an investigation into the observed sex difference in ...€¦ · christopher s. amenson and peter...

Journal of Abnormal Psychology1981, Vol. 90, No. I, 1-13

Copyright 1981 by the American Psychological Association, Inc.0021-843X/81/9001-0001S00.75

An Investigation into the Observed Sex Differencein Prevalence of Unipolar Depression

Christopher S. Amenson and Peter M. LewinsohnUniversity of Oregon

The rates of occurrence of unipolar depression as measured by self-report andby diagnosis based on semistructured interviews were examined in a longitudinalstudy with a general community sample (N = 998). Number of new cases, av-erage age of onset, and average duration of episodes were comparable in menand women, but women with a history of previous depression were much morelikely to become depressed again than men with a similar history. Data on alarge number of demographic and psychological variables were collected on thesubjects. Controlling for these factors (both singly and in combination) did noteliminate the sex difference in unipolar depression. Neither the artifact hypothesis(according to which the "true" prevalence of depression is equal for men andwomen) nor the psychosocial hypothesis (according to which women are moredisadvantaged on relevant psychosocial variables) was supported by the data.

A consistent finding in the depression lit-erature is the preponderance (often 2:1) offemale depressives. In a comprehensive re-view of epidemiological studies of depres-sion, Weissman and Klerman (1977) notethat this sex difference is observed both instudies of treated cases and in communitysurveys in the United States and in mostWestern countries. (However, some studiesin non-Western countries have not found apreponderance of female depressives.) Themagnitude of the sex difference varies some-what across studies and may be a functionof the depression measure employed and thepopulation studied. Sex differences on moodand symptom ratings and on diagnostic cat-egories are always reported when heteroge-neous samples are employed, but two studies(Hammen & Padesky, 1977; Padesky &Hammen, Note 1) employing a very ho-mogeneous group of college students did notfind a sex difference on two self-report symp-tom rating scales.

From previous studies it is impossible to

This study is based on portions of a doctoral disser-tation submitted by the first author and done under thedirection of the second author. This research was sup-ported in part by Research Grants MH32085 andMH28168 from the National Institute of Mental Health.

Requests for reprints should be sent to Peter M. Lew-insohn, Human Neuropsychology Laboratory, Depart-ment of Psychology-Straub Hall, University of Oregon,Eugene, Oregon 97403.

ascertain whether the elevated prevalence ofdepression (number of cases of diagnosableepisodes of depression at any point in time)among women occurs because women whohave never been depressed are more suscep-tible to becoming depressed (elevated inci-dence), because women who have been pre-viously depressed are more likely to becomedepressed again, and/or because womenhave longer lasting episodes of depression.

The present study focused on sex differ-ences in the occurrence of symptoms ofdepression and of diagnosable episodes ofdepression in a longitudinal study of a het-erogeneous, community-based sample. Thisallowed us to (a) attempt to replicate the sexdifference findings on reported depressivesymptoms and on diagnosis of depression,(b) compare self-report of symptoms anddiagnosis as measures of depression, and (c)examine the effects of incidence, history,recurrence rate, and duration of episodes ofdepression in determining the prevalence ofdepression among women. Following thesedescriptive analyses, several proposed expla-nations of the sex differences in depressionare examined.

Three general hypotheses have been ad-vanced to account for the sex difference: theartifact hypothesis, the biological hypothe-sis, and the psychosocial hypothesis. Theartifact hypothesis maintains that the actual

1

CHRISTOPHER S. AMENSON AND PETER M. LEWINSOHN

prevalence of depression is equal amongwomen and among men but that women per-ceive, acknowledge, report, and seek help fordepression more freely than men do. Thus,observed sex differences in reported symp-toms (cf. Radloff, 1975) and mental healthservice utilization (Rosenthal, 1961) are in-terpreted as representing sex biases thatover-estimate the number of female depres-sives (Phillips & Segal, 1969). Sex biaseshave also been hypothesized for self-labelingand self-presentation regarding depression(Padesky & Hammen, Note 1). Anotherpossible artifact could be constituted by atendency on the part of diagnosticians tooverdiagnose depression in women. From theartifact hypothesis it was predicted that,when symptom level is held constant, womenmore frequently than men will (a) labelthemselves as depressed; (b) be in psycho-therapy; (c) be taking psychotropic medi-cation; and (d) be diagnosed as depressed.

The biological hypothesis postulates thatwomen have a unique vulnerability todepression associated with events in the re-productive cycle or in the sex linked genes.In their review of endocrinological findings,Weissman and Klerman (1977) concludethat although premenstrual tension, use oforal contraceptives, and the postpartum pe-riod all seem to increase rates of depressionfor women, these effects are not of sufficientmagnitude to account for the large sex dif-ference in prevalence rates. Additional re-search is needed to further clarify these is-sues, but the biological hypothesis was notexamined in the present study.

The main thrust of the present study wasto test aspects of the psychosocial hypothesis.Psychosocial explanations of the male-fe-male differential take two major forms. Oneform hypothesizes that interactions betweensex and demographic variables are criticalto understanding the sex differential. Thus,Gove (1972), building on the finding that theincreased prevalence of psychopathologyamong women occurs only for marriedpersons, suggests that negative aspects of thehousewife role are critical determinantsfor the preponderance of psychopathologyamong women. Radloff (1975) makes useof results showing that more symptoms ofdepression are reported not only by women

but also by persons who are young, poorlyeducated, and of low income to suggest that"relative helplessness may be a common de-nominator in the risk factors for depression"(p. 263). The present study investigates thebases of the proposals of Gove and Radloffby examining the relationships between age,education, income, and marital status anddepression. Since the explanations men-tioned above are based on the assumptionthat women are likely to occupy less re-warding societal roles and hence will exhibitlower self-esteem than men, the presentstudy also examines the influence of role sat-isfaction with the prediction that women willbe less satisfied with their jobs, their friends,and their families and will exhibit lower self-esteem.

The second source of potential psychoso-cial explanation is derived from existing psy-chological theories of depression. Each ofthese postulates a cause and effect relation-ship between certain variables and depres-sion, and each theory can be extrapolatedto generate a set of "if-then" predictionssuch that (/"the theory is correct then womenshould be higher on x than men, where x isthe characteristic (or condition) postulatedby the theory to lead to depression. Usingan "integrative" (e.g., Akiskal & McKinney,1973) and "empirical" (e.g., Becker, 1977)stance, according to which many anteced-ents may increase the probability for theoccurrence of depression, the following pre-dictions were formulated from cognitive the-ories (Beck, 1967; Abramson, Seligman, &Teasdale, 1976; Ellis, 1962), from reinforce-ment theories (Lewinsohn, Youngren, &Grosscup, 1979), and from stress theories(Klerman, 1974; Paykel et al., 1969): Ascompared to men, women (a) have lowerexpectations for positive outcomes and higherexpectancies for negative outcomes (Beck,1967); (b) are more likely to attribute suc-cess experiences to "external" causes andfailure experiences to "internal" causes(Abramson, Seligman, & Teasdale, 1976);(c) believe more that outcomes are indepen-dent of their responses (Seligman, 1975);(d) tend more to subscribe to "irrationalbeliefs" (Ellis, 1962); (e) have less response-contingent positive reinforcement (Lewin-sohn, Youngren, & Grosscup, 1979), as re-

SEX DIFFERENCE IN DEPRESSION

fleeted in lower rates of occurrence and sub-jective enjoyment of pleasant activities ingeneral and of interpersonal events in par-ticular; and (f) experience more aversive-ness, as reflected in higher rates of occur-rence and subjective aversiveness of unpleas-ant events and greater occurrence of stress-ful life events (Klerman, 1974), especiallythose involving exits from the social field(Paykel et al., 1969).

For any of the above variables to be de-terminants of the sex difference in depres-sion, it is necessary that the predicted sexdifference on the variable be found. The sec-ond requirement is that controlling for thevariable reduces the apparent sex differencein prevalence of depression.

Method

ParticipantsSubjects were recruited in March 1978 through an

announcement inviting paid participation in psycholog-ical research, which was mailed to 20,000 Eugene andSpringfield, Oregon, residents randomly selected fromthe county voter registration list. Two thousand personsexpressed interest by returning a form in which theygave their name, address, phone number, year of birth,occupation, education, and marital status. In June 1978these 2,000 subjects (63% female) were mailed a 938-item questionnaire that included the Minnesota Mul-tiphasic Personality Inventory (MMPI) Lie scale items.Participants were asked to return (by mail) a signed,informed consent form, the completed questionnaire,and a rating of the likelihood of their moving away fromthe area during the next year. The questionnaire wascompleted by 1213 subjects by September 1978 and allbut 90 of these subjects volunteered to continue in alongitudinal study aimed at "the understanding of psy-chological health and its relationship to what people do,think, and feel."' Ninety-three subjects were excluded(86 because they reported a high probability of movingand 7 because they scored six or higher on the MMPI-Lie Scale). Of the remaining 1030 subjects, 998 con-tinued to participate in the study until its conclusion inJune 1979. These 998 subjects are considered to be thereference sample for this study. Because the sample isself-selected it cannot be considered to be a ran-dom or a representative sample of the population ofEugene-Springfield.

Inspection of demographic characteristics of this sam-ple revealed that the sample differs from the populationof Eugene-Springfield2 in that the majority (69%) arefemale, ages 25 to 34 are overrepresented, there aremore divorced and fewer never-married persons, moreare employed, more have gone to or completed college,and there is an excess of middle income and few highincome subjects.

More crucial to the purposes of the present study isthe comparison of male and female subjects shown in

Table 1, which suggests that sex differences on demo-graphic characteristics are minor and generally parallelthose found in the larger population from which thesample was drawn (i.e., all significant sex differencesin our sample are also found in the population of Eu-gene-Springfield).

Longitudinal Design

Depression, other aspects of psychopathology, andpsychosocial variables were assessed at two times. Thefirst assessment (Tl) is defined as the date on which thesubject returned the first extensive questionnaire (Ql)(i.e., June or July 1978 for most subjects). The secondassessment (T2) is defined as the date on which thesubject came to the University of Oregon PsychologyClinic for a diagnostic interview. Subjects were inter-viewed between October 1978 and June 1979, with theaverage time between Tl and T2 being 8.2 months(range = 3 to 11 months). Two weeks before their in-terviews, subjects were mailed a 205-item questionnaire(Q2), which they completed and brought to the inter-view. Immediately prior to the interview, subjects com-pleted two symptom rating scales.

Case Finding

So that all episodes of depression and other psycho-pathology that occurred between Tl and the end of thestudy could be counted, potential cases were selected tobe interviewed on the basis of high scores on the Centerfor Epidemiologic Studies Depression Scale (CES-D)(Radloff, 1977). The CES-D was part of Ql, and sub-jects were subsequently mailed the CES-D on a tri-monthly basis. Completion rates were high (99%, 99%,and 94% for the three mailings). Any subject who scored18 or above on any administration of the CES-D wasbrought in for an interview. For reasons unrelated tothe present study, subjects were also selected to be in-terviewed if they scored high on certain nondepression-related symptom ratings or if they had changed theirmarital status. By these criteria 598 subjects were se-lected to be interviewed. To be certain that episodes ofdepression were not being overlooked in the remaining400 subjects, a random sample of 100 of these subjectswas also interviewed. All but one of these 100 subjectsdid not experience an episode of depression or otherpsychopathology at Tl, at T2, or at any time during thestudy. Therefore it is assumed that only three cases ofdepression were missed. Since it was impossible to pro-ject the characteristics of these hypothetical cases, thestatistical analyses of prevalence and incidence dataassume that none of the 300 subjects not selected for

' In order to maximize response rates, subjects whodid not promptly return these and subsequent question-naires were reminded with followup letters and phonecalls. Subjects requiring assistance to complete the ques-tionnaires were aided in person or by phone by projectstaff members.

2 More detailed information is available from the sec-ond author.


Table 1Comparison of Demographic Characteristics ofSubjects Participating for the Duration of thisStudy

Characteristic

RaceWhiteOther

Age (years)18-2425-3435-4445-5455-6465 and older

Marital statusMarriedDivorcedSeparatedWidowedNever married

EducationLess than high

schoolHigh school

graduateSome collegeCollege graduate

Family incomeLess than $4000$4000-57999$8000-$ 11, 999Greater than

$12,000

Employment statusEmployedUnemployed-seeking

employmentUnemployed-not

seeking employment

OccupationUpper managerLower managerClerical-salesSkilled or semiskilledHomemakerStudentRetired

%ofmen

(n = 312)

96.73.3

9.642.013.811.512.110.9

61.410.3

1.61.0

25.7

4.5

20.933.140.8

8.714.219.4

57.7

78.1

4.8

17.0

15.412.216.728.60.38.0

15.1

%ofwomen

(« = 686)

98.11.9

13.039.512.215.310.79.2

57.915.62.56.9

17.2*

4.5

29.4*35.929.9**

11.618.821.9

47.7*

64.6**

6.3

29.0**

12.87.3

25.1*8.5**

29.3**7.96.1*

Note. Significance levels were determined by the methodof comparing percentages derived from different samplesizes described by Lawshe and Baker (1950).* p < .05. * * / > < . 01.

an interview was depressed at any time during the study.This erroneous assumption had an extremely small im-pact on our findings.

An additional assumption was used for the compu-tation of prevalence, incidence, and history of depressiondata. Of the 698 subjects selected to be interviewed,9.6% refused to be interviewed, 8.8% had moved too faraway to be interviewed, and the diagnostic informationobtained from 1.1% was deemed unreliable due to poormemory or extreme defensiveness. These 130 selectedbut undiagnosed subjects did not differ from the 568diagnosed subjects on demographic characteristics, basisfor selection, or prior CES-D scores. In subsequent cal-culations we assume that the 130 subjects not inter-viewed experienced episodes of depression with the samefrequency as the 568 subjects who completed the inter-view. The above assumptions all allow calculation ofprevalence, incidence, and history of depression for ourreference sample of 998 subjects.

Assessment of Depression

Diagnostic interview. Diagnoses of depression andother psychopathological syndromes were based on in-formation gathered from participants in a 2-hour semi-structured interview, the Schedule for Affective Disor-ders and Schizophrenia (SADS; Endicott & Spitzer,1978). Interviewers blind to questionnaire data and se-lection procedure made symptom ratings for Tl, T2 andany other period for which psychological disturbancewas suggested. Decision rules specified by the ResearchDiagnostic Criteria (RDC) (Spitzer, Endicott & Rob-ins, 1978) were used to combine the information ob-tained during the interview into specific RDC diagnosticcategories. For each episode of disturbance, the inter-viewer recorded the diagnosis, subject's age at onset,and duration of the episode. In this scheme it is possiblefor episodes to temporally overlap so that a single personmay have more than one diagnosis at a specific time.Each subject was categorized into one of six groups (atTl and again at T2), based on his or her RDC diagnosisat that time. The categories are: major depressive dis-order, minor or intermittent depressive disorder, bipolardepression, other disorders (including anxiety disorders,psychoactive substance abuse, and schizophrenia), mixed(persons with both depressive and nondepressive disor-ders), and not mentally ill. Depression at Tl or T2 isdefined as assignment to the first two groups (i.e., pureunipolar depression). Past history of depression includesany episode of depression, regardless of the presence ofother psychopathology.

The interviewers were a carefully selected group of28 graduate and advanced undergraduate students whowere enrolled in a year-long didactic and practical di-agnostic interviewing course. They underwent extensivedidactic and experiential training in the use of theSADS-RDC. They were required to demonstrate com-petence in the SADS-RDC methodology and decisionrules on two written examinations. To complete training,interviewers were required to agree with over 90% ofthe symptom ratings provided by experts for a video-taped SADS interview and to agree with the "expert"


diagnosis on four of five case histories.3 Initially, ratersconducted interviews in pairs, and interviewer pairingswere rotated to prevent consensual drift. Once an in-terviewer had achieved agreement on three of four con-secutive ratings, he or she did only every fourth interviewin a pair. The kappa statistic (Cohen, 1960), whichmeasures how well two raters agree beyond the levelpredicted by chance, was used to measure reliability.For individual symptom ratings, kappa (requiring exactagreement) was found to average .75. In other words,our raters agreed on 75% of the instances when bychance alone they would have disagreed. For RDC di-agnoses, weighted kappa was computed by using a com-puter program provided by Spitzer and Endicott. Forthe first four diagnostic categories used in this study,weighted kappa was .88 for not mentally ill, .96 formajor depressive disorder, .88 for minor or intermittentdepressive disorders, and .78 for bipolar depression. Forthe several diagnoses in the last two categories (otherdisorders and mixed diagnosis), kappa ranged from 0to 1.0, but for all diagnoses that occurred in more thanfour cases, kappa was at least .6. Even for infrequentdiagnoses, the median kappa was .80.

CES-D. The Center for Epidemiologic Studies-Depression Scale is a self-report measure of frequencyof occurrence of 20 depressive symptoms designed foruse in general community samples. Each symptom israted for the past week. The CES-D has been shown topossess adequate psychometric properties and to cor-relate substantially with other self-report measures(Radloff, 1977).

BSI. The Brief Symptom Inventory is a shortenedversion of the SCL-90 (Derogatis, Lipman, & Covi,1973), which has been shown to possess good concurrentvalidity (Derogatis & Cleary, 1977). The BSI is com-posed of 53 items that reflect nine primary symptomdimensions. The depression subscale consists of ratingsof intensity of experience of six depressive symptoms.The BSI depression scale (intensity ratings) and theCES-D (frequency ratings) were found to correlatehighly (r = .79, p < .001). All analyses done with theCES-D were also done with the BSI. Results were quitesimilar, therefore only results utilizing the CES-D arereported.

Mental health service utilization. Eight items in Qlwere designed to elicit information about help-seekingbehaviors. In separate questions subjects rated whetherthey had presently (at Tl) or in the past been in psy-chotherapy, used psychotropic medication, receivedelectroconvulsive therapy, or been hospitalized fordepression. During the SADS interview subjects wereasked to verbally report on help-seeking behaviorsatT2.

Self-labeling of depression. As part of Ql, subjectswere given a one-paragraph, written definition of clinicaldepression and were asked to rate their current depres-sion level on a 5-point scale. Subjects were also askedto state whether they had ever been depressed, how longtheir longest episode of depression lasted, and at whatage their first episode began.

Measures of Independent VariablesDemographic variables. Subjects reported their sex,

age, race, marital status, educational level (Rollings-

head & Redlich, 1958), family income, occupationalstatus (Hollingshead & Redlich, 1958), and employ-ment status on three occasions (at initial subject re-cruitment, at Tl, and at T2). This allowed assessmentof changes in status and assurance that the same personwas completing all phases of the study.

Psychological variables. So that as many of thesevariables could be assessed as possible in a questionnaireof manageable size, the original inventories were short-ened by a two-step procedure. The first step consistedof selecting individual items that have been shown inprevious studies to be especially strongly related todepression. The number of these items for each measurewas further reduced by using a computer program de-veloped by Serlin and Kaiser (1976) to select the subsetof the original item set that yields maximum internalconsistency as measured by coefficient a. In this waythe following scales were created:

1. Feelings of satisfaction. Fifteen items shown to berelated to "happiness" and "feelings of satisfaction"were chosen from among those used by Andrews andWithey (1976) and Cambell, Converse, and Rogers(1976). On the basis of interitem correlations and factoranalysis, items were grouped into six subscales repre-senting feelings about (a) self (a = .96), (b) family(a = .91), (c) job (a = .96), (d) neighborhood (a = .89),(e) friends (a = .94), and (f) leisure time activities(a = .90).

2. Self-esteem. A 23-item self-esteem scale, similarto the Semantic Differential Inventories constructed byCoyne and Holtzman (1966), was developed by Flippoand Lewinsohn (1971). Five items (a = .85) were se-lected from this scale to be rated both according to theinstructions "how you see yourself" and "how you wouldlike to be." Real self-esteem and "real minus ideal" self-esteem scores were computed.

3. Expectancies of positive and negative outcomes.The Subjective Probability Questionnaire (Mufioz &Lewinsohn, Note 2) was designed to operationalizeBeck's "cognitive triad" by having subjects rate theprobability of occurrence of negative and positive out-comes. Six positive items (a = .86) and four negativeitems (a = .75) were selected from among the itemsshown to most strongly discriminate between depressedand nondepressed persons (Lewinsohn, Larson, & Mu-fioz, Note 3).

4. Locus of control. The Multidimensional Multiat-tributional Causality Scale (Lefcourt, von Baeyer,Ware, & Cox, 1979) affiliation items encompass a threefactor design of causality (internal vs. external, stablevs. unstable, and success vs. failure). For each cell, twoof the three items provided were arbitrarily chosen.

5. Perception of control. Three items (a = .62) wereselected from an original group of seven items designedto assess perceptions of control over one's life (e.g., Ihave little control over the things that happen to me.).(Lewinsohn, Note 4)

6. Irrational beliefs. The Personal Beliefs Inventory(Mufioz & Lewinsohn, Note 5) samples irrational be-liefs that have been hypothesized to be associated withdepression. Five items (a = .66) were selected from the

3 We are grateful to Jean Endicott and Robert Spitzerfor providing these and other training materials.


subset of items shown to most strongly discriminate be-tween depressed and nondepressed populations (Lew-insohn et al., Note 3).

7. Pleasant events. Twenty items were selected fromthe mood-related events of the Pleasant Events Schedule(PES) (MacPhillamy & Lewinsohn, Note 6). The fre-quency of occurrence (a = .90) and the enjoyability(a = .93) of the events were rated separately and thecross-product score was assumed to represent a measureof positive reinforcement (Lewinsohn, Weinstein, &Shaw, 1969).

8. Interpersonal events. Twenty items were selectedfrom the subset of Interpersonal Events Schedule (IBS)(Youngren, Lewinsohn, & Zeiss, Note 7) that best dis-criminated between depressed and nondepressed per-sons. Subjects rated the frequency of occurrence (a =.84) and comfort (ct = .92) of the events, and the cross-product score was computed and assumed to reflect in-terpersonal reinforcement.

9. Unpleasant events. Twenty items were selectedfrom the mood-related items (Lewinsohn & Amenson,1978) of the Unpleasant Events Schedule (UES) (Lew-insohn, Note 8). Subjects rated the frequency (a = .84)and aversiveness (a = .90) of the events. The cross-prod-uct score was assumed to represent experienced aver-siveness (Lewinsohn & Talkington, 1979).

10. Stressful life events. Twenty-three items werearbitrarily chosen from the Social Readjustment RatingScale (Holmes & Rahe, 1967). Life change unit scoresare assumed to reflect the stressful impact of significantlife changes.

Results

Sex Differences in Prevalence, Incidence,Previous History, and Duration ofEpisodes of Unipolar Depression

As can be seen in Table 2, a significantlyhigher percentage of women than of menmet the RDC criteria for a diagnosis ofunipolar depression at the time of the initialquestionnaire (Tl) and at the follow-up in-terview (T2).4 The elevated prevalence ofdepression among women is not paralleledby a general increase in prevalence in otherdiagnostic categories.5

Further examination of Table 2 indicatesthat the difference in prevalence occurs be-cause women with a history of depressionare more likely than men to develop a newepisode of depression. The elevated preva-lence among women is not due to sex dif-ferences in duration of episodes or age atfirst onset. The finding of no significant sexdifference in incidence of depression seemsinconsistent with the finding of a prepon-derance of women among persons with a

history of depression. However, a differencein incidence of less than 1%/year (unde-tectable with a sample size of 1000) couldaccount for the 13% difference in previousepisodes accumulated over a period of 15 to20 years.

Sex Differences on Other Measuresof Depression

As expected, women scored higher on allof the other measures of depression andmental health service utilization. These re-sults and their level of significance are shownin Table 3.

Testing the Artifact Hypothesis

To test the predictions that at equal levelsof depressive symptomatology women aremore likely than men to label themselves asdepressed and to seek help for depression,men and women were divided into high,medium, and low symptom level groupsbased on their Tl CES-D scores. This sub-division resulted in male and femalesubgroups with very similar mean CES-Dscores.6 Contrary to predictions from theartifact hypothesis, no differences were found(all x2s < 1) in treatment seeking or self-la-beling among men and women reportingequal levels of symptoms. Analogous anal-yses were also done by grouping men andwomen according to Tl BSI Depressionscale scores. Results were comparable.

The degree to which males and femaleswith comparable self-reported symptomswould be diagnosed as depressed was ex-amined next. For this analysis men andwomen were grouped according to scores onthe CES-D taken just prior to the diagnosticinterview. The chi-square analyses revealedno significant interviewer or subject gendereffects on diagnosis; that is, men and women

4 The percentages at T2 probably overestimated theactual prevalence of depression in this sample, since thistime point varies in a nonrandom fashion over an 11-month span, with subjects being more likely to be in-terviewed near times when they scored above 18 on theCES-D.

5 Data for other categories are available from the sec-ond author.

6 Scores are available from the second author.

SEX DIFFERENCE IN DEPRESSION 7

Table 2Comparison of Men and Women on Occurrence and Duration of Episodes of Depression at Tl, atT2, and in Lifetime History as Assessed in SADS-RDC Interview

DepressionMen

(» = 312)Women

(« = 686)

Prevalence% depressed at Tl% depressed at T2

History"% ever having been depressedM no. of episodes/depressiveM duration of episodes (weeks)Mdn duration of episodes (weeks)6

M age of first onset (years)Mdn age of first onset (years)

Percent developing new episodes ofdepression/year0

Of all subjectsOf subjects with a history of depressionOf subjects never before depressed

5.1' 7.1

48.81.1

71.624.327.023.0

9.812.96.9

11.4**12.8**

62.3**1.4*

71.819.527.023.0

16.2*21.8**

7.1

Note. Tl = Time of first assessment; T2 = time of second assessment. Significance levels were determined by themethod of comparing percentages derived from different sample sizes described by Lawshe and Baker (1950)." These figures are for all episodes of depression that occurred in the subject's lifetime, including any episodes thatoccurred during the course of the study.b Median duration is a better description of length of episode, since the mean duration is elevated by a few episodesof very long (> 15-year) duration.' These figures are prorated for one year based on the occurrence of new episodes over an average time span of8.3 months between Tl and T2.*p<.05 . **p<.01.

with equal symptom levels were equallylikely to be diagnosed as depressed by maleand by female interviewers.

Testing the Psychosocial Hypothesis

Each demographic and psychological vari-able was tested for three properties necessaryfor it to be a determinant of the sex differ-ences in prevalence of episodes of depressionand in self-report of depressive symptoms:(a) Does the variable have a significant re-lation with depression? (b) Do males andfemales differ significantly on the variable?and (c) Does controlling for the variable re-duce the variance in depression attributableto sex?

Demographic variables. Multiple corre-lation analysis was used as an overall test ofthe relations of demographic variables withCES-D (i.e., self-reported depressive symp-toms) and with diagnosis of depression (i.e.,frequency of occurrence of episodes of clin-

ical depression). The multiple correlation ofdemographic variables (excluding sex) was.21 (p < .001) with CES-D and .11 (p< .05)with diagnosis of depression. Analysis ofvariance (for nominal categories) and cor-relations (for ordinal categories) found youth(p < .05), divorce or separation (p < .05),low education (p < .05), low family income(p < .01), and unemployment (p < .01) tobe associated with high CES-D scores. Onlydivorce or separation (p < .05) exhibited asignificant relation to diagnosis of depres-sion.

Analyses of variance and chi-square anal-yses revealed no significant differences be-tween male and female subjects on age orfamily income (F < 1), but women wereless likely to be in the never-married status(p < .05), and to be lower on educational(p < .05) and occupational (p < .05) leveland more likely to be unemployed (p < .001)than men.

Hierarchical multiple regression analyses

8 CHRISTOPHER S. AMENSON AND PETER M. LEWINSOHN

Table 3Sex Differences in Measures of Depression and Mental Health Service Utilization

Measureof men

(« = 312)of women(n = 686)

Historical3

Ever receiving psychotherapy fordepression

Ever hospitalized for depressionEver receiving psychotropic drugs for

depressionLabeling self as having been depressed for

2 weeks or more

15.12.5

10.9

51.7

23.1**4.8*

25.3**

58.8*

Time 1Receiving psychotherapyTaking psychotropic drugsLabeling self as moderately or very

depressedScoring above 18 on CES-Depression ScaleScoring above 1 .0 on BSI-Depression Scale

Time 2Receiving psychotherapyTaking psychotropic drugsScoring above 18 on CES-Depression ScaleScoring above 1 .0 on BSI-Depression Scale

3.24.8

11.616.017.0

(n = 181)1.83.8

13.214.0

4.26.2

16.0*23.3**21.1*

(n = 387)4.5*8.2*

21.4**16.0

Note. Significance levels were determined by the method of comparing percentages derived from different samplesizes described by Lawshe and Baker (1950). CES = Center for Epidemiologic Studies; BSI = Brief SymptomInventory.a These measures were obtained via retrospective questions completed by the subjects at Time 1.*p< .05 . **p<.01.

were used to determine to what extent thedemographic variables could reduce the sexdifferences in CES-D scores and diagnosisof depression. In this technique the variancein depression measure due to sex is measuredprior to and after the variation due to otherdemographic variables is extracted. For CES-D, extracting the variance due to other dem-ographics reduced the zero-order correla-tion of .11, F(l, 972) = 10.8, p < .001, toa residual correlation of .06, F( 1, 972) = 3.9,p < .05. Thus, simultaneously controlling fordemographic variables reduced the correla-tion between CES-D score and sex by ap-proximately 45%. For diagnosis of depres-sion, extracting the variance due to demo-graphics did not alter the magnitude of thesex difference. The residual correlation(r = .17), F(l, 444) = 12.8, p < .001, isidentical to the zero order correlation (r =.17), F(\, 444) = 12.4,p< .001. The resultsthus indicate that even after the effects ofthe demographic variables have been par-

tialled out, sex continues to be significantlyassociated with CES-D and with diagnosisof depression.

Psychological variables. Multiple regres-sion analyses demonstrated strong relationsof the psychological variables with CES-D(R = .73, p < .001) and with diagnosis ofdepression (R = .5l, p < .001). With theexception of the internal attribution for fail-ure measure, all of the psychological vari-ables were found to be associated (at varyinglevels of strength and statistical significance)with both CES-D and diagnosis of depres-sion.7

Women did not score significantly morein the "depressed" direction than men onany of the psychological variables. Contraryto the psychosocial hypotheses, men wereless satisfied with their friends (p < .05) andtheir neighborhood (p < .05), were less likely

7 Details of this analysis are available from the secondauthor.


to attribute success to internal (p < .001) orfailure to external (p < .05) causes, engagedin fewer (p < .001) and derived less enjoy-ment from (p < .001) pleasant events, andengaged in fewer (p < .05) interpersonal in-teractions. Since 27 variables were exam-ined, results at the p < .05 significance levelshould be interpreted cautiously. Nonethe-less these results do not implicate any ofthese 27 variables as determinants of the sexdifferences in CES-D and diagnosis ofdepression.

To test the possibility that nonsignificantsex differences on the psychological vari-ables could significantly alter the relationsbetween sex and CES-D or diagnosis ofdepression, hierarchical multiple regressionanalyses using all the psychological vari-ables, with sex entered last, were performedon the CES-D and diagnosis of depressiondata. The relation between sex and CES-Dscore (zero order r = .11), F(l, 972) = 10.8,p = .001, was unaltered by extracting thevariance in CES-D due to the 27 psycholog-ical variables (residual r = .12), F(l,972) = 11.9,/> < .001. Similarly, the relationbetween sex and diagnosis of depression(zero order r = .17), F(l, 444) =12.8,p < .001, was also changed very little byprior extraction of variance in diagnosis ofdepression due to the psychological variables(residual r = .15), F(l, 444) = 6.9, p < .01.Given that the psychological variables takentogether were unable to reduce the magni-tude of the sex differences in CES-D anddiagnosis of depression, no individual vari-able could have significantly reduced the sexdifferences. Individual covariance analysesthat partialed out each psychological vari-able one at a time supported this conclusion.

Discussion

Limitations Imposed by SampleRecruitment Strategy

Since the major goal of this study was totest numerous hypotheses about the sex dif-ference in the occurrence of depression, itwas vital to collect a large amount of data(subjects answered over 1500 questions overa 1-year span) on the participants. Our ef-forts in this regard were at the expense of

obtaining the participation of a larger pro-portion of the original, randomly selectedsample of 20,000. Only 5% of the originalsample volunteered to participate in the lon-gitudinal study, but 84% of these completedthe study. This self-selection bias seriouslylimits the generalizability of the epidemio-logic findings generated by this sample. Itis likely, for example, that our prevalencefigures for unipolar depression (11.5% forwomen and 5.1% for men) overestimate theactual prevalences (which have been esti-mated at 6% for women and 4% for men byLehman, 1971) because persons with a pro-pensity toward depression are overrepre-sented in the final sample. However, thissample bias does not vitiate the testing ofhypotheses regarding sex differences on thedata unless one posits that an additional sexspecific selection bias is present. The dataon the characteristics of the male and femalesubgroups argue against a sex specific selec-tion bias. The male and female subsamplesdiffer only in ways that males and femalesdiffer in the larger population (e.g., moremales are college graduates), and large andcomparable (for males and females) vari-ances obtain for all of the independent vari-ables.

Epidemiologic Findings

Our results replicate previous studies inshowing a substantially elevated prevalenceof unipolar depression in women as com-pared with men. Unexpectedly, however, theincidence of depression (i.e., number of per-sons without a history of previous depressionwho became depressed during the course ofthe study) in men and in women was quitecomparable. Women did not have longerlasting episodes; nor were there any differ-ences in age of first onset. The major dif-ference between the sexes was observed inpersons with a history of previous depression.Women with a history of previous depressionwere much more likely to become depressedagain (21.8%) than men with a history ofprevious depression (12.9%).

Artifact Hypothesis Findings

Contrary to the artifact hypothesis, menand women matched for symptom level were


equally likely to label themselves as de-pressed, to seek help for depression, orto be diagnosed as depressed by trained in-terviewers.

Our failure to observe a sex bias in self-labeling is consistent with earlier studies(I^ammen & Padesky, 1977). In fact thepresent study suggests that self-labeling isquite accurate. Self-labeling and clinical di-agnosis of depression were in agreement for81% of female and 92% of male subjects.The only bias that was found was that, com-pared to men, women diagnosed as depressedless frequently labeled themselves as de-pressed, %2(2) = 14.0, p < .01. This findingsuggests that studies which rely on self-di-agnosis may actually underestimate the rateof occurrence of depression among women.

The absence of a sex bias in help-seekingbehavior is also consistent with earlier find-ings (Padesky & Hammen, Note 1) thatmale and female college students of com-parable depression level exhibited no sexdifferences in their ratings of the degree ofdepression required to seek treatment.

It is important to note that our negativeconclusions regarding the artifact hypothesisare predicated on the assumption that self-report of intensity (measured by the BriefSymptom Inventory—Depression Scale) orof frequency (measured by the Center forEpidemiologic Studies—Depression Scale)of depressive symptoms is a valid measureof the presence of these symptoms. Althoughhighly improbable, it remains possible thata sex bias that has equivalent effects on self-report, self-labeling, help seeking, and di-agnosis exists that accounts for the observedsex differences in depression. Two types ofevidence can be cited against this argument.Strong evidence exists for the concurrentvalidity of the Center for EpidemiologicStudies and the Brief Symptom InventoryDepression scales (Radloff, 1977; Derogatis& Cleary, 1977). Second, Clancy and Gove(1974) found women to be less likely to re-port symptoms and to show more evidenceof common response biases that could beassociated with under-reporting of symp-toms, such as answering items in a sociallyapproved direction and nay saying. Thus, ourassumption that groups matched on the BSIand on the CES-D were indeed comparable

in the actual occurrence of depression symp-toms seems reasonable. We interpret ourfindings as supporting Weissman and Kler-man's (1977) conclusion that "we must re-gard the sex differences as real findings andexamine the possible explanations" (p. 103).

Psychosocial Hypothesis Findings

Demographic variables. Radloff s (1977)findings that high CES-D scores are relatedto youth, divorce and separation, low in-come, low education, and unemploymentwere replicated, but controlling for thesevariables did not eliminate the sex differenceon the depression measures. Our study alsoreplicates a recently reported study by Rad-loff and Rae (1979), in which it was foundthat controlling for psychosocial variablesdid not eliminate sex differences on theCES-D. Gove's (1972) hypothesis that thehousewife role is the critical determinant ofthe sex difference in depression was not sup-ported in the present study. On all measuresof depression, women scored higher thanmen regardless of marital status, andhousewives were no more depressed thanworking wives or nonmarried women. Ourfindings are consistent with Radloff s (1975)finding that housewives do not score signif-icantly higher on the CES-D than workingwives but inconsistent with her finding of nodifferences on the CES-D between unmar-ried men and unmarried women.

In contrast to the statistically significantrelations that were found between income,educational level, and employment statusand CES-D, there were no significant rela-tions between these demographic variablesand diagnosis of depression; consequently,controlling for these variables did not reducedifferences between men and women in num-ber of diagnosable cases. The fact that di-agnosis of depression was not found to berelated to these social class variables is im-portant in its own right. The relation be-tween social class and depression has beenambiguous in that studies employing mea-sures of depressive moods (Levitt & Lubin,1975) or depressive symptoms (Langner &Michael, 1963) have reported significant re-lations, whereas studies using diagnoses ofdepression (Hollingshead & Redlich, 1958)


have not. The findings of the present studyare consistent with the proposition that so-cial class and other demographic measuresare related to self-report of depressive symp-toms but not to diagnosis of depression.Since the CES-D and diagnosis of depressionwere highly correlated (r = .7), this diver-gence of results deserves some attention. Inthe present study only 50% of persons scor-ing above 18 on the CES-D on the day oftheir interview were diagnosed as depressedaccording to the RDC criteria, and only 50%of diagnosed cases scored above 18 on theCES-D on the day of the interview. Eithermeasure by itself thus would have led to alarge number of misclassifications, if theother measure were used as the criterion. Wesuggest that the differences between thesetwo measures are two-fold. For a diagnosisof depression, a person must experience clearsymptoms continuously for at least twoweeks, whereas the CES-D score may beelevated due to some short-lived event. Inthe present study it was frequently observedthat subjects scoring high on the CES-D whodid not meet the RDC criteria reported atransient physical or emotional disturbanceat the time they took the CES-D. For ex-ample, the first person of this type that wasinterviewed reported having knee surgery inthe week prior to completing the CES-D.The second difference is that the diagnosisof depression requires symptoms to be or-ganized into a specific syndrome, whereasCES-D items (symptoms) may be related tomedical events or other stressful life eventsthat may cause the person to be fearful, tosleep poorly, or to have trouble concentrat-ing (all of these are items on the CES-D).The reported relationships between the CES-D and low socioeconomic status may thusreflect a more general relationship betweensocial class and stressful life events, ratherthan being unique to depression. The presentresults clearly indicate that the CES-D score(and probably other self-report measures ofdepression as well) and the diagnosis ofdepression are only partially overlappingdomains and that a high CES-D score shouldnot be equated with clinical diagnosis ofdepression,

Psychological variables. As required byour overall strategy, most of the psycholog-

ical variables that had been included weresignificantly related to depression as mea-sured by the CES-D and by diagnosis. Theonly exception was the attribution of cau-sality measure, which did not consistentlyshow the expected pattern of attributing suc-cess to external factors and failure to inter-nal factors that has been postulated byAbramson, Seligman, and Teasdale (1976)to characterize depressives.

Contrary to the psychosocial hypothesis,women did not score in the more depresseddirection on the psychological variables thanmen. In fact, women expressed more satis-faction with their neighborhood and withtheir friends, were more likely to attributesuccess experiences to internal causes andfailure experiences to external causes, andreported themselves as engaging in morepleasant activities and enjoying them more.Thus the psychosocial variables could not(and did not) reduce the difference in therate of occurrence of depression betweenmen and women. It appears that women aremore likely to be depressed in spite of scoringas more "healthy" on psychological vari-ables related to depression.

Future Research Directions

Since none of the variables included in thisstudy singly or in combination eliminatedthe observed sex difference, the question re-mains, why are women who have been de-pressed in the past more likely to becomedepressed again than men? It is, of course,possible that critical psychosocial variableswere omitted. Two immediately come tomind. Interpersonal dependency has beenhypothesized to be an antecedent for depres-sion by Hirschfeld et al. (1976), and Young(1975) found that recently widowed or di-vorced women who were high in degree of"traditional sex role acceptance" were morelikely to be depressed than widowed or di-vorced women who were lower on the sexrole scale. Another phenomenon that maybe relevant is the greater social rejection ofdepression in men than in women reportedby Hammen and Peters (1977, 1978). Intheir study, subjects read a case history ofeither a male or a female college studentdescribed as having one of three reactions


to stress: depression, anxiety, or flat-affectdetached response. Results of the studyshowed that depression elicited more rejec-tion of males than of females, and the sexdifference in rejection was more pronouncedfor depression than for anxiety or flat-affectdetached responses. If, as the study suggests,depression is less socially acceptable for menthan for women, then men would be morelikely to be punished for expressing theirfeelings of depression. Alternatively, womenmay elicit more social reinforcement (sec-ondary gain) from their depression. Theseconsiderations would not be expected to af-fect the occurrence of first episode of depres-sion but a differential social response couldconceivably result in a higher probabilitythat women with a previous history ofdepression will become depressed again thanthat men with a similar history will.

The results of the present investigationsuggest that unipolar depression may be evenmore common than is usually thought. Theprevalence and incidence figures were higherthan expected, and 49% of the men and 62%of the women had experienced a diagnosableepisode of depression some time during theirlives. In order to be cost effective, therefore,treatment and prevention programs fordepression must be broadly applicable. Fi-nally, it is clear that individuals with a pre-vious history of depression, and especiallyfemales, are at a very high risk for depres-sion, and treatment and prevention programsshould be aimed at them.

Reference Notes1. Padesky, C. A., & Hammen, C. L. Sex differences

in depressive symptom expression and help-seekingamong college students. Unpublished mimeograph,University of California at Los Angeles, 1977.

2. Mufioz, R. F., & Lewinsohn, P. M. The SubjectiveProbability Questionnaire. Unpublished mimeo-graph. University of Oregon, 1976.

3. Lewinsohn, P. M., Larson, D., & Mufioz, R. F. Themeasurement of expectancies and other cognitionsin depressed individuals. Unpublished mimeograph,University of Oregon, 1980.

4. Lewinsohn, P. M. Mastery scale. Unpublished mim-eograph, University of Oregon, 1971.

5. Munoz, R. F., & Lewinsohn, P. M. The PersonalBeliefs Inventory. Unpublished mimeograph, Uni-versity of Oregon, 1976.

6. MacPhillamy, D. J., & Lewinsohn, P. M. The Pleas-ant Events Schedule. Unpublished mimeograph,University of Oregon, 1971.

7. Youngren, M. A., Lewinsohn, P. M., & Zeiss, A. M.The Interpersonal Events Schedule. Unpublishedmimeograph, University of Oregon, 1975.

8. Lewinsohn, P. M. The Unpleasant Events Schedule.Unpublished mimeograph, University of Oregon,1975.

ReferencesAbramson, L. Y., Seligman, M. E. P., & Teasdale, J.

Learned helplessness in humans: Critique and refor-mulation. Journal of Abnormal Psychology, 1976,86, 78-97.

Akiskal, H. S., & McKinney, W. T., Jr. Depressivedisorders: Toward a unified hypothesis. Science, 1973,182, 20-29.

Andrews, F. M., & Withey, S. B. Social indicators ofwell-being. Americans' perceptions of life quality.New York: Plenum Press, 1976.

Beck, A. T. Depression: Clinical, experimental, and the-oretical aspects. New York: Hoeber Medical Press,1967.

Becker, J. Affective disorders. Morristown, N.J.: Gen-eral Learning Press, 1977.

Cambell, A., Converse, P. E., & Rogers, W. L. Thequality of American life. New York: Russell SageFoundation, 1976.

Clancy, K., & Gove, W. Sex differences in mental ill-ness: An analysis of response bias in self reports.American Journal of Sociology, 1974, 80, 205-216.

Cohen, J. A coefficient of agreement for nominal scales.Educational and Psychological Measurement, 1960,20, 37-46.

Coyne, L., & Holzman, P. Three equivalent forms ofa semantic differential inventory. Educational andPsychological Measurement, 1966, 26, 665-674.

Derogatis, L. R., & Cleary, P. A. Confirmation of thedimensional structure of the SCL-90. A study in con-struct validation. Journal of Clinical Psychiatry,1977, 33, 981-989.

Derogatis, L. R., Lipman, R. S., & Covi, L. SCL-90:An outpatient psychiatric rating scale—Preliminaryreport. Psychopharmacology Bulletin, 1973, 9, 13-27.

Ellis, A. Reason and emotion in psychotherapy. NewYork: Lyle Stuart, 1962.

Endicott, J., & Spitzer, R. L. A diagnostic interview:The Schedule for Affective Disorders and Schizo-phrenia. Archives of General Psychiatry, 1978, 35,837-844.

Flippo, J. R., & Lewinsohn, P. M. Effects of failureupon the self-esteem of depressed and nondepressedsubjects. Journal of Consulting and Clinical Psy-chology, 1971, 36, 151.

Gove, W. R. The relationship between sex roles, maritalstatus, and mental illness. Social Forces, 1972, 51,34-44.

Hammen, C. L., & Padesky, C. A. Sex differences inthe expression of depressive responses on the BeckDepression Inventory. Journal of Abnormal Psy-chology, 1977, 86, 609-614.

Hammen, C. L., & Peters, S. D. Differential responsesto male and female depressive reactions. Journal ofConsulting and Clinical Psychology, 1977, 45, 994-1001.


Hammen, C. L., & Peters, S. D. Interpersonal conse-quences of depression, responses to men and womenenacting a depressed role. Journal of Abnormal Psy-chology, 1978, 87, 322-332.

Hirschfeld, R. M. A., et al. A measure of interpersonaldependency. Journal of Personality Assessment, 1976,41, 610-618.

Hollingshead, A. B., & Redlich, F. C. Social class andmental illness. New York: Wiley, 1958.

Holmes, T. H., & Rahe, R. H. The Social ReadjustmentRating Scale. Psychosomatic Medicine, 1967, / / ,213-218.

Klerman, G. L. Depression and adaptation. In R. J.Freidman & M. M. Katz, (Eds.), The psychology ofdepression: Contemporary theory and research.Washington, D.C.: V. H. Winston, 1974.

Langner, T. S., & Michael, S. T. Life stress and mentalhealth. New York: Glencoe Press, 1963.

Lawshe, C. E., & Baker, P. C. Three aids in the eval-uation of the significance of the difference betweenpercentages. Educational and Psychological Mea-surement, 1950, 10, 263-270.

Lefcourt, H. M., von Baeyer, C. L., Ware, E. E., &Cox, D. J. The multidimensional-multiattributionalcausality scale: The development of a goal specificlocus of control scale. Canadian Journal of Behav-ioural Science, 1979, //, 286-304.

Lehman, E. E. Epidemiology of depressive disorders. InR. R. Fieve (Ed.), Depression in the 1970's: Moderntheory and research. Princeton, N.J.: Excerpta Med-ica, 1971.

Levitt, E. E., & Lubin, B. Depression. New York:Springer, 1975.

Lewinsohn, P. M., & Amenson, C. Some relations be-tween pleasant and unpleasant mood related activitiesand depression. Journal of Abnormal Psychology,1978,57,644-654.

Lewinsohn, P. M., & Talkington, J. Studies on themeasurement of unpleasant events and relations withdepression. Applied Psychological Measurements,1979,5,83-101.

Lewinsohn, P. M., Weinstein, M., & Shaw, D. Depres-sion: A clinical research approach. In R. D. Rubin

& C. M. Frank (Eds.), Advances in behavior therapy,1968. New York: Academic Press, 1969.

Lewinsohn, P. M., Youngren, M. A., & Grosscup,S. J. Reinforcement and depression. In R. A. Depue(Ed.), The psychobiology of depressive disorders:Implications for the effects of stress. New York:Academic Press, 1979.

Paykel, E. S., et al. Life events and depression: A con-trolled study. Archives of General Psychiatry, 1969,21, 753-760.

Phillips, D., & Segal, B. Sexual status and psychiatricsymptoms. American Sociological Review, 1969, 34,58-72.

Radloff, L. Sex differences in depression: The effectsof occupational and marital status. Sex Roles, 1975,/, 249-265.

Radloff, L. The CES-D Scale: A self-report depressionscale for research in the general population. AppliedPsychological Measurement, 1977, /, 385-401.

Radloff, L., & Rae, D. S. Susceptibility and precipi-tating factors in depression: Sex differences and sim-ilarities. Journal of Abnormal Psychology, 1979, 88,174-181.

Rosenthal, S. H. Changes in a population of hospitalizedpatients with affective disorders, 1945-1965. Amer-ican Journal of Psychiatry, 1961, 123, 671-681.

Seligman, M. E. P. Helplessness. San Francisco: Free-man, 1975.

Serlin, R. C., & Kaiser, H. F. A computer program foritem selection based on maximum internal consis-tency. Educational and Psychological Measurement,1976, 36, 757-759.

Spitzer, R. L., Endicott, J., & Robins, E. ResearchDiagnostic Criteria: Rationale and reliability. Ar-chives of General Psychiatry, 1978, 35, 773-782.

Weissman, M. M., & Klerman, G. L. Sex differencesand the epidemiology of depression. Archives of Gen-eral Psychiatry, 1977, 34, 98-111.

Young, K. A. Sex role acceptance and depression inmiddle aged women. Dissertation Abstracts Inter-national, 1975, 36, 4068.

Received March 4, 1980Revision received September 2, 1980 •

an investigation into the observed sex difference in ...€¦ · christopher s. amenson and peter...

Documents