You are on page 1of 21

Sex Differences in Distress: Real or Artifact?

Author(s): John Mirowsky and Catherine E. Ross

Source: American Sociological Review, Vol. 60, No. 3 (Jun., 1995), pp. 449-468
Published by: American Sociological Association
Stable URL:
Accessed: 11/05/2010 11:28

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to
American Sociological Review.

John Mirowsky CatherineE. Ross

TheOhioState University TheOhioState University
Womenreportgreater distress than men, but do womengenuinely experience
greater distress, suggesting a heavier burdenof hardshipand constraint? Or
do they merely report thefeelings in standard indexes morefrequently? Per-
haps women discuss their emotions morefreely. Or perhaps the indexes tap
"feminine"emotions such as depression rather than "masculine"ones such
as anger. This study analyzes data from a 1990 U.S. sample of 1,282 women
and 749 men. Results show that men keep emotions to themselves more than
women, and that women express emotions more freely than men. However,
these factors do not explain the effect of sex on reported levels of distress-
an effect that remains significant with adjustmentfor these factors. Our re-
sults also contradict the idea that the sex difference in distress would dimin-
ish if the indexes of distress contained more items that tap anger Adjusting
for emotional reserve and expressiveness, women experience anger more of-
ten than men, as they do sadness, anxiety, malaise, and aches. In fact, being
female has twice the effect on the frequency of anger that it has on the fre-
quency of sadness. Womenreportfeeling happy as often as men, but adjust-
ing for emotional expressiveness reveals a negative effect of being female on
happiness. Overall, women experience distress about 30 percent more often
than men. Wediscuss thepossibility that drug abuse and heavy drinkingmask
male distress, but find little evidence that those behaviors ameliorate dis-
tress. Weconclude that women genuinely suffer more distress than men.

A mericansurveysfind that womenre- freely than men, and thus appear more dis-
port higher average levels of depres- tressed. Or, women may respond to stressors
sion and anxiety than men (Aneshensel 1992; with somewhat different emotions than men.
Mirowsky and Ross 1986). What explains Thus, if surveys ask more questions about
these sex differences in reports of distress? responses typical of women than about those
One possibility is that women genuinely suf- typical of men, women may falsely appear
fer greater distress than men. If so, the dif- more distressed.
ference in distress reflects and reveals
women's relative disadvantage in American
society. Another possibility is that women
may simply express their emotions more DEPRESSION
Theories of gender inequality, gender roles,
*Direct correspondence to John Mirowsky, De- or gender-based exposure to social stressors
partment of Sociology, Ohio State University, explain women's elevated distress as the con-
300 Bricker Hall, 190 North Oval Mall, Colum- sequence of inequality and disadvantage
bus, OH 43210-1353 (Internet:MIROWSKY.1 @ (Gove and Tudor 1977; Pearlin 1989; Ross
OSU.EDU). We are indebted to the National Science and Huber 1985). According to the struc-
Foundation for the grant (SES-8916154) to tured-strain view, different positions in the
Catherine Ross that supported the "Work, Fam- social structure expose individuals to differ-
ily, and Well-Being" data collection. Sampling,
pretesting, and interviewing were conducted by ent amounts of hardship and constraint.
the Survey Research Laboratoryof the University Women's positions at work and in the family
of Illinois. An earlier version of this paper was disadvantage them compared to men because
presented at the 1994 meeting of the American of their greater burden of demands and limi-
Sociological Association in Los Angeles. tations. This burden creates stress and frus-

American Sociological Review, 1995, Vol. 60 (June:449-468) 449


tration and is manifested in higher levels of Gove 1978; Gurin, Veroff, and Feld 1960;
distress. Wood, Rhodes, and Whelan 1989). It seems
Two alternativeviews say that women sim- inconsistent that women would be more dis-
ply express themselves differently than men, tressed than men but equally happy or more
thus creating a false impression of greater happy. Second, women report more physi-
distress (Nolen-Hoeksema 1987). According ological symptoms than men as well as more
to the response-bias view, women express all psychological ones, despite the fact that
emotions more freely than men. According to women live longer than men. In fact, com-
this view, women are more aware than men pared to men, women report more physical
of their emotions, they are also more likely symptoms, more acute conditions, more non-
to talk about emotions to others, to be open fatal chronic diseases, more physician visits,
and expressive, and to think that discussing and more hospital stays, but men have higher
personal well-being is acceptable ratherthan rates of most fatal chronic diseases and
stigmatizing. Thus, when women and men higher mortality rates (Verbrugge 1986). To
are questioned about depression and anxiety many researchers,these facts say that women
the women report it more frequently. Alter- genuinely suffer more of the nonfatal disor-
natively, the gendered-response theory says ders caused or exacerbated by stress, where-
women respond to the ubiquitous stress of as men suffer more of the damaging and
life with somewhat different emotions than deadly disorders caused or exacerbated by
men. In particular, women might feel anx- risky and abusive behavior (Verbrugge 1985,
ious and depressed where men might feel 1986). However, others read these same facts
agitated and angry. If surveys ask more ques- as saying that the sex ratio of disease inverts
tions about types of distress typical of as the mode of assessment becomes more
women than about those typical of men, then objective and less susceptible to response
women may falsely appear more distressed. bias-going from symptom reports to diag-
Below we summarize the arguments and re- noses to death certificates. The inconsistency
search concerning the response-bias and between the rates of symptoms or complaints
gendered-response alternatives to the view and the rates of fatal disease leads some re-
that structured strain accounts for women's searchersto view women's greaterdistress as
greater distress. partially or wholly response bias (Ritchey,
LaGory, and Mullis 1991, 1993).
The fact that, compared to men, women re-
Expressiveness and Response Bias
port as much happiness and live longer sug-
According to the response-bias hypothesis, gests that response tendencies may account
men and women differ in their likelihood of for higher distress reported by women. To
expressing emotions or feelings to an inter- determine whether a specific response ten-
viewer (Phillips and Segal 1969; Ritchey, dency actually accounts for women's higher
LaGory, and Mullis 1993; Seiler 1975). His- levels of distress, research must answer three
torically, women have specialized in emo- empirical questions (Clancy and Gove 1974;
tionally supportive and expressive activities Ross and Mirowsky 1984). First, does the re-
and men in competitive ones. To women, sponse tendency significantly increase re-
emotions carry the signals between con- ports of distress? If not, the tendency cannot
nected beings; to men they reveal weak- account for any increases in reported dis-
nesses. Thus women express emotions more tress. Second, do women more than men tend
freely and men keep emotions more hidden, to respond that way? If not, then the ten-
perhaps even from themselves. The response- dency cannot account for differences be-
bias hypothesis claims that women's greater tween women and men. And third, does ad-
apparent distress flows largely or strictly justing for the response tendency substan-
from differences in reporting and not from tially decrease the estimated effect of sex on
differences in experience. distress? If not, then response bias accounts
Proponents of this view cite two kinds of for only a small part of the effect of sex on
evidence to support it. First, women do not reporteddistress.
report less happiness than men, and some- Only a few previous studies have tested re-
times they report more (Bradburn 1969; sponse tendencies in all three ways. How-

ever, intuitively none of the tendencies exam- According to the gendered-response hy-
ined seem like clear measures of expressive- pothesis, women may appearmore distressed
ness. Previous studies have tested three re- than men simply because the standard in-
sponse tendencies: yea-saying or acquies- dexes of distress ask more questions about
cence, the tendency to give socially desirable depression and anxiety than about anger and
responses, and the perceived social undesir- hostility. Men and women may experience
ability of symptoms of psychophysiological similar levels of frustration,but men get an-
distress (Clancy and Gove 1974; Ross and gry at others and women get upset with
Mirowsky 1984). Yea-saying does not affect themselves. Although the reasoning behind
reports of distress. Even if it did, women the gendered-response theory is clear and
yea-say less often than men do. The tendency plausible, to date research results do not
to give socially desirable responses decreases strongly support it. One clinical study found
reports of distress, but women show this ten- surprisingly high levels of anger and hostil-
dency as much as men. Finally, men do not ity among depressed female patients, but did
rate the symptoms as more undesirable than not systematically compare these levels to
women rate them. In the end, the effect of those in male patients (Weissman and Paykel
sex on reports of psychophysiological dis- 1974). Another clinical study that followed
tress remains significant after adjusting for 180 women and 50 men through treatment
yea-saying, the tendency to give socially de- for recurrent depression found that the
sirable responses, and the perceived unde- women reportedsignificantly more anger and
sirability of the symptoms. These results hostility than did the men at baseline (Frank,
clearly fail to supportresponse bias as an ex- Carpenter, and Kupfer 1988). After eight
planation of women's greater reported dis- weeks of treatment this difference disap-
tress. However, being emotionally expressive peared.Although this seems to contradict the
is not the same thing as yea-saying or giving gendered-response hypothesis, it could rep-
a socially desirable response, and women resent a selection effect. If women are not
might be more willing to report symptoms supposed to get angry, those who do may
than men even though agreeing with men on wind up in treatment because it is a deviant
the undesirability of those symptoms. response (Rosenfield 1982). However, a
community survey of 2,248 U.S. adults
found that the women reported significantly
Types of Distress and "Gendered
more "manifest irritation"(Gove 1978), and
Responses" a survey of 451 married couples in the rural
According to the gendered-response hypoth- Midwest found that wives reported signifi-
esis, men and women differ in the nature of cantly greater marital hostility than did their
their emotional responses to stress (Dohren- husbands (Conger, Lorenz, Elder, Simons,
wend and Dohrenwend 1976, 1977): Men get and Ge 1993).
angry and hostile-women get sad and de-
pressed. Sex differences in type of response
Testing Response-Biases and Gendered-
emerge when sex-role socialization channels
the psychodynamics of depression (see the Responses
review by Rosenfield 1980). According to Female expressiveness may mask the distinc-
this thesis, depression emerges from the frus- tion between feminine and masculine types
trationof desires and aspirations. Frustration of distress. Women may reportgreaterhostil-
generates rage and hostility. Depression re- ity and anger than men simply because
sults when a person turns that anger inward, women more freely report their emotions.
punishing the self for failure and inadequacy. Thus, the failure to find clear evidence of
Men are socialized for competitive and com- gendered responses may be seen as evidence
bative roles that allow, and even encourage, of response bias. While this may seem like a
the outward expression of anger and hostil- reasonableinterpretation,it reveals a possible
ity. Women are socialized for nurturingand circularity in the arguments against genuine
supportiveroles that discourage such expres- sex differences in distress: If men do report
sions. Thus, frustrationmakes men angry and more anger than women, then women's
makes women depressed. greater depression and anxiety is dismissed

as the "feminine response" to burdens and procedure, which ensures the inclusion of
frustrations shared equally with men (gen- unlisted numbers (Waksberg 1978). The in-
dered response); if men do not report more terviewers called primarily during evenings
anger than women do, then men's lower dis- and weekends. Unanswered working num-
tress is dismissed as "masculine stoicism" bers were called back 10 times before being
and reserve about burdens and frustrations dropped. Non-household numbers were
shared equally with women (response bias). dropped (businesses, dorms, etc.). In each
Any true tests of the response-bias and household the adult (18 years old or older)
gendered-response hypotheses must avoid with the most recent birthday was selected
this circularity. It must be possible to accept as respondent. (This is an efficient way to
or reject each hypothesis independently. randomly select a respondent within the
Independenttests requireexplicit measures household [O'Rourke and Blair 1983]). Up
and adjustments for expressiveness, and the to 10 callbacks were made to selected re-
examination of various forms of distress. We spondents who could not be interviewed im-
measure expressiveness directly by asking mediately. Of the selected respondents, 82.3
people whether they keep their emotions to percent completed interviews, yielding a to-
themselves, and indirectly by assessing the tal of 2,031 respondents (1,282 females and
tendency to reportboth positive and negative 749 males), ranging in age from 18 to 90.'
emotions (an unobtrusive latent factor). We
examine various outcomes, including de- I The method produces a sample with charac-
pressed mood (sadness), positive mood (hap-
teristics broadly representative of the U.S. popu-
piness), anxiety, anger, malaise, and physi-
lation as a whole. For example, the mean house-
ological symptoms (aches). hold income in the U.S. in 1990 was $37,922
Accepting the response-bias perspective compared to $38,632 in the sample. Among the
requires that we find support for all three of adults age 25 and older the median education is
the following hypotheses: 12.4 years in the U.S. compared to 12.0 in the
sample. In the U.S., 11.3 percent of householders
HI: Women are more likely than men to ex- are Black and 6.6 percent Hispanic, compared to
press their feelings. 10.1 percent and 5.4 percent of the sample respec-
H2: Expressiveness increases reports of dis- tively. Sixty percent of U.S. adults are married,
tress. compared to 60.6 percent of the sample. The per-
centage employed in the population and sample,
H3: Adjustment for the tendency to express respectively, are 71.9 percent and 74.0 percent for
emotions explains sex differences in dis- men and 54.3 percent and 57.1 percent for
tress (the effect of sex on distress be- women. Of the adults aged 20 and older, the age
breakdowns of the population and of the sample
comes insignificant with adjustment for
respectively are 22.9 percent and 21.5 percent in
response tendencies). their twenties, 23.6 percent and 25.6 percent in
Accepting the gendered-response perspec- their thirties, 17.7 and 18.9 percent in their for-
tive requires that we find support for both of ties, 12.3 percent and 12.2 percent in their fifties,
the following hypotheses: and 23.6 percent and 21.3 percent in their sixties
or older (statistics from United States Bureau of
H4:Women have higher levels of depression; the Census 1993).
men have higher levels of anger. Although the sample approximates the demo-
graphic profile of the population in many ways, it
H5: People with higher levels of anger have contains a lower fraction of males. Males com-
lower levels of depression. prise 47.8 percent of the U.S. population age 18
and over but are only 36.2 percent of the sample.
Research suggests that most of this difference
METHODS probably arises because women are more likely
to answer the telephone than their male partners,
Sample and the person who answers sometimes gatekeeps
for the person selected (O'Rourke and Lakner
This research draws on a fall 1990 telephone 1989). A comparison of the sex ratios by marital
survey of a national probability sample of status shows the effect. For the married popula-
U.S. households. Sampling followed the tion age 18 and over the ratio of females to males
Waksberg-Mitofsky random-digit dialing is 1.02. The sample's corresponding ratio is 1.75.

The mean age is 43.5, with 23.4 percent of and unable to shake the blues (ax= .82). Hap-
the sample under age 30 and 21.2 percent piness averages the frequency of feeling
age 60 or older. Of the total sample, 60.6 happy, feeling hopeful about the future, and
percent are marriedand 17.5 percent are mi- enjoying life (ax= .79). Anxiety averages the
norities (including 10.1 percent Black and frequency of worrying a lot about little
5.4 percent Hispanic). There is one adult in things, feeling tense or anxious, and feeling
24.9 percent of the sample's households, two restless (ax = .82). Anger averages the fre-
in 59.5 percent, three in 10.6 percent, and quency of feeling annoyed with things or
four or more in 5.0 percent. The mean people, feeling angry, and yelling at some-
household income is $38,632 ($44,080 for one (a = .71). Malaise averages the fre-
married persons and $30,267 for others; quency of feeling everything is an effort,
$33,512 for minorities and $39,737 others). feeling that you just can't get going, having
The mean education is 13.2 years, with 27.9 trouble keeping your mind on what you are
percent holding a bachelors degree or higher doing, and having trouble getting to sleep or
and with 13.3 percent having less than a staying asleep (a = .73). Aches averages the
high school degree. frequency of having aches and pains, having
headaches, and feeling weak all over (a =
Although some researchers invoke emo-
Distress is measured using six indexes rep- tional expressiveness as an explanation of
resenting sadness, happiness, anger, anxiety, sex differences in measured distress, we
malaise, and aches.2 For each symptom of know of none who has measured expressive-
distress, respondents were asked, "On how ness and shown that sex differences in dis-
many of the past 7 days have you ... ?" Re- tress vanish when controlling for it. Part of
sponses are coded from 0 to 7, from never the reason no one has done this may be that
experiencing the symptom to experiencing it differences in the willingness to express
every day. Each index represents the mean emotions are easier to imagine than to mea-
frequency of its component items. Sadness sure. Our analyses below take two ap-
averages the frequency of feeling sad, lonely, proaches to assessing possible response bias:
one self-evident and the other unobtrusive.
The first measure is a self-evident ques-
(The populationand sample sex ratios [FI/M"]
are .80 and .989 amongthe never married,1.40 tion about how much the respondent agrees
and 1.86 among the divorcedor separated,and with the statement, "I keep my emotions to
5.00 and4.88 amongthe widowed). myself." The responses indicate self-evalu-
This underrepresentationof males raises a ated emotional reserve (called "reserve" for
question:Does the sex differencein distressob- short). The response categories are
served in samples arise becausedistressedmen "strongly disagree" (coded -2), "disagree"
escape samplingmore often than do distressed (coded -1), "agree"(coded 1), and "strongly
women?A survey of marriedcouples indicates
thatthe answeris no (Ross,Mirowsky,andHuber
1983; Ross and Mirowsky 1984). The survey 1977). Items in these threeindexes load on ex-
used the same procedureas we use,in this study ploratoryfactors representingdepressedaffect,
to select andinterviewa sampleof 680 U.S. mar- positiveaffect,andenervation,respectively(Ross
ried couples. Comparedto their own husbands, and Mirowsky1984). The items in each index
thewives reportedsignificantlymorefrequentde- correlatehighly with the factorbeing measured,
pressed affect and enervationand significantly but only moderatelywith the other exploratory
less frequentpositive affect (CES-Dsubscales). factors,andshow the samepatternof factorcor-
This sample of couples avoids the gatekeeping relationsfor men and women.Itemsin the anxi-
effect because it containsonly marriedpersons ety andachesindexescome fromthe psychologi-
whose spouses also completedinterviews.Thus cal andphysiologicalcomponentsof the Langner
the observedsex differencein distresscannotre- index,respectively(Langner1962;Mirowskyand
sult froma sex differencein the effect of distress Ross 1983). Itemsin the angerindex are similar
on the probabilityof beingsampled. to thosein Congeret al.'s (1993) indexof hostil-
2 Itemsin the sadness,happiness,and malaise ity betweenmarriagepartners,except they were
indexescome fromthe Centerfor Epidemiologi- modifiedto reflect angerwith anyone (not just
cal Studies DepressionScale (CES-D) (Radloff with one's spouse).

agree" (coded 2). (Six women and 2 men Female is a dummy variable representing
said "don't know" and were coded 0). This sex, coded 1 for females and 0 for males.
measure is simple and direct. However, it Sociodemographic variables are adjusted
has several disadvantages. Someone who is in some of the models. Age is coded in num-
circumspect about emotions may be disin- ber of years. Minority status is coded 0 for
clined to say so. Or people may be unaware non-Hispanic whites and 1 for others. Mari-
of their own emotional expressiveness com- tal status is coded 1 for marriedor living to-
pared to others'. On the other hand, people gether as marriedand 0 otherwise. Education
who are wary of sharing emotions with is coded as the number of completed years
friends and acquaintances may be relieved of formal schooling.
to report them as an anonymous respondent.
Perhaps more to the point, the conscious re-
straint of one's emotions may indicate a
heightened level of disturbing emotion, a The analysis has three parts. The first part
dysfunctional coping strategy, or a sense of describes sex differences in the degree of
needing to tell someone. emotional expression and tests the response-
The second measure is an unobtrusive la- bias hypothesis. The second part describes
tentfactor implicit in the reports of positive sex differences in the type of emotional ex-
and negative moods (happiness and sadness). pression and tests the gendered-response hy-
In this crossed 2 x 2 measurement, people pothesis. The third part analyzes the impact
who report more of both emotions are con- of sex differences in the degree and type of
sidered more emotionally expressive. We call emotional expression on regression esti-
the factor "expression"for short. The model mates of the sex difference in distress. The
defines expression and depression as cross- introduction to the results in each part ex-
cutting factors, each indicated by reports plains the reasoning behind its models and
both of sadness and of happiness: Expression analyses.
increases reportsof happiness and of sadness
net of the level of depression, whereas de-
Analysis 1: Response Bias and Female
pression increases reports of sadness and de-
creases reports of happiness net of the level Expressiveness
of expression. (For details and justification Analysis 1 explores the possibility that
of the 2 x 2 measurementdesign, see Figure women seem more depressed than men be-
2 below.) cause women report their emotions more
Crossed and balanced measurement was freely.
designed originally to deal with the problem Self-evident appraisal of expressiveness.
of agreement bias in indexes of perceived The statement, "I keep my emotions to my-
control (Mirowsky and Ross 1991). Such in- self," voices emotional reserve-the self-per-
dexes generally ask people whether they ceived restraint of emotional expression.
agree (or disagree) with statements that ex- Cross-tabulatingthe responses by sex shows
press either instrumental beliefs or fatalistic that more men than women claim to keep
ones. The tendency to agree (or to disagree) emotions to themselves. Sixty-eight percent
with statements regardless of their content of the men either agree or strongly agree
crosscuts the actual level of perceived con- with the statement, compared to 50 percent
trol and needs to be factored out. Agreement of the women. Thus, in our sample men are
bias is not a problem in most measures of 1.360 times more likely than women to say
distress, which usually ask people to report they keep emotions to themselves. (Alterna-
the frequency of symptoms rather than ask- tively, the odds of agreeing with the state-
ing for agreement or disagreement with ment are 2.125 times greaterfor men than for
statements implying distress. However, the women.) The men's average score exceeds
idea of a crosscutting factor and the method the women's by about a third of a standard
of crossed and balanced measurement seem deviation. Both the chi-square test of the
appropriate for distinguishing the level of cross-tabulation and the t-test of the differ-
depression from the degree of emotional ex- ence in mean scores are statistically signifi-
pression. cant at p < .001.

2.0 men, but also that they are more depressed

IMen than men. The main difference in results be-
tween our two measures is that the unobtru-
01.5 Women
U sive factor relates to measures of distress
more as one would expect for an index of
C 1.0 1 emotional expressiveness. With the direct
question about emotional reserve, people
C who claim to be the most reserved actually
report more days of sadness than others, not
fewer. By definition, people with low scores
.0 on the expression factor reportfewer days of
Strongly Disagree Agree Strongly sadness, and fewer of happiness.
Disagree Agree
"I keep my emotions to myself."
The basic model shows the logic behind
the crosscutting factor model in its simplest
Figure 1. Relationship between Sadness Index and clearest form. The basic model, shown
Score and Emotional Reserve" in Table 1 and Figure 2, hypothesizes that
a Sadness increases with emotional reserve. scores on the sadness index equal the sum of
Women are sadderthan men in each category. the depression and expression factors,
whereas scores on the happiness index equal
the difference. In the terms of structural
Does male reserve account for the fact that equation modeling, the model fixes the met-
men report being depressed less often than ric loadings of the sadness index to +1 on
women? No. We find that people who claim both the depression and expression factors,
to keep emotions to themselves report more whereas it fixes the metric loadings of the
days of depressed mood, not fewer. The sad- happiness index to -1 on depression and +1
ness index from the CES-D averages the on expression. The basic model treats the
number of days from the previous week the sadness and happiness indexes as exact lin-
respondent reported feeling sad, lonely and ear composites of the two crosscutting fac-
blue. Figure 1 shows that mean scores for tors, which overstates the precision of the in-
sadness increase with self-reported emo- dexes. Despite the simplification, the basic
tional reserve (r= .122, p < .001; 71= .144).3 model produces results consistent with those
People who strongly agree that "I keep my of a detailed model that follows.
emotions to myself" average more than The results are consistent with the hypoth-
twice the frequency of sadness than people esized crosscutting factors in that the model
who disagree. Women are sadder than men fits the observed variances and covariances
within each category of emotional reserve. well and shows significant variance in each
This simple analysis suggests that male re- factor. The metric effect of being female on
serve cannot account for sex differences in expressiveness is .170 (row 1, column 3 of
depression. In fact, if the only difference be- Table 1). The metric effect of being female
tween the sexes were that men had greater on depression is .182 (row 1, column 2 of
emotional reserve, men would be sadderthan Table 1). According to the model, women
women because people who say they keep feel more depressed than men, and they ex-
their emotions to themselves report more press their feelings more freely.
sadness than others, not less. Both the depression and expression factors
Unobtrusive appraisal of expressiveness. increase women's scores on the sadness in-
The crosscutting factor model indicates that dex, but their influence on the happiness in-
women express emotions more freely than do dex tends to cancel out. In these data women
reportabout the same frequency of happiness
3 The r and r7 coefficients of correlation are
as men. (Women actually report slightly
.130 and .141 respectively for men and .142 and
.170 respectively for women. The deviation from
more happiness than men, but the difference
linearity is significant for women (p = .003) but is not statistically significant [p = .422]).
not for men (p = .339). The covariance structure How can women be sadder than men but not
models and regression analyses use only the lin- less happy? The model says that the appar-
ear component of the correlation. ent inconsistency reflects the presence of the

Table 1. Metric Coefficients from the Basic Crosscutting Factor Modela Testing Sex Differences in
Depression and Emotional Expressiveness: U.S. Men and Women Ages 18 to 90, 1990

(1) (2) (3)
"I keep my Depression Expression
IndependentVariables emotions to myself." Factorb Factorb
(1) Female -.442*** .182** .170***
(-8.531) (3.095) (4.729)
(2) "I keep my .165*** .OOOC
emotions to myself." (6.934)
*p <.05 **p< .01 ***p< .001 (two-tailedtests)
a EQS Model (Bentler 1989): N = 2,027; %2 = .001; d.f. = 1;p = .978; normedfit = 1.000; nonnormedfit
= 1.009; comparativefit = 1.000.
b The depressionand expressionfactors are both indicatedby a three-itemindex of sadness and a three-
item index of happiness.The metricloadings of the indexes on the depressionfactor are fixed to +1 and -1
respectively. The metric loadings on the expressionfactor are both fixed to +1. The regressionresidualsof
the depressionand expressionfactorshave a covarianceof -.284 (t = -12.539).
c The parameteris fixed to zero. When freed it is not significantlydifferentfrom zero.

Note: Numbersin parenthesesare t-values.

crosscutting expression factor. Depression Interestingly, the basic crosscutting factor

shifts women's balance of emotions away model shows no direct effect of self-evalu-
from happiness toward sadness, but greater ated reserve on the expression of positive and
emotional expressiveness means they report negative emotions. As in the simple analysis
more happiness than men do at the same
level of depression.4 tions into one group and the positive emotions
into another. The correlation of the indexes based
4 The presence of a crosscutting expression fac- on the two groups has one negative component
tor can explain why positive and negative emo- (because depression increases negative emotions
tions load on separate factors in exploratory while decreasing positive ones) and one positive
analyses (Ross and Mirowsky 1984) and why component (because expression increases both
separate indexes of positive and negative emo- types of emotions), which sum to a correlation
tions often have little or no correlation with each near zero.
other (Bradburn 1969). Crosscutting factors cre- The low correlation between indexes of nega-
ate larger correlations between items that share tive and positive affect leads some researchers to
the same pattern of loadings on both factors than view them as orthogonal traits-as if sadness has
between ones with different patterns of loadings. no bearing on happiness and vice versa. The alter-
For example, feeling sad and feeling blue corre- native is to view depression and expression as
late highly because depression and expression in- crosscutting traits, both of which influence the re-
crease both. Likewise, feeling happy and feeling ported frequency of positive and negative emo-
hopeful about the future correlate highly because tions. To some extent this is a conceptual choice,
depression decreases both and expression in- similar to choosing between rotated and unrotated
creases both. In contrast, feeling sad and feeling factors. We believe it is best to think in terms of
happy have a smaller correlation (in absolute crosscutting factors. Treating sadness and happi-
terms) because depression increases sadness ness as unrelatedimplicitly denies the existence of
while decreasing happiness, whereas expression a dimension of depression that shifts the balance
increases both. The correlation between feeling of affect from positive to negative emotions. That
sad and feeling happy has one negative compo- in turn implicitly denies the impact of real condi-
nent (because they representopposite poles of the tions (such as poverty) on depression. In contrast,
depression continuum) and one positive compo- viewing depression and expression as crosscutting
nent (due to expression), which sum to a correla- factors allows us to assess the causes and conse-
tion near zero. Exploratory factor analyses sepa- quences of one factor without interference from
rate items into groups defined by higher correla- the other. In the present case, this means we can
tions within the groups than between them. Thus, assess the sex differences in distress with the dif-
exploratory factor analyses put the negative emo- ferences in expression filtered out.

Z5 U

"Ikeep my
to myself."

U Index
U 996
o ~~~~~~~~~Happiness


Figure 2. Basic Crosscutting Factor Modela

a The basic crosscuttingfactor model shows that women are more expressive than men but also are more

depressed(standardizedcoefficients). Table 1 shows the fit statistics and metric coefficients.

Note: All coefficients are significantat p < .05 or better.

of Figure 1, depression increases with emo- male" on "blue" net of the "depression" and
tional reserve (b = .165, row 2, column 2 of "expression" factors). Only this one of the
Table 1). Again, our results imply that men six items shows signs of sex bias in report-
would report more depression than would ing. The rest co-vary with sex in proportion
women if the sexes differed only in emo- to their factor loadings. Those covariances
tional reserve. only differ for men and women by small, ran-
A detailed version of the crosscutting fac- dom amounts, showing that, for the most
tor model reiterates that women are more ex- part, men and women describe their depres-
pressive than men, but that they are also sion in comparable terms.
more depressed. Figure 3 and Tables 2a and An alternativemodel ignores the direct ef-
2b show the detailed model. The model ad- fect of sex on reports of feeling blue by fix-
justs for random error and nonrandom bias ing that parameter to zero. The alternative
in the reporting of individual symptoms, and does not fit as well (AX2= 4.551, Ad.f. = 1,
for possible spuriousness due to marital sta- p < .05), but the estimated metric effect of
tus, age, and education. The detailed model sex on depression drops by only 11.9 percent
is truerthan the basic one in the sense that it (from .185 to .163), and it remains quite sig-
allows for error, bias, and spuriousness. The nificant (the t-value goes from 2.994 to
core results remain essentially the same. 2.679). The robustness of the estimate re-
Women feel depressed more often than men. flects three facts. First, sex affects only one
Women are more expressive and less re- of the six items, feeling blue, net of the de-
served about emotions than men, but these pression and expression factors. Second, far
differences do not account for the sex differ- more of the variance in reports of feeling
ence in depression. blue is attributableto the mediating factors
The detailed model also indicates that than to the direct effect of sex net of those
women report feeling blue less often than factors. Third, even among women, feeling
men who are equally depressed and expres- blue co-varies substantially with the other
sive (note the direct negative effect of "fe- symptoms.

Table 2a. Metric Coefficients from the Detailed Crosscutting Factor Modela Testing Sex Differences
in Depression and Emotional Expressiveness: U.S. Men and Women Ages 18 to 90, 1990
(1) (2) (3)
"I keep my Depression Expression
IndependentVariables emotions to myself." Factorb Factorb

(1) Female -.456*** .185** .148***

(-8.860) (2.994) (4.274)

(2) Marriage -.235*** -.389*** .OOOC

(-4.761) (-6.537)
(3) Age .005*** -.004** .002
(3.682) (-2.435) (1.786)

(4) Education -.024* -.050*** -.014*

(-2.422) (-4.375) (-2.096)

(5) "I keep my emotions .151*** .OOOC

to myself." (5.838)

*p <.05 **p<.01 ***p< .001 (two-tailedtests)

a EQS Model (Bentler 1989): N = 2,019; x2 = 30.204; d.f. = 28; p = .354; normedfit = .994; nonnormed
fit = .999; comparativefit = 1.000. This model meets the criteriathat (a) all of the coefficients have a two-
tailed p-value of .10 or less and that (b) a Lagrangemultipliertest indicates the overall fit cannot be im-
provedby freeing any of the parametersfixed to 0, +1, or -1.
b The depressionand expressionfactors areboth indicatedby threereportsof sadness and threereportsof
happiness.Table 2b details the measurementmodel.
c The parameteris fixed to zero. When freed it is not significantlydifferentfrom zero.
Note: Numbersin parenthesesare t-values.

Table 2b. Metric Factor Loadings for the Detailed Crosscutting Factor Model of Sex Differences in
Depression and Emotional Expressiveness: U.S. Men and Women Ages 18 to 19, 1990
IndependentVariables Blue Lonely Sad Happy Enjoy Life Hopeful

(1) Depression 943*** 1.OOOa 1.oooa -1.134*** -1.oooa -1.o00a

factor (38.517) (-37.825)
(-2.328) (-4.470)
(2) Expression 1.OO0a 1.OO0a 1.300 1.OOa 1.oooa 1.475
factor (16.708) (14.481)
(3.856) (4.663)
(3) Female -.122
(4) Marriage -.306
(5) Age .008
(6) Education .051

*p <.05 p < .01 p < .001 (two-tailedtests)

a Coefficients withoutt-values are fixed to the values shown. Blanks representcoefficients fixed to zero.
Note: A t-value in parentheses(t) measuresthe coefficient's standard-error distancefrom 0; A t-value in
brackets It) measuresits standard-error distancefrom the theoreticalmetricloading of +1 or -1.

/ / my*emotions
OR;>\ \
L0 ~tomyef".615
a Womenar or xpesvetanmnuls or eresdsanadze oefcBlue *-be

.61 621

9 ~~~~~~~~~~~~~~~Lo
Mote:Arrcoeffiagen DEPRESSION .6 f


I 0
Happy.4- e

Enjoy * e

Hope -e

Education?-------------- -PO

Figure 3. Detailed Crosscutting FactorModela

a Women are more expressive than men but also more depressed (standardizedcoefficients). Table 2

gives the fit statistics and metriccoefficients.

Note: All coefficients are significantatp < .05 or better,except for the directeffect of age on the expres-
sion factor. Dashed lines representdirecteffects of exogenous variableson indicatorsnet of their factors.

Analysis 2: "Masculine" and "Feminine" or people, feeling angry, and yelling at

Types of Distress someone. These indicators form a progres-
sion of escalating anger. The pattern of co-
Surveys might give a false impression by variances among the indicators and between
asking mostly about feminine types of dis- them and other variables mixes aspects of a
tress. The analyses up to this point have con- latent factor with those of a Guttman scale.
centrated on feelings of depression. By that Loadings on the anger factor represent the
measure women exhibit more distress than former, and the sequential direct effects rep-
men. Does the relationship hold for more resent the latter. Yelling at someone is the
"masculine" forms of distress? The escalat- highest expression of anger on the scale, be-
ing anger model and the hierarchical distress cause it crosses the line from emotion to be-
model address this question. havior. The results show that women are an-
Sex and escalating anger. Do women get grier than men and are more likely to ex-
depressed and men get angry? Gendered-re- press their anger by yelling.
sponse theory suggests that the answer is Anger has another thing in common with
yes, but our data indicate that the answer is depression, in additionto showing higher lev-
no. Figure 4 and Table 3 show the escalating els among women than among men: People
anger model. Three reports indicate the who say they keep emotions to themselves
level of anger: feeling annoyed with things report feeling angry more frequently than


Female "Ikeep !
my emotions
/ / \ ~~~~~~~to
myself." _ Yelled e

(0 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~8


a Angry w e


ANGER - 0* Annoyed *-e

Education ' U
Figure 4. Escalating Anger Modela
a Women are angrierthanmen and yell more thanmen who are equally angry(standardizedcoefficients).
Table 3 gives the fit statistics and metric coefficients.
Note: All coefficients are significant at p < .05 or better. Dashed lines representdirect effects of exog-
enous variableson indicatorsnet of theirfactors.

others-not less. Anger also shows some pat- A simple but importantresult underlies the
terns that are distinct from those of depres- hierarchical factor model: Various measures
sion. In particular,age has a substantialnega- of distress are positively correlatedwith each
tive effect on anger (/3 = -.544), and neither other. Depression, anxiety, anger, and physi-
marriagenor education have direct effects on cal symptoms correlate positively and sig-
it (although marriageincreases the likelihood nificantly. Positive mood correlates nega-
of escalating to yelling and education de- tively with all psychophysiological measures
creases the likelihood). With respect to sex, of distress. Thus, people who feel sad, for
though, anger and depression can be viewed example, also tend to feel unhappy,anxious,
as interchangeable measures of distress. angry, depleted, and ill.
Sex and the hierarchy of distress. A con-
firmatoryhierarchical factor model using six
Analysis 3: Adjusted Regressions
indexes shows that, at the most general level,
women feel distressed more often than do Response bias and gendered response. Re-
men. They experience distress more fre- sponse-bias theory implies that adjusting for
quently than men regardless of the form or emotional reserve and expression will elimi-
level of generality. Figure 5 illustrates the hi- nate or greatly reduce the estimated effect of
erarchical factor model. Distress may take sex on levels of distress. Table 4 shows six
either emotional or physical form. Emotional sets of regressions. The first in each set re-
distress appears as depression, indicated by gresses an index of distress on sex adjusting
sadness and lack of happiness, or as agita- for age, minority status, marital status, and
tion, indicated by anger and anxiety. Physi- education. The second adjusts also for the
cal distress appears as malaise and aches. expression index (Expression = (Sadness +

Table 3. Metric Regression Coefficients and Factor Loadings for the Escalating-Factor Modela Test-
ing Sex Differences in Anger: U.S. Men and Women Ages 18 to 90

IndependentVariables Anger Factor Annoyed Angry Yelled
Female .146** .534***
(2.844) (5.945)
Marriage .358***
Age -.021
Education -.070***
"I keep emotions .050**
to myself." (2.513)
Anger factor 1.817*** 1.Ob 1.O0b
Annoyed .305***
Angry 374***
*p<.05 * < .01 p < .001 (two-tailedtests)
a EQS Model (Bentler 1989): N = 2,031; X2= 11.557; d.f. = 10; p = .316; normedfit = .994; nonnormed
fit = .998; comparativefit = .999. This model meets the criteriathat (a) all of the coefficients have two-
tailed p-values of .10 or less and that (b) a Lagrangemultipliertest indicates the overall fit cannot be
improvedby freeing any of the parametersfixed to 0, +1, or -1. The table does not show the regressionof
keeping emotions to oneself because it is the same as in the first column in Table 2a.
b Coefficients without t-values are fixed to the values shown. Blanks
representcoefficients fixed to zero.
Note: A t-value in parentheses(t) measuresthe coefficient's standard-error distancefrom 0; A t-value in
brackets( t} measuresits standard-error distancefrom the theoreticalmetric loading of + 1.

Happiness)12), and the third adds adjustment comes more negative and more statistically
for keeping emotions to oneself. The first significant with the added adjustment for
row of Table 4 shows the metric effect of be- keeping emotions to oneself. Overall, the re-
ing female on each type of distress, first gressions show that, adjustingfor expressive-
without and then with the adjustmentsfor ex- ness, women are more distressed than men
pressiveness and reserve. For the five nega- on all six measures, and that adjusting for re-
tive emotions the effect of sex gets smaller serve increases the estimated differences.
with adjustmentfor the expression index, but Gendered-responsetheory has implied that
it gets larger with adjustment for emotional women's distress would most exceed men's
reserve. Only the effect of being female on when measured as sadness and unhappiness
sadness is smaller with the two adjustments and least when measured as anger. Our re-
than without them, but still it remains quite gressions show just the opposite. Sex differ-
significant. For happiness, the effect of sex ences are smallest for sadness and happiness,
changes sign when adjustment for the ex- and the sex difference in anger is twice as
pression index is added, from positive but large as that for sadness and happiness. Sex
very nonsignificant to negative and nearly has its largest metric effect on anxiety. In
significant for a two-tailed test (or significant terms of days per symptom per week, the
for a one-tailed test; t = -1.937). The nega- gender gap in anxiety and anger far exceeds
tive effect of being female on happiness be- that in sadness and unhappiness.

4co Sadness


Happiness .494


An er


< X // -
05t ~ ~ ~ ~~X ~ Anxiety 09

my emotions /
to myself:' L'
e | , PHYSICAL ~_ - Malaise 0

/ ~~PHYSICAL________ ____

DISTRESS .4 Ahs .7
Female--_-- ~ ~ ~ ~ i

Figure 5. Hierarchical Factor Modela EXPRESSION

a Women are more distressedthanmen by variousmeasures(standardizedcoefficients).
Note: All coefficients are significant at p < .05 or better.OverallX2= 8.929, d.f. = 9, p = .444. Dashed
lines representdirect effects of exogenous variableson indicatorsnet of theirfactors.

Relative frequency of distress. How large -.089, which means it explains only 0.8 per-
is the effect of sex on the frequency of dis- cent of the variance in income. However,
tress? Statistically, the metric coefficients in minority families get 15.7 percent less in-
Table 4 are quite significant. But do they come-an average of $6,224 less per year.
represent a major increment in the burden of That is a substantial sum of money even
distress for women as compared to men? though it is small compared to the standard
The standardized effect on distress of being deviation of $26,380 within groups.)
female shown in Figure 5 is .075, which A measure of relative frequency shows that
seems small. However, the size of the stan- women experience symptoms of distress
dardized coefficient is deceptive for two roughly 30 percent more often than men.
reasons. First, dichotomous variables tend to Taking the regression coefficient for sex as a
have small standardized coefficients for percentage of the mean symptom score
purely technical reasons: Standardizedcoef- among men shows roughly the percentage
ficients shrink as the number of possible increase in the frequency of symptoms for
values decreases, and two (male or female) women as compared to men. For example,
is the smallest number possible for a vari- Table 4 shows that the adjusted effect of be-
able. Second, large variances in distress ing female on the frequency of sadness is
among persons of the same sex diminishes .198. The mean frequency of sadness for men
the standardized effect of sex, even if the is .680 days per symptom per week.5 Thus,
average difference between men and women
is substantial. (As an example of the prin- 5 The intercepts in the following bivariate re-
ciple, in these data minority status has a gressions represent the crude means for males,
standardized effect on family income of and the coefficients of being female (F) represent

Table 4. Metric Coefficients for the Regression of Six Indexes of Distress on Sex and Sociodemo-
graphic Variables, without and with Adjustment for Expressiveness and Emotional Reserve:
U.S. Men and Women Ages 18 to 90, 1990

Independent Sadness Happiness Anger

Variables Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 Model 1 Model 2 Model 3

Female .234*** .127* .198*** .048 -.118+ -.188** .387*** .337*** .413***
(3.447) (2.076) (3.153) (.635) (-1.937) (-3.056) (5.251) (5.092) (5.468)

Age -.002 -.005** -.005** .009*** .005** .006*** -.033*** -.033*** .034***
(-.960) (-2.582) (-3.021) (4.196) (2.882) (3.339) (-15.547) (-15.603) (-15.759)

Minority .073 .022 .030 .069 -.010 -.018 .090 .086 .090
(.833) (.275) (.378) (.706) (-.132) (-.238) (.950) (.897) (.941)

Married -.527*** -.497*** -.458*** .446*** .492*** .452*** .004 .007 .027
(-7.776) (-8.154) (-7.516) (5.870) (8.141) (7.485) (.058) (.095) (.370)

Education -.067*** -.064*** -.060*** .059*** .064*** .061*** -.015 -.014 -.013
(-5.214) (-5.501) (-5.222) (4.076) (5.584) (5.298) (-1.049) (-1.025) (-.891)

Expression .785*** .784*** 1.220*** 1.221*** .073+ .073+

Index (21.788) (21.924) (34.157) (34.475) (1.687) (1.675)

"I keep my .149*** -.539*** .077*

emotions (5.682) (-5.885) (2.413)
to myself."

Intercept 1.960 -.518 -.623 4.328 .475 .582 3.000 2.769 2.715
R2 .053 .234 .246 .032 .388 .398 .117 .118 .121

Independent Anxiety Malaise Aches

Variables Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 Model 1 Model 2 Model 3

Female .584*** .546*** .626*** .294*** .273*** .327*** .298*** .284*** .302***
(5.650) (5.284) (5.963) (4.528) (4.206) (4.954) (4.823) (4.587) (4.798)

Age -.021 *** -.022*** -.002*** -.001 -.002 -.002 .005** .005** .005a*
(-6.951) (-7.283) (-7.584) (-.574) (-.859) (-1.174) (2.974) (2.775) (2.653)

Minority -.166 -.185 -.175 -.035 -.045 -.039 -.080 -.086 -.084
(-1.284) (-1.391) (-1.326) (-.413) (-.536) (-.464) (-1.001) (-1.082) (-1.055)

Married -.306** -.296** -.250* -.374*** -.341*** -.310*** -.280*** -.277*** -.266***
(-5.361) (-2.881) (-2.425) (-5.361) (-5.289) (-4.799) (-4.559) (-4.495) (-4.293)

Education -.007*** -.076*** -.072*** -.109*** -.109*** -.106*** -.094*** -.094*** -.093***
(-3.936) (-3.888) (-3.680) (-8.853) (-8.832) (-8.624) (-8.084) (-8.043) (-7.949)

Expression .280*** .279*** .155t** .154*** .099** .099**

Index (4.627) (4.621) (4.071) (4.063) (2.733) (2.725)

"I keep my .174*** .117*** .041

emotions (3.930) (4.191) (1.534)
to myself."

Intercept 4.164 3.279 3.156 2.588 2.028 2.016 2.072 1.758 1.729
R2 .045 .055 .062 .065 .072 .080 .065 .069 .070

*p < .05 **p< .01 ***p< .001 (2-tailed tests)

+p < .05 (one-tailed tests)
Note: Numbers in parentheses are t-values.

adjusting for emotional reserve and expres- (Cooperstock and Parnell 1975; Gove 1993;
siveness, women are sad about 29.1 percent Gove and Clancy 1975; Ritchey et al. 1993;
more frequently than men (100[.198/.680] = Seiler 1975). Most available evidence indi-
29.1). The outline below shows the adjusted cates that sex differences in distress are not
rate fractions (as subscripts) for the six in- due to reporting tendencies. In some cases
dexes in Table 4 and for the composite in- the response tendency is not correlated with
dexes that represent higher-orderfactors up distress, and in others it is not correlatedwith
to the level of general distress (see Figure 5). sex or is correlated in the wrong direction
(Clancy and Gove 1974; Gove and Geerken
1977; Gove, McCorkel, Fain, and Hughes
1976; Ross and Mirowsky 1984). However,
few studies examine all the links necessary
to determine whether response tendencies
Happiness-3.3% bias the observed association between sex
Agitation 29.5%
and distress. Furthermore, previous studies
Anger 28.7%
measure tendencies such as yea-saying and
giving socially desirable responses that do
not appearto representexpressiveness. Some
Malaise 37.6%
researchersstill maintainthat sex differences
in distress are due to reporting differences.
In proportional terms, the smallest differ- For instance, a recent study of the homeless
ence is for happiness, which is about 3.3 per- interpretedthe higher reports of physical and
cent less common for women than for men. psychological symptoms by the women as
(Put the other way, not feeling happy is about due to a "tendency for women to perceive
15.4 percent more common for women than and report symptoms more freely and for
for men). The largest difference is for mal- men to underreport"(Ritchey et al. 1991:45).
aise, which occurs 37.6 percent more fre- The response-bias hypothesis makes three
quently for women than for men. Women re- claims: (1) that women express their emo-
port feeling angry about 28.7 percent more tions more freely than men, (2) that expres-
frequently than men. Women's risk as a frac- siveness increases the reporting of symp-
tion of men's is greater for agitation than for toms, and (3) that adjusting for expressive-
depression, and it is greater for physical dis- ness eliminates the association between sex
tress than for emotional distress. At the high- and distress. Our results support the first
est level of generality, the symptoms of dis- proposition-that women are more expres-
tress we measure occur 29.4 percent more sive than men. Women score higher on both
frequently for women than for men. indicators of potential response bias: the un-
obtrusive sum of both positive and negative
mood and disagreement with the self-evident
DISCUSSION statement that, "I keep my emotions to my-
self." Our results support the second claim
Response Bias
for the unobtrusiveexpression factor but not
For two decades researchers have debated for the direct question about emotional re-
whether sex differences in reported psycho- serve. Scores on all six indexes increase with
physiological distress are real or an artifact the expression factor (although not quite sig-
of differences in the expression of emotions nificantly for anger), but they also increase
the more that respondentsclaim to keep emo-
tions to themselves. Our results contradict
the increment in the crude mean for females: Sad- the third proposition for both measures. The
ness = .680 + .238F; Happiness = 5.783 + .062F; estimated effect of sex on the measures of
Anger = 1.441 + .309F; Anxiety = 2.052 + .55OF;
negative mood remains quite significant af-
Malaise = .869 + .313F; Aches = .844 + .330F.
The bivariate coefficients of sex are significant at ter adjusting
for the expression of positive
p < .001 (two-tailed test) for the five negative and negative mood. And adjusting for self-
emotions. The unadjusted difference in Happi- reported reserve actually increases the esti-
ness scores is not significant (t = .802). mated effect of sex on all six measures.

Broadly speaking, the effect of sex on mea- and expression of one's anger reduces de-
sures of distress diminishes somewhat with pression. However, our results show that sad-
adjustment for emotional expression, but it ness, malaise, and the other indexes of dis-
does not vanish. tress all correlate positively with reported
Reserve and expressiveness are distinct anger, not negatively. The hierarchical factor
traits with distinct relationships to measures model shows that anger can be viewed as one
of distress, although women are more expres- manifestation of distress on a par with sad-
sive than men by both standards.As a mea- ness, unhappiness, anxiety, malaise and
sure, asking people if they keep emotions to aches. Second, the gendered-response view
themselves has the advantage of simplicity assumes that women feel less angry toward
and directness. However, people who say others than do men. On the contrary, our re-
they keep emotions to themselves actually sults show that women feel more angry to-
reportmore symptoms than others, not fewer. ward others. Thus, our results confirm those
The expression factor that crosscuts the de- of earlier studies showing that women report
pression factor behaves more as one would greater anger, hostility, and manifest irrita-
expect of a measure of emotional expressive- tion than do men (Conger et al. 1993; Frank,
ness. By definition the expression factor in- Carpenter, and Kupfer 1988; Gove 1978).
creases the reported frequency of both posi- Third, it assumes that women feel less angry
tive and negative mood. The expression fac- toward others than men at any given level of
tor also increases the reported frequency of distress. However, the hierarchical factor
anger, anxiety, malaise, and aches (although model shows that women are more agitated
the anger coefficient was not significant in than men experiencing similar levels of emo-
the hierarchical factor model). Interestingly, tional distress, which means that women are
the covariance structuremodels indicate that more angry and more anxious than men who
self-evaluated reserve does not significantly are equally depressed. Similarly, the escalat-
affect the unobtrusive expression factor. Re- ing anger model shows that women are more
gardless of which is a better measure, nei- likely to yell at someone than men who are
ther reserve nor expressiveness explains the equally angry. Thus, our results contradict
observed sex difference in distress. the idea that men and women suffer equal
frustration but transform it into different
emotions. By all our measures, women suf-
Masculine and Feminine Distress fer more distress than men.
Freudian ideas about the psychodynamics of
depression, coupled with contemporaryideas
Related Issues
about sex-role socialization, suggest the pos-
sibility of gendered response to stress Alcoholism, Drug Abuse, and Antisocial
(Rosenfield 1980). According to the theory, Behavior. Our analysis examines sex differ-
frustration naturally produces hostility and ences in distress. A focus on misery and suf-
anger. However, women learn to repress fering seems justified on its own, without
those feelings as they are contrary to nurtur- reference to other values. It is worse to feel
ing and supportive feminine roles. Repress- distressed, sad, lonely, worried, tense, anx-
ing hostility and anger toward others redi- ious, angry, annoyed, run down, and unable
rects punishment inward, producing depres- to concentrate or to sleep than to feel happy,
sion. If true, this dynamic raises the possi- hopeful about the future, and to enjoy life.
bility that men and women differ only in the However, disorder may take behavioral
type of distress and not in the amount of it. forms as well as emotional ones. The distinc-
In theory, men become angry; women be- tion raises the possibility that gendered re-
come depressed. In fact, women become an- sponses may occur across realms of disorder,
gry as well as depressed. Depression is even though they do not occur within the
anger's companion: not its substitute. emotional realm (Horowitz and White 1987).
The gendered-response explanation of In other words, women and men may experi-
women's greater reported distress assumes ence equal levels of frustrationand hardship
three things that all appearto be false. First, that produce emotional problems in women
the explanation assumes that the recognition and behavioral problems in men. In particu-

lar, surveys of the general population find active drugs even if women use them more
that women qualify for psychiatric diagnoses often.
of affective disorders more frequently than It is uncertain whether men use drugs to
men, whereas men qualify for diagnoses of cope more frequently than women do. It is
alcoholism and drug abuse more frequently clear that alcoholism and drug abuse are tem-
than women (Aneshensel, Rutter, and Lach- poraryand counterproductiveescapes at best.
enbruch 1991). Gove and Tudor (1977) ar- For cross-realmgendered response to explain
gue that symptoms or diagnoses from differ- women's greater distress, some behavioral
ent realms should not be combined-that disorder characteristic of men must reduce
they representinherently distinct phenomena distress. If the disorder does not lower dis-
that may be interrelated but should not be tress, then it cannot account for lower male
confounded. Indeed, research shows that levels of distress. On this count, there seems
some stressors associated with depression little or no support for such a cross-realm
and anxiety differ from those associated with gendered response. For the most part, stud-
alcoholism and drug abuse (Aneshensel et al. ies find that distress increases with increased
1991). However, the possibility remains that levels of antisocial behavior, alcoholism, and
women feel more distressed than men be- drug abuse, which are the main problems
cause the men transform their frustrations found more commonly in men than in
into behavioral disorder. women (Boyd et al. 1984; Dohrenwend,
The possibility of cross-realmgendered re- Dohrenwend, Shrout, Egri, and Mendelsohn
sponses extends to a broaderarena the ques- 1980; Endicott and Spitzer 1972). Alcohol-
tion addressed in our analysis. Do women ism, drug abuse, and antisocial personality
merely appear more distressed than men be- multiply the odds of a major depressive epi-
cause the men convert distress into other sode by 4.1, 4.2, and 5.1 respectively (Boyd
realms of disorder? This question is beyond et al. 1984). In sum, there is no evidence that
the scope of the data analyzed here, but con- men's use of alcohol or illegal drugs explains
ceptual distinctions and empirical observa- men's lower distress levels as an artifact of
tions suggest that the answer is no. Concep- sex differences in the expression of prob-
tually, problems like alcoholism, drug abuse, lems. Heavy drinking and drug abuse prob-
and antisocial behavior are not in themselves ably do not protect men from feeling dis-
distress. If men escape distress by that route tressed.
then the sex difference in distress is genuine, Isolation and guilt. Our review of the lit-
not false. Empirically, though, alcoholism, erature finds a few instances in which men
drug abuse, and antisocial behavior probably score higher than women on something
produce more distress than they avoid. If viewed as a sign of distress. A study of
men were not as inclined toward behavioral symptoms in the Center for Epidemiologic
disorder, the gender gap in distress might be Studies' Depression scale (CES-D) finds
even larger. that married men report 42.6 percent more
What pacifier for frustration not included days than their wives of feeling disliked and
in our study might possibly explain men's that people are unfriendly (Ross and Mir-
lower distress? Alcoholism and drug abuse owsky 1984). There is no evidence that
are the chief candidates. Clearly, men drink these perceptions protect men against other
more heavily and use illegal drugs more fre- forms of depression. A subindex of the two
quently than women. However, women actu- items correlates +.39 with depressed affect
ally may depend on drugs for emotional re- for the men and +.36 for the women (.59
lief more frequently than men. Women use and .52 respectively corrected for attenua-
prescribed psychoactive drugs far more than tion). Despite feeling disliked more often,
men-over one woman in S compared to husbands report feeling lonely and de-
less than one man in 10 (Verbrugge 1985). pressed less frequently than do their wives.
The fact that women use psychoactive drugs A study of symptoms in the Psychiatric
on the advice of their doctors, whereas men Evaluation Research Interview (PERI) finds
use psychoactive drugs on their own author- that more men than women say they have
ity, is a legal distinction. It makes it appear done something wrong or evil and deserve
that men have more problems with psycho- to be punished (Newmann 1984). Despite

that self judgement, the men feel less sad "Exclusion Criteria of DSM-III: A Study of
and depressed than the women. Perhaps, as Co-Occurrence of Hierarchy-FreeSyndromes."
equity theory says, the guilt of the advan- Archives of General Psychiatry. 41:983-89.
taged is not as bad as the misery of the dis- Bradburn, Norman M. 1969. The Structure of
Psychological Well-Being. Chicago, IL:
advantaged (Mirowsky and Ross 1985). Aldine.
Men report feeling disliked and guilty more Clancy, Kevin and Walter R. Gove. 1974. "Sex
than do women. Were it not for men feeling Differences in Mental Illness: An Analysis of
disliked and guilty, the effect of sex on de- Response Bias in Self-Reports." American
pression would be even greater than it is. Journal of Sociology 80:205-15.
Conger, Rand D., Frederick 0. Lorenz, Glen H.
Elder, Ronald L. Simons, and Xiaojia Ge.
Conclusion 1993. "Husband and Wife Differences in Re-
We find that women experience distress sponse to Undesirable Life Events." Journal of
Health and Social Behavior 34:71-88.
about 30 percent more frequently than men. Cooperstock, Ruth and Penny Parnell. 1975.
Women's extra burden of distress cannot be Comment on Clancy and Gove. American Jour-
dismissed as mere bias due to greaterexpres- nal of Sociology 81:1455-57.
siveness or to a "feminine"ratherthan "mas- Dohrenwend, Bruce, and Barbara S. Dohren-
culine" emotional response. wend. 1976. "Sex Differences in Psychiatric
Disorder." American Journal of Sociology
John Mirowsky is Professor of Sociology at The 82:1447-59.
Ohio State University. He is co-author, with . 1977. "Reply to Gove and Tudor."
Catherine E. Ross, of Social Causes of Psycho- American Journal of Sociology 82:1336-45.
logical Distress (Aldine de Gruyter, 1989). He is Dohrenwend, Bruce, Barbara S. Dohrenwend,
principal investigator on a grant from NIA, "Ag- Patrick E. Shrout, Gladys Egri, and Frederick
ing, Status, and the Sense of Control, "for which S. Mendelsohn. 1980. "Nonspecific Psycho-
Catherine Ross is co-principal investigator. Cur- logical Distress and Other Dimensions of Psy-
rently he is researching the effects of age on de- chopathology." Archives of General Psychia-
pression, anxiety, fear, mistrust, and the sense of try. 37:1229-36.
control. Endicott, Jean and Robert L. Spitzer. 1972.
"What! Another Rating Scale? The Psychiatric
Catherine E. Ross is Professor of Sociology at Evaluation Form." The Journal of Nervous and
The Ohio State University. She is studying the ef- Mental Disease. 154:88-104.
fects of work and family on men's and women's Frank, Ellen, Linda L. Carpenter, and David
sense of control, health, and mental health. Her Kupfer. 1988. "Sex Differences in Recurrent
latest research examines socioeconomic status Depression: Are There Any That Are Signifi-
and health, with a focus on the processes by cant?" American Journal of Psychiatry.
which educational attainment affects health. She 145:41-5.
is principal investigator on a grant from NIMH, Gove, Walter R. 1978. "Sex Differences in Men-
"Community,Crime, and Health," which will ex- tal Illness among Adult Men and Women: An
amine the ways in which neighborhood affects Evaluation of Four Questions Raised Regard-
subjective well-being. ing the Evidence on the Higher Rates of
Women." Social Science and Medicine
12B: 187-98.
REFERENCES . 1993. "Higher Rates of Physical Symp-
Aneshensel, Carol S. 1992. "Social Stress: toms among Homeless Women Do Not Appear
Theory and Research." Annual Review of Soci- to be Due to Reporting Bias: A Comment on
ology 18:15- 38. Ritchey et al." Journal of Health and Social
Aneshensel, Carol S., Carolyn M. Rutter, and Pe- Behavior 34:178-81.
ter A. Lachenbruch. 1991. "Social Structure, Gove, Walter R. and Kevin Clancy. 1975. "Re-
Stress, and Mental Health: Competing Concep- sponse Bias, Sex Differences, and Mental Ill-
tual and Analytic Models" American Sociologi- ness: A Reply." American Journal of Sociol-
cal Review 56:166-78. ogy 81:1463-72.
Bentler, Peter M. 1989. EQS StructuralEquations Gove, Walter R. and Michael R. Geerken. 1977.
Program Manual. Los Angeles, CA: BMDP. "Response Bias in Surveys of Mental Health:
Boyd, Jeffrey H., Jack D. Burke, Ernst Gruen- An EmpiricalInvestigation."American Journal
berg, Charles E. Holzer, Donald S. Rae, Linda of Sociology 82:1289-1317
K. George, Marvin Karno, Roger Stolzman, Gove, Walter R., James McCorkel, Terry Fain,
Lary McEnvoy, and Gerald Nestadt. 1984. and Michael D. Hughes. 1976. "Response Bias

in Community Surveys of Mental Health: Sys- cal Measurement 1:385-401.

tematic Bias or Random Noise?" Social Sci- Ritchey, Ferris J., Mark LaGory, and Jeffrey
ence and Medicine 10:497-502. Mullis. 1991. "Gender Differences in Health
Gove, Walter R. and Jeanette Tudor. 1977. "Sex Risks and Physical Symptoms among the
Differences in Mental Illness: A Comment on Homeless." Journal of Health and Social Be-
Dohrenwend and Dohrenwend." American havior 32:33-48.
Journal of Sociology 82:1327-35. . 1993. "A Response to the Comments of
Gurin, Gerald, Joseph Veroff, and Sheila Feld. Walter R. Gove on Gender Differences in
1960. Americans View Their Mental Health. Health Risks and Physical Symptoms among
New York: Basic. the Homeless." Journal of Health and Social
Horowitz, Allan and Helene Raskin White. 1987. Behavior 34:182-85.
"Gender Role Orientations and Styles of Pa- Rosenfield, Sarah. 1980. "Sex Differences in De-
thology among Adolescents." Journal of pression: Do Women Always Have Higher
Health and Social Behavior 28:158-70. Rates? Journal of Health and Social Behavior
Langner, Thomas S. 1962. "A Twenty-Two Item 21:33-42.
Screening Score of Psychiatric Symptoms In- . 1982. "Sex Roles and Societal Reactions
dicating Impairment." Journal of Health and to Mental Illness: The Labeling of Deviant De-
Human Behavior 3:269-76. viance." Journal of Health and Social Behav-
Mirowsky, John. 1985. "Depression and Marital ior 23:18-24.
Power: An Equity Model." American Journal Ross, CatherineE. and Joan Huber. 1985. "Hard-
of Sociology 91:557-92. ship and Depression." Journal of Health and
Mirowsky, John, and Catherine E. Ross. 1983. Social Behavior 26:312-27.
"The Multidimensionality of Psychopathology Ross, Catherine E. and John Mirowsky. 1984.
in a Community Sample."American Journal of "Components of Depressed Mood in Married
CommunityPsychology 11:573-91. Men and Women." American Journal of Epi-
. 1986. "Social Patterns of Distress. An- demiology. 119:997-1004.
nual Review of Sociology 12:23-45. Ross, Catherine E., John Mirowsky, and Joan
1991. "Eliminating Defense and Agree- Huber. 1983. "Dividing Work, Sharing Work,
ment Bias from Measures of the Sense of Con- and In-Between: Marriage Patterns and De-
trol: A 2 x 2 Index." Social Psychology Quar- pression." American Sociological Review 48:
terly 54:127-45. 809-23.
Newmann, Joy P. 1984. "Sex Differences in De- Seiler, Lauren H. 1975. "Sex Differences in Men-
pressive Symptoms." Pp. 301-23 in Research tal Illness: Comment on Clancy and Gove's In-
in Community and Mental Health, vol. 4, ed- terpretations."American Journal of Sociology
ited by J. R. Greenley. Greenwich, CT: JAI. 81:1458-62.
Nolen-Hoeksema, Susan. 1987. "Sex Differences United States Bureau of the Census. 1993. Statis-
in Unipolar Depression: Evidence and Theory." tical Abstract of the United States. 113th ed.
Psychological Bulletin 101:259-82. Washington, DC: Government Printing Office.
O'Rourke, Diane and Johnny Blair. 1983. "Im- Verbrugge, Lois M. 1985. "Gender and Health:
proving Random Selection in Telephone Sur- An Update on Hypotheses and Evidence."
veys." Journal of Marketing Research 20:428- Journal of Health and Social Behavior 26:156-
32. 82.
O'Rourke, Diane and Edward Lakner. 1989. . 1986. "From Sneezes to Adieux: Stages
"Gender Bias: Analysis of Factors Causing of Health for American Men and Women." So-
Male Underrepresentation in Surveys." Inter- cial Science and Medicine 22:1195-1212.
national Journal of Public Opinion Research. Waksberg, Joseph. 1978. "Sampling Methods for
1:162-76. Random Digit Dialing." Journal of the Ameri-
Pearlin, LeonardI. 1989. "The Sociological Study can Statistical Association 73:40-46.
of Stress." Journal of Health and Social Behav- Weissman, Myrna M. and Eugene S. Paykel.
ior 30:241-56. 1974. The Depressed Woman. Chicago, IL:
Phillips, Derek L. and Bernard F. Segal. 1969. University of Chicago Press.
"Sexual Status and Psychiatric Symptoms." Wood, Wendy, Nancy Rhodes, and Melanie
American Sociological Review 34:58-72. Whelan. 1989. "Sex Differences in Positive
Radloff, Leonore. 1977. "The CES-D Scale: A Well-Being: A Consideration of Emotional
Self-Report Depression Scale for Research in Style and Marital Status." Psychological Bul-
the General Population." Applied Psychologi- letin 102:249-64.