, S., Seltzer, M., & Congdon, R. (1988). Anintroduction to HLM: Computer program and user's guide. Unpub-lished manual, University of Chicago, Department of Education.

Cleary, P. D, & Mechanic, D. (1983). Sex differences in psychologicaldistress among married people. Journal of Health and Social Behav-ior, 24, 111-121.

Crosby, F. J. (1984). Job satisfaction and domestic life. In M. D. Lee &R. N. Kanungo (Eds.), Management of work and personal life (pp.41-60). New York: Praeger.

Crosby, F. J. (1991). Juggling. New York: Free Press.Deaux, K. (1984). From individual differences to social categories.

American Psychologist, 39, 105-116.Derogatis, L. R. (1975). The SCL-90-R. Baltimore, MD: Clinical Psy-

chometrics.Derogatis, L. R. (1983). Description and bibliography for the SCL-90-R

& other instruments of the psych opathology rating scale series. Balti-more, MD: Johns Hopkins University School of Medicine.

Faludi, S. (1991). Backlash: The undeclared war against Americanwomen. New York: Crown.

Frankenhaeuser, M., Lundberg, U, & Chesney, M. (1991). Women,work and health: Stress and opportunities. New \brk: Plenum Press.

Fujita, F., Diener, E., & Sandvik, E. (1991). Gender differences in nega-tive affect and well-being: The case for emotional intensity. Journalof Personality and Social Psychology, 61, 427-434.

Golding, J. M. (1988). Gender differences in depressive symptoms:Statistical considerations. Psychology of Women Quarterly, 12, 61-74.

Goldstein, H. (1987). Multilevel models in educational and social re-search. Oxford, England: Oxford University Press.

Gore, S., & Mangione, T. W (1983). Social roles, sex rolesand psycholog-ical distress: Additive and interactive models of sex differences.Journal of Health and Social Behavior, 24, 300-312.

Greenberger, E., & O'Neil, R. (1990). Parents' concerns about theirchild's development: Implications for fathers' and mothers' well-be-ing and attitudes toward work. Journal of Marriage and the Family,52, 621-635.

Gurin, G., Veroff, J., & Feld, S. (1960). Americans view their mentalhealth. New >t>rk: Basic Books.

Hatfield, E., Cacioppo, J. T., & Rapson, R. (1992). The logic of emo-tion: Emotional contagion. In M. S. Clark (Ed.), Review of personal-ity and social psychology, Volume 14: Emotion and social behavior(pp. 151-177). Newbury Park, CA: Sage.

Haynes, S. G, Eaker, E. D., & Feinleib, M. (1983). Spouse behavior and

coronary heart disease in men: Prospective results from the Fram-ingham heart study. American Journal of Epidemiology, 110, 1-22.

Hochschild, A., & Machung, A. (1989). The second shift: Working par-ents and the revolution at home. New York: Viking Press.

House, J. S., Strecher, V, Metzner, H. L., & Robbins, C. A. (1986).Occupational stress and health among men and women in the Te-cumseh community health study. Journal of Health and Social Behav-ior, 27, 62-77.

Karasek, R. A., Schwartz, J., Theorell, T., Pieper, C, Russell, S., &Michela, J. (1982). Final report: Job characteristics, occupation andcoronary heart disease. New York: Columbia University, Departmentof Industrial Engineering and Operations Research.

Karasek, R. A., & Theorell, T. (1990). Healthy work: Stress, productivityand the reconstruction of working life. New York: Basic Books.

Kessler, R. C, & McRae, J. A., Jr. (1982). The effect of wives' employ-ment on the mental health of married men and women. AmericanSociological Review, 47, 217-227.

LaCroix, A. Z., & Haynes, S. G. (1987). Gender differences in thehealth effects of workplace roles. In R. C. Barnett, L. Biener, & G. K.Baruch (Eds.), Gender and stress (pp. 96-121). New York: Free Press.

Levenson, R. W, & Gottman, J. M. (1983). Marital interaction: Physio-logical linkage and affective exchange. Journal of Personality andSocial Psychology, 49, 85-94.

Ortner, S. B., & Whitehead, H. (1981). Sexual meanings: The culturalconstruction of gender and sexuality. Cambridge, MA: CambridgeUniversity Press.

Parsons, T. (1949). The social structure of the family. In R. Anshen(Ed.), The family: Its function and destiny (pp. 173-201). New York:Harper.

Parsons, T, & Bales, R. (Eds.). (1955). Family, socialization, and interac-tion process. New \brk: Free Press.

Pearlin, L. I. (1975). Sex roles and depression. In N. Datan & L. H.Ginsberg (Eds.), Life-span developmental psychology (pp. 191-207).San Diego, CA: Academic Press.

Pleck, J. H. (1985). Working wives, working husbands. Beverly Hills, CA:Sage.

Raudenbush, S. W, Becker, B. J., & Kalaian, S. (1988). Modeling multi-variate effect sizes. Psychological Bulletin, 103, 111-120.

Raudenbush, S., & Bryk, A. S. (1986). A hierarchical model for study-ing school effects. Sociology of Education, 59, 1-17.

Raudenbush, S. W, Rowan, B., & Kang, S. J. (1991). A multilevel, multi-variate statistical model for studying school climate in secondaryschools with estimation via the EM algorithm. Journal of Educa-tional Statistics, 16, 295-330.

Repetti, R. L., Matthews, K. A., & Waldron, I. (1989). Employmentand women's health: Effects of paid employment on women's mentaland physical health. American Psychologist, 44,1394-1401.

Rodin, J., & Ickovics, J. R. (1990). Women's health: Review and re-search agenda as we approach the 21st century. American Psycholo-gist, « , 1018-1034.

Rosenfeld, S. (1980). Sex differences in depression: Do women alwayshave higher rates? Journal of Health and Social Behavior, 21, 33-42.

Ross, C. E., Mirowsky, J., & Huber, J. (1983). Dividing work, sharingwork, and in-between: Marriage patterns and depression. AmericanSociological Review, 48, 809-823.

Seber, G. A. F. (1977). Linear regression analysis. New York: Wiley.Simon, R. (1992). Parental role strains, salience of parental identity

and gender differences in psychological distress. Journal of Healthand Social Behavior, 33, 25-35.

Skolnick, A. (1991). Embattled paradise. New \brk: Basic Books.Spence, J. X, Deaux, K., & Helmreich, R. L. (1985). Sex roles in contem-

porary American society. In G. Lindsey & E. Aronson (Eds.), The

GENDER, JOB EXPERIENCES, AND DISTRESS 805

handbook of social psychology (Vol. 2, pp. 149-178). New York: Ran-dom House.

Sullins, E. S. (1991). Emotional contagion revisited: Effects of socialcomparison and expression style on mood convergence. Personalityand Social Psychology Bulletin, 17,166-174.

Thoits, P. A. (1992). Identity structures and psychological well-being:Gender and marital status comparisons. Unpublished manuscript,Sociology Department, Vanderbilt University, Nashville.

Verbrugge, L. M. (1986, March). Sex differences in physical health-making good sense of empirical results. Paper presented at the meet-ings of the Society of Behavioral Medicine, San Francisco, CA.

Veroff, J., Douvan, E., & Kulka, R. A. (1981). The inner American. NewYork: Basic Books.

Wethington, E., & Kessler, R. C. (1989). Employment, parental respon-sibility, and psychological distress: A longitudinal study of marriedwomen. Journal of Family Issues, 10, 527-546.

Appendix

Role Quality Measures

Job Role Quality

Job Rewards

1. Challenging or stimulating work2. Liking your co-workers3. The income4. Being able to work on your own5. The job security6. The recognition you get7. Doing work you consider significant8. Your supervisor's respect for your abilities9. Being needed by others

10. Being able to set your own work schedule11. Your work contributing to the good of a larger community12. Having a variety of tasks13. Having hours that fit your needs14. Being able to work as part of a team or group15. Your job being flexible enough that you can respond to non-

work situations16. Liking your supervisor17. Making good money compared with other people in your

field18. Being able to make decisions on your own19. Your supervisor's concern about the welfare of those under

him or her20. The sense of accomplishment and competence you get from

doing your job21. Having the authority you need to get your job done22. Having friendly co-workers23. The job's fitting your skills24. The appreciation you get25. The opportunities for advancement26. The freedom to decide how you do your work27. Your supervisor paying attention to what you have to say28. Having an impact on other people's lives29. The opportunity for learning new things30. Helping others31. Having supportive co-workers32. The benefits your job offers, for example, paid sick leave

Job Concerns

1. The job's dullness, monotony, lack of variety2. Having little chance for the advancement you want or de-

serve3. The job's not using your skills4. Your supervisor's lack of competence5. Lack of respect at your workplace for someone who does

your job

6. Having to do tasks you don't feel should be a part of your job7. Being exposed to illness or injury8. Lack of job security9. Having to juggle conflicting tasks or duties

10. The income11. Having too much to do12. Your supervisor's lack of appreciation for your work13. Facing discrimination at your job because of your race, reli-

gion, or ethnic background14. Making less money than other people in your line of work15. The physical conditions on your job (noise, crowding, tem-

perature, environmental hazards, etc.)16. The possibility of unemployment17. The job's taking too much out of you18. Your supervisor's having unrealistic expectations for your

work19. Facing discrimination at your job because of your sex or

gender20. Having to do things against your better judgment21. The job's being physically strenuous22. Limited opportunity for professional or career development23. Having to deal with emotionally difficult situations24. Not being able to get your own job done because of other

people or red tape25. Lack of support from your supervisor26. The sexual harassment you experience27. Making less money than you feel you deserve28. People in jobs like yours being laid off or unemployed

Marital Role Quality

Marital Role Rewards

1. Having a partner who is easy to get along with2. Your partner's doing (his or her) fair share at home3. The physical affection4. Your partner being proud of you5. Your partner's financial contributions to the household6. Your partner's appreciation of you7. Your partner actively encouraging you8. Your sexual relationship9. Your partner's contribution of (his or her) fair share to the

family's finances10. Good communication11. Socializing as a couple12. Your partner's backing you up in what you want to do13. Your partner's concern for members of your family, such as

your parents, brothers or sisters, etc.14. Your partner's finding you physically attractive15. Enjoying the same activities

(Appendix continues on next page)

806 BARNETT, MARSHALL, RAUDENBUSH, AND BRENNAN

16. Having a partner who really talks to you17. The ability of you and your partner to work out conflicts18. Having a partner who is a good friend19. Doing things together for fun20. Your partner doing (his or her) share to make the relationship

work21. Your partner liking you as a person22. Being able to disagree without threatening the relationship23. Your relationship with your partner's family24. Having a partner who is a good listener25. Your partner giving you constructive criticism when you

need it26. Having a lot in common with your partner

Marital Role Concerns

1. Poor communication2. Conflicts over money3. Your partner's job or career problems4. Your partner not understanding who you really are5. \bur partner wanting more than you can give emotionally6. Not getting enough attention from your partner7. Your partner not doing (his or her) share at home8. Your partner's job instability

9. The lack of physical affection10. Your partner's not being home enough11. Your partner being emotionally dependent on you12. Arguing or fighting13. Your partner being too self-absorbed14. Your partner not backing you up in what you want to do15. Lack of companionship16. Your partner's personal problems right now17. Your partner not being tuned in to you18. Your sexual relationship19. Your partner being critical of you20. Not getting along21. Your partner being faithful to you22. \bur partner being overly involved in (his or her) job23. Your partner not earning enough24. Your partner taking you for granted25. Having to do more than your fair share26. You and your partner not feeling emotionally connected to

each other

Received February 28,1992Revision received November 2,1992

Accepted November 13,1992 •

Low Publication Prices for APA Members and AffiliatesKeeping You Up-to-Date: All APA members (Fellows; Members; Associates, and StudentAffiliates) receive—as part of their annual dues—subscriptions to the American Psychologist andAPA Monitor.

High School Teacher and International Affiliates receive subscriptions to the APA Monitor,and they can subscribe to the American Psychologist at a significantly reduced rate.

In addition, all members and affiliates are eligible for savings of up to 60% (plus a journalcredit) on all other APA journals, as well as significant discounts on subscriptions from coop-erating societies and publishers (e.g., the American Association for Counseling and Develop-ment, Academic Press, and Human Sciences Press).

Essential Resources: APA members and affiliates receive special rates for purchases of APAbooks, including the Publication Manual of the APA, the Master Lectures, and Journals in Psychol-ogy: A Resource Listing for Authors.

Other Benefits of Membership: Membership in APA also provides eligibility for low-costinsurance plans covering life, income protection, office overhead, accident protection, healthcare, hospital indemnity, professional liability, research/academic professional liability, stu-dent/school liability, and student health.

For more information, write to American Psychological Association,Membership Services, 750 First Street, NE, Washington, DC 20002-4242, USA