You are on page 1of 23

Translate this document to:

AUTHOR: Susan Averett; Sanders Korenman

TITLE: The Economic Reality of The Beauty Myth

SOURCE: The Journal of Human Resources v31 p304-30 Spr '96

The magazine publisher is the copyright holder of this article and it is reproduced with permission.
Further reproduction of this article in violation of the copyright is prohibited.

PUBLISHER ABSTRACT AB We investigate income, marital status, and hourly


pay differentials by body mass (kg/m[sup2) in a sample of 23- to 31-year-olds
drawn from the 1988 NLSY. Obese women have lower family incomes than women whose
weight-for-height is in the "recommended" range. Results for men are weaker and mixed.
We find similar results when we compare same-sex siblings in order to control for family
background (for example, social class) differences. Differences in economic status by body
mass for women increase markedly when we use an earlier weight measure or restrict the
sample to persons who were single and childless when the early weight was reported.
There is some evidence of labor market discrimination against obese women. Differences in
marriage probabilities and spouse's earnings, however, account for 50 to 95 percent of
their lower economic status. There is little evidence that obese African American women
suffer an economic penalty relative to other African American women.

I. INTRODUCTION
Eating disorders (such as anorexia nervosa and bulimia) are a major social problem in the United
States. Estimates of the prevalence of anorexia nervosa range from 1 to 4 percent of the female
population (Autry et al. 1986). Although estimates of the prevalence of bulimia nervosa run as high as
20 percent of college and high school females, studies based on representative samples suggest the
actual proportion is less than 5 percent of college females and less than 1.5 percent of males.(FN1)
Concern about eating disorders has led to a recent explosion of consciousness-raising efforts, perhaps
best exemplified by the best-selling book The Beauty Myth by Naomi Wolf.
There is also growing awareness of the social stigma attached to being overweight (examples from
the popular press include Kolata, Brody, and Rosenthal 1993; Coleman 1993; Lampert 1993). These
articles leave little doubt that Americans (especially women) experience great social and psychological
pressure with respect to body size, and they provide poignant accounts of ridicule and
discrimination experienced by obese persons.
Until very recently, economists have had little to say on the issue of body weight in contemporary
industrialized societies; our search of the Journal of Economic Literature index uncovered only one
study of obesity and economic outcomes, a cross-sectional analysis of wage rates (Register and
Williams 1990).(FN2) However, there are several new studies of the economic effects of obesity and
appearance more generally (Gortmaker et al. 1993; Loh 1993; Hamermesh and Biddle 1993; Sargent
and Blanchflower 1994). We summarize these below.
In this paper, we describe economic differentials by body mass for a sample of men and women
aged 23 to 31 in 1988. In so doing, we hope to contribute to a literature that describes social and
psychological pressures that may contribute to the development of eating disorders and to gender
differences in prevalence rates.(FN3) Accurate information about economic differentials by body mass
can aid in the formulation of appropriate public health and social policies. Raising awareness of the
social and psychological pressures surrounding fear of weight gain, and of the stereotyping of obese
persons, represents an important step in addressing associated social and public health problems;
however, we hypothesize that attempts to change a variety of attitudes and behaviors surrounding
body weight, including the social stigma attached to being overweight, eating disorders such as
anorexia nervosa and bulimia, and obsession with body image, diet, and weight loss, will face greater
difficulties if economic differences reinforce social and psychological pressures.(FN4)
That there is an economic aspect to behavior surrounding body weight would hardly be surprising.
A recent Business Week article (Armstrong and Mallory 1992) reported that U.S. companies had 1991
sales of $8.4 billion in products and services for serious dieters. A less restrictive definition that
includes expenditures on items such as health club fees and artificial sweeteners raises the figure to
$33 billion, roughly the GDP of Pakistan, Egypt, or Hungary (World Bank 1992, Table 3).
Naomi Wolf provides additional anecodotal evidence that U.S. college students invest heavily in
appearance (weight loss) related human capital. She finds that audiences of college women have little
trouble answering a series of specific questions about the caloric content of different foods, the
number of calories one must consume to lose weight at different rates, the number of calories
consumed by different amounts of physical exercise and so forth. An economist is naturally led to ask
if there is an economic return to such investments. A number of recent articles suggest that there is.

II. STUDIES OF OBESITY AND ECONOMIC STATUS


Register and Williams (1990) examine the effect of obesity on wage rates in a sample of 18- to
25-year-olds from the 1982 round of the National Longitudinal Survey of Youth (NLSY). In
employment-selectivity corrected wage equations, the pay differential is minus 12 percent for obese
women and minus 5 percent for obese men. They interpret their results as evidence of
discrimination against obese women, although they are aware of other interpretations. In particular,
they mention potential problems of reverse causality: "... the obese may have such status [obesity]
because of low earnings, to the extent that income level affects food and nutritional consumption
behavior" (p. 138, emphasis in original).(FN5)
Loh (1993) followed up Register and Williams (1990) with an analysis of wage levels of full-time
workers in the 1982 NLSY and of wage changes between 1982 and 1985. He finds no effect of
obesity on male or female wage levels in 1982, although wages grew about 5 percent less between
1982 and 1985 for obese men and women.
Sargent and Blanchflower (1994) study a British cohort of 23-year-olds in the 1981 wave of the
National Child Development Study. They report that women who were obese at age 16 earn about 7
percent less per hour at age 23 than their nonobese peers; there are no significant differences for
men. In addition to standard controls, they include a set of controls for social class (occupation of the
household head when the respondent was 11) and IQ (a test of math and reading ability administered
at ages 7, 11, and 16). They also report that women who were obese at age 16 suffered a similar
wage disadvantage at age 23, whether or not they remained obese at age 23.
Hamermesh and Biddle (1993) study the effects of physical attractiveness on earnings for men and
women in several data sets. Although they focus on interviewer ratings of appearance, they include a
control for obesity status (also determined by interviewer observation) in their analysis of the 1977
quality of Employment Survey. In log wage models that also control for a subjective rating of beauty,
they find that obese women earned about 12 percent less than their counterparts of "average" weight,
although this difference was not statistically significant.
Gortmaker et al. (1993) estimate effects of obesity(FN6) on several social and economic outcomes
in the NLSY. They relate an indicator of obesity--being above the 95th percentile of National Center
for Health Statistics standards of the Body Mass Index (BMI; defined as weight in kilograms divided by
the square of height in meters)--at ages 16 to 24 to ages 23 to 31 values of household income, years
of education completed, an index of self-esteem (measured in 1987), and probabilities of being
married, graduating from college, and being poor. In unadjusted comparisons, obese women exhibit
substantial disadvantages in all outcomes except self-esteem. When Gortmaker et al. use multivariate
regressions to adjust for baseline (1979) values of income, education, marital status, maternal and
paternal education, work-limiting chronic health conditions, height in 1981, self-esteem in 1980, age
in 1981, and race or ethnic group, statistically significant differentials remain in marital status,
income, proverty, and years of education; differences in later self-esteem and in the fraction
completing college are small and not significant. Differences by obesity status for men are smaller
and, except for the lower fraction married at ages 23 to 31 among men who were obese at ages 16 to
24, were not significant at the 0.01 level.
Gortmaker et al. investigate the importance of two explanations for the deficits in social and
economic status among obese women. First, they find no evidence to support the hypothesis that
obesity differentials are confounded by health status, since controlling for work-related health
limitations does not change their results. Second, they reject the hypothesis that socioeconomic origin
or ability accounts for the obesity differentials, because significant differentials in income, marriage,
and years of education remained "after we controlled for base-line differences in potentially
confounding factors" (p. 1011). Gortmaker et al. conclude that discrimination may explain the
residual (adjusted) deficits in socioeconomic status among obese women:

In summary, overweight during adolescence has important social and economic consequences that are greater than
those associated with many other chronic physical and health conditions. Discrimination against people who are
overweight may account for these results. The recent Americans With Disabilities Act prohibits discrimination in
employment and in establishments serving the public. Our data suggest that the extension of this act to include
overweight persons should be considered.

We build upon and extend these analyses in several respects. First, we estimate differences in
hourly pay at ages 23 to 31. This wage analysis focuses on the area of central importance to labor
market antidiscrimination policies such as the Americans with Disabilities Act. Gortmaker et al. do not
study labor market outcomes. Like Register and Williams, and Sargent and Blanchflower, we find
evidence consistent with pay discrimination against obese women. There is also evidence that obese
women are more likely than other women to report having experienced gender-based discrimination
in the labor market.
Second, in addition to studying hourly wages, we study marriage probabilities, spouse's earnings,
and family income. We are therefore able to assess the relative contributions of marriage-market
factors (probability of marriage and spouse's earnings) and labor market factors (employment and
earnings) to the overall differences in family income between obese persons and persons of
recommended weight. This accounting exercise allows us to assess the potential for labor market
antidiscrimination measures to increase the economic status of obese women.
Third, we use same-sex sibling differences as an alternative (and perhaps more complete) way to
control for social class or family background differences between obese persons and others. For
women, estimates based on sibling differences are broadly consistent with those based on
conventional cross-sectional estimates. These sibling differences address the concern raised by
Stunkard and Sorensen (1993) in their editorial that accompanied the Gortmaker et al. piece: "The
possibility, indeed the probability, that a common factor or factors influence both obesity and
socioeconomic status greatly complicates the attribution of causality. In addition to control for social
factors, as was carried out so effectively by Gortmaker et al., further assessment of the relation will
also require control for parental obesity" (p. 1037, italics added).
Fourth, we attempt to address the possibility of bias from reverse causality in contemporaneous
relations between economic status and body weight. We estimate contemporaneous (in other words,
at ages 23 to 31) relationships between economic or social outcomes and body weight, in addition to
relationships between outcomes at ages 23 to 31 and obesity at ages 16 to 24 (the latter are similar
to those estimated by Gortmaker et al.). Contemporaneous relationships between social and economic
outcomes and obesity (at ages 23 to 31) are weaker than lagged relationships (namely, those
between outcomes at ages 23 to 31 and obesity at ages 16 to 24). As noted, women who are obese
at ages 16 to 24 have lower economic status at ages 23 to 31. However, although most women who
are obese at ages 16 to 24 are also obese at ages 23 to 31 (as Gortmaker et al. note), only about 30
percent of women who are obese at ages 23 to 31 were obese at ages 16 to 24. Moreover, women
who become obese in their mid- to late 20s (the majority of women who are obese at ages 23 to 31)
appear to be better-off financially than those who were obese at both ages, and they do not differ
greatly from those in the recommended weight range. This finding suggests that contemporaneous
social and economic differentials are most likely not the result of adverse labor or marriage market
outcomes causing weight gain; using a lagged BMI measure (which should be less affected by reverse
causality) strengthens the adverse association between obesity and economic status.
A final extension of previous research is our estimation of separate models by black or Hispanic
identification of sample members. Unlike Gortmaker et al., who report that "The addition of interaction
terms to the models to determine whether the relation of obese to subsequent social and economic
characteristics varied according to race or ethnic group did not alter the results (p. 1011)," we find
some statistically significant and large racial differences in the obesity differentials; the social and
economic penalties attached to being overweight appear to be smaller among black women.

III. METHODS, DATA, VARIABLES

A. THE NLSY DATA


The sample is derived from the National Longitudinal Survey of Labor Market Experience of Youth
(NLSY), which has been conducted annually since 1979 (CHRR 1992). At baseline (1979), respondents
were aged 14 to 21. The NLSY oversamples black, Hispanic, and economically disadvantaged nonblack
and non-Hispanic youths. Respondents were asked to report their current weight in the 1981, 1982,
1985, 1986, 1988, 1989, and 1990 interviews. Height information was collected in 1981, 1982, and
1985. Height information was not collected after 1985, presumably because individuals are assumed
to have attained adult stature. (Sample members were aged 20 to 27 in 1985.)
We focus on labor market and marriage-market outcomes measured in the 1988 interview. Our
sample consists of 5,090 women and 4,951 men who were interviewed in 1988 and for whom we had
the requisite height, weight, and hourly wage information (if they were employed; nonemployed
persons are also included in the sample).(FN7)

B. VARIABLES
The explanatory variables of chief interest are categories of the Body Mass Index (BMI; Bray
1978), which, as noted, is defined as weight in kilograms divided by the square of height in meters.
Although there are many ways to combine height and weight to estimate amount of body fat, these
ways tend to be highly correlated and are considered reliable (Kannel 1983; Abraham 1983). In
addition, Bray reports a correlation between the BMI and various anthropometric measures (such as
skinfold thickness) of 0.7 to 0.8.
We use BMI categories that correspond to weight-for-height tables formulated by Metropolitan Life
Insurance Company (1983). The recommended (based on associated mortality risks) BMI range is 20-
25 for men and 19-24 for women. We follow convention (Bray 1979) in referring to persons below the
recommended range as "underweight," men with BMIs between 25 and 29 and women with BMIs
between 24 and 29 as "overweight," and men or women with BMIs 30 and over as "obese." We must
emphasize that the recommended weight refers to a range associated with low mortality risks, and
may not correspond to social norms about what might constitute an overweight or underweight
appearance. We adopt these ranges because they are conventional, widely used, and are a convenient
way to classify the sample. The Metropolitan Life recommended weight tables are appropriate for
individuals aged 18 to 30 (Greenwood 1983); we use weight data for a sample aged 16 to 24 in 1981
and 23 to 31 in 1988.
We use BMI measures at two ages: an average of the 1981 and 1982 BMIs, and an average of the
1988 and 1989 BMIs (based upon 1985 height and 1988 and 1989 weights; height information was
last collected in 1985). For simplicity, we refer to these measures as the 1981 and 1988 BMIs. We
examined height data for inconsistencies and flagged 177 women and 240 men who appeared to
"shrink" more than two inches in height between either the 1981 and 1982 interviews, or the 1981 or
1982 interview and the 1985 interview. For these individuals, we assigned heights based on a
combination of the three reports.(FN8) In addition, if 1981 and 1982 heights were both missing, they
were set to the 1985 value.
Outcome variables are taken from the 1988 interviews when respondents are aged 23 to 31. Basic
control variables include age (or actual work experience for wage models), highest grade completed
as of 1988, region, SMSA, and black or Hispanic identification. We also conduct some analyses
separately by race/Hispanic identification. In addition to these basic control variables, we included a
set of more detailed controls in order to investigate hypotheses about the source of earnings and
income differentials by body mass. These variables include an indicator of work-related health
limitations in 1988 (equal to 1 if the respondent reported that health limited ability to work or amount
or type of work; and 0 otherwise), the respondent's score on the Armed Forces Qualifications Test (a
test of academic ability and achievement administered in 1980), and an index of low self-esteem
based on responses to a battery of ten questions asked in 1980. To compute the index of low self-
esteem we summed the number of "low esteem" responses and formed a variable with mean of 0 and
standard deviation of 1; higher values indicate lower self-esteem. In some models we also enter
controls for marital status in 1988, number of children, and age of youngest child in 1988. Finally,
although we do not attempt to control for sex discrimination in any models, we present proportions
who responded yes to question last asked in 1983: "Have you ever experienced discrimination at
work? ... Was that on the basis of sex?" (No questions were asked about weight-based
discrimination.)
Most of the variables used in the analyses are standard and self-explanatory; two require additional
explanation. Actual labor market experience is based on reports of weeks worked each year after age
18. A good experience measure may be important for a study of body mass and wages because, for
example, if childbearing is associated with higher body mass it may lead higher body mass to be
associated with lower wages through the effect of childbearing on the accumulation of labor market
experience (for example, Mincer and Polachek 1974; Filer 1993).
Our principal measure of economic well-being is the income-to-needs ratio in 1988, based on 1987
income (reported in the 1988 interview) and family composition as of the 1988 interview. It is defined
as the family income of the respondent divided by the U.S. Census poverty line for the family based
on its size and age composition (number of adults and children).(FN9)

C. SAMPLE DESCRIPTION
Table 1 presents a cross-tabulation of the distribution of the sample by BMI categories in 1981
and 1988. In 1988, at ages 23 to 31, about half the sample women are in the recommended (19 to
23) BMI range, 30 percent are between 24 and 29, 13 percent had BMIs over 30, and 7 percent were
below 19. Especially notable is the increase with age (between 1981 and 1988) in the fraction in the
top two categories: the fraction above 30 rose from five to 13 percent, and the fraction above 23 rose
from 24 to 43 percent. The substantial increase between 1981 and 1988 in the fraction of women in
the overweight and obese categories highlights the possibility that reverse causality may bias
estimates of contemporaneous relations between BMI and economic status (specifically, that adverse
economic outcomes may cause weight gain). Although relatively few women (16 percent) who were
obese in 1981 exited the obese category between 1981 and 1988, only about 31 percent (199 out of
646) of women who were in the obese category at ages 23 to 31 were also in the obese category at
ages 16 to 24.
There was a similar increase with age (between 1981 and 1988) in the fraction of men in the two
heaviest BMI categories; the fraction with BMIs above 29 rose from 4 to 11 percent, and the fraction
above 24 rose from 25 to 48 percent, a slightly greater increase than among women. The similar
pattern of weight gain for men and women suggests that the increase for women was not entirely due
to biological aspects of pregnancy and childbirth.(FN10) However, we explore this possibility below.
In Table 2 we present sample means and frequencies for women classified by BMI at ages 16 to 24
(1981). According to all indicators, obese women are of lower socioeconomic status than women of
recommended weight. Family income, income/needs, spouse's earnings, hourly wages, years of
schooling, Armed Forces Qualifications Test score, the fraction in managerial/professional occupations,
and the fraction employed are lower in the higher BMI categories, while the fractions poor, minority,
and those in unskilled manufacturing occupations are higher. In addition, obese women are more
likely to report health limitations and having experienced sex discrimination. Overweight and obese
women are somewhat more likely to have children and are less likely to be married. The fact that
body weight is inversely associated with economic status in developed societies is well documented
(Kannel 1983). One goal of this study is to use sibling comparisons to investigate whether this
relationship reflects the influence of family background (for example, social class) on weight, or
whether there appears to be an association between weight and economic status in later life, net of
family background differences.
Table 3 presents means and frequencies for men by BMI category at ages 16 to 24 (1981). The
patterns for men differ from those for women. Income appears to be lowest among underweight men;
differences across the other BMI categories are modest. Men in the lowest BMI category appear less
likely to be married, more likely to be divorced or separated, and less likely to have children. A larger
proportion are black. Spouse's earnings are slightly lower among obese men. Armed Forces
Qualifications Test scores are lower for underweight men; obese men are more likely to report health
problems.

IV. RESULTS
Table 4 summarizes the results of multivariate models of income, marriage, spouse's earnings,
and hourly wages for women and men. We estimated each model for the full sample (separate models
by sex) using, alternatively, BMI at ages 23 to 31 (1988) and BMI at ages 16 to 24 (1981). In
addition, each model is estimated for the subsample of persons who were childless and single (never
married) in 1982 using the 1981 BMI. The models contain a set of basic controls for age (or actual
work experience in wage models), years of schooling completed, and dummy variables for region (3),
residence in an SMSA (2), and black and Hispanic identification.

A. RESULTS FOR WOMEN


The figures in the table are coefficients of BMI categorical variables from least squares
regressions, except for those under the heading "P(married)," which are derivatives based on logit
coefficients, evaluated at the sample means. These derivatives may be interpreted as percentage
point differences in the fraction married. The reference group for all models are persons in the
"recommended" BMI range. Same-sex sibling differences are estimated by least squares for the three
continuous outcomes,(FN11) and fixed-effects logits (Chamberlain 1980) for the dichotomous outcome
(marital status in 1988). Because necessarily sample sizes are much smaller for sibling analyses,
coefficient estimates are less precise.(FN12) We therefore look to the sibling analyses for general
support for, or obvious contradiction of, evidence from cross-sectional analyses. Cross-sectional
estimates have the advantage of being based on larger samples, but may suffer from bias induced by
unmeasured family background differences.
In the first panel of the table we present differentials in outcomes at ages 23 to 31 (1988)
according to BMI category at ages 16 to 24 (1981). The cross-sectional models suggest that women
who were obese or overweight at ages 16 to 24 have, at ages 23 to 31, lower family income and lower
hourly wages (30+ category only for the latter), are less likely to be married, and have lower spousal
income (if married) than women in the recommended BMI range. The pattern of coefficients from the
sibling analysis is similar, suggesting that family background heterogeneity bias is not serious in
cross-sectional estimates.(FN13)
In the second panel we present differentials in outcomes at ages 23 to 31 (1988) according to BMI
category at ages 23 to 31. Differentials by BMI category in family income, spouse's earnings, and
wages are similar to but smaller (in absolute value) than those in the first panel. There is no evidence
of lower probabilities of marriage among heavier women in the second panel. What do we learn from
the comparison between the first two panels of the table? First, differences in income, marriage,
spousal income, and wages by BMI category in the second panel (ages 23 to 31 BMI, ages 23 to 31
outcomes) do not appear to be, on net, biased downward (in other words, are not made "too
negative") by (reverse) causality running from obesity to economic outcomes at ages 23 to 31. If
anything, they appear to be biased upward. Second, the fact that heavier women are no less likely to
be married according to the second panel but are less likely to be married according to the first panel
suggests that the source of upward bias in the second panel may be that marriage (or perhaps
childbearing) raises weight. As we shall see later, however, controlling for marital status, the presence
of children, and the age of the youngest child has no effect on the estimated wage differentials, and
only a modest effect on the income differentials by BMI.
We also ran models similar to those in Table 4 in which we included an interaction between
obesity in 1981 and 1988, as well as dummy variables for obese in 1981 and obese in 1988. In
models of income-to-needs, the coefficients (standard errors) of the dummy variables were as follows:
obese in 1981, -0.36 (0.15); obese in 1988, 0.05 (0.05); interaction term, 0.08 (0.17). Clearly, the
largest penalty is associated with obesity at the younger ages. Women who became obese between
1981 and 1988 appear to be no worse off than women of recommended weight. These results do not
support the hypothesis that our estimates are biased because poor economic outcomes cause
subsequent weight gain.
The wage models with interactions were slightly different. The coefficients (standard errors) were:
obese in 1981, -0.08 (0.08); obese in 1988, -0.04 (0.03); interaction term, -0.05 (0.09). Thus,
women who were obese in both 1988 and 1981 had the lowest wages in 1988, roughly 17 percent
below those of women of recommended weight (p < .01). Women who became obese between 1981
and 1988 had only slightly lower wages than women of recommended weight.
Finally, in the third panel of the table we present differentials in outcomes at ages 23 to 31 (1988)
according to BMI at ages 16 to 24 (1981) for the subsample of women who were single (never
married), childless, and not pregnant in 1982. Results are similar to those in the first panel, but
differences across BMI categories are larger (in absolute value). Differences in the two marriage-
market outcomes are especially dramatic.

B. RESULTS FOR MEN


Results for men are presented in the bottom half of Table 4. Significant results in the first panel
include lower income and lower spouse's earnings among men who were underweight at ages 16 to
24, and lower wages among those who were obese at these ages. When we relate outcomes at ages
23 to 31 to BMI at the same age (second panel for men), we find that heavier men are more likely to
be married. Results for men tend to be weaker than those for women. For example, unlike results for
women, results from sibling differences for men are often not consistent with those based on standard
cross-sectional analyses. However, as with the results for women, a comparison of the first and
second panels suggests, if anything, that differences in outcomes at ages 23 to 31 between obese or
overweight men and men in the recommended weight range according to age 23 to 31 BMI category
are biased upward by reverse causality, possibly the result of weight gain associated with marriage.
This idea is supported by the finding (reported in the last panel) that among men who were single at
ages 16 to 24, obesity is indeed associated with lower odds of being married at ages 23 to 31.(FN14)

C. ACCOUNTING FOR INCOME DIFFERENCES


How much of the difference in income between obese women or underweight men and their
counterparts in the recommended BMI ranges is accounted for by labor market differences (wages and
employment), and how much by marriagemarket differences (probability of being married and
spouse's earnings if married)? The answer to this question depends somewhat on the sample (for
example, full sample or single/childless only) and the age at which the BMI is measured. The following
figures are based on BMIs at ages 16 to 24 (1981), the full samples of men and women, and have
been adjusted for race, education, age, region, and urban residence.(FN15)
The vast majority (80 percent) of the difference in family income between obese women and those
in the recommended weight range results from differences in the marriage market (lower fraction
married and lower spouse's earnings if married), compared to about 17 percent from the labor
market. Using the age 23 to 31 (1988) BMI measure raises the proportion of the obesity differential
in adjusted income accounted for by labor market differences to about one third, and lowers the
proportion due to marriage-market outcomes to about one half. On the other hand, restricting the
sample to women who were single and childless at ages 16 to 24 raises the proportion of the obesity
differential that is accounted for by marriage-market differences to 96 percent ($7,014 out of
$7,332); labor market differences account for 12 percent in this subsample. (The figures sum to more
than 100 percent because obese women have slightly higher unearned family income.) In short,
although obese women appear to be disadvantaged in the labor market, marriage-market differences
account for the great majority their substantial deficits in economic status. The reverse is true for
underweight men.

D. DETAILED WAGE AND INCOME MODELS


In Tables 5 and 6 we present more detailed analyses of wage and income differences for women
in order to investigate alternative hypotheses concerning the source of the obesity differences in
wages and income.(FN16) A leading hypothesis is that the lower wages of obese women reflect
differences in health status. Both underweight and obesity are associated with health problems; in
fact, these categories are defined by their associations with health risks. Although health differences
could explain the lower earnings of underweight men and obese women, it is not obvious why a
gender difference in these relationships should exist (namely, the hypothesis that the relationship
between economic status and BMI is confounded by unmeasured health status would predict wage
penalties for underweight women and obese men).
A second hypothesis is that low self-esteem is associated both with obesity and with adverse labor
market and marriage-market outcomes.(FN17) To explore this hypothesis we include an index of low
self-esteem (measured in 1981) in the wage and income models. Finally, in the wage models, we also
control for occupational status to investigate whether residual differences in wages are explained by
occupational sorting or segregation.(FN18) In the first column of Table 5 we present (for convenience)
the coefficients reported in the corresponding column of Table 4. In the second column we add to the
variables included in the first column: dummy variables for married and divorced/separated in 1988, a
dummy variable for the presence of children in 1988, an interaction of this variable with the age of the
youngest child, and controls for Armed Forces Qualifications Test score, a dummy variable if the
respondent reported a work-related health limitation in 1988, and the index of low self-esteem
measured in 1981. In the third column we add seven dummy variables representing the occupational
categories listed in Table 2. The remaining three columns are sibling-differenced versions of the same
models. Adding controls for marital status, children, occupation, Armed Forces Qualifications test
score, health, and self-esteem has little effect on the wage differentials by body mass for women; the
coefficient of the 30+ BMI category is stable across the cross-sectional specifications. Sister-
differenced estimates are similar in magnitude to the cross-sectional estimates, although the
estimated effect of obesity is not statistically different from zero and appears more sensitive to the
addition of occupation controls.
Systematic differences in pay linked to a personal characteristic which remain after human capital
differences have been accounted for are often interpreted as evidence of pay discrimination. In this
sense, the wage equation estimates provide evidence of pay discrimination against obese women.
(FN19) Such pay differentials may also reflect unmeasured productivity differences correlated with
body mass. Although it is not possible to determine the importance of the two factors from
information in the NLSY, the hypothesis that at least part of these pay differentials results from labor
market discrimination is bolstered by reports by obese and overweight women that they are
subjected to labor market discrimination because of their weight (for example, Coleman 1993;
Kolata, Brody, and Rosenthal 1992) and, indirectly, by the responses to the sex discrimination
question summarized in Table 2.
In Table 6 we present models of the (log) income-to-needs ratio. As in Table 5, the first column
reports results from the specification used in Table 4, which we repeat here for convenience. In the
second and fifth columns of Table 6, we add the Armed Forces Qualifications Test score, the health
limitations dummy variable, and the index of low self-esteem. Adding these controls has little effect on
the obesity differential in either the cross-sectional or sibling analyses, despite their large effects on
income in many of the models.
In the third column we add dummy variables for married and divorced or separated in 1988, a
dummy variable for the presence of children in 1988, and an interaction of this variable with the age
of the youngest child, in order to assess the degree to which marriage and children appear to
"mediate" the relationships between obesity and economic outcomes. Controlling for marital status
and the presence and age of children has surprisingly little effect on the cross-sectional income gap
for the heaviest women [compare Columns (2) and (3) in Table 6]. However, adding these controls
has a larger effect on the sibling differences [Columns (5) and (6)], so it is difficult to reach a
definitive conclusion.

E. RACE DIFFERENCES
There is a literature that suggests that there are cultural differences in norms pertaining to ideal
body type (for example, Furnham and Alibhai 1983). In particular, there may be a smaller social
penalty attached to being overweight for African American than white women. Consistent with this
hypothesis, black women are more likely to be above recommended body weight, but they are less
likely to perceive themselves to be overweight (Dawson 1988).
This literature suggests several testable hypotheses. The most obvious is that if social pressure is
an important determinant of weight, the prevalence of overweight should be higher among African
Americans than among whites, especially for women. The figures in Tables 2 and 3 confirm that, for
our sample, this is indeed the case. Second, since marriage markets continue to be highly segmented
by race, especially for black women (Kalmijn 1993), we would expect these social norms to be
reflected in marriage probabilities, so that a given weight differential should be associated with larger
differences in marriage probabilities for whites than blacks. Finally, one might also predict that the
difference in hourly wages between obese and recommended-weight black women should be smaller
(in absolute value) than the corresponding difference among white women, although this prediction
requires more explanation.(FN20)
In Table 7 we repeat the analyses of Table 4 for subsamples of Hispanic, non-Hispanic black, and
non-Hispanic white women. Figures in Table 7 are cross-sectional differences in income, marriage,
hourly wages, and spouse's earnings for the full sample according to categories of BMI at ages 16 to
24 (1981) and 23 to 31 (1988). Sibling sample sizes were too small to permit separate analyses of
sibling differences for black and Hispanic subsamples.
We find smaller differences by body mass for African American women than for white women. For
example, regression-adjusted differences in the log of family income/needs at ages 23 to 31 (1988)
between women who were in the 30+ BMI range and those in the 19-24 (reference) range were -0.42
for whites, -0.21 for Hispanics, and -0.08 for blacks. The lower penalty among overweight black
women is also apparent in models of marriage and hourly pay. However, differences across BMI
categories in spouse's earnings (conditional on marriage) do not appear to be smaller among African
Americans than among whites or Hispanics.
In order to determine whether the substantial race differences that appear in these tables are
statistically significant, we reran the models presented in the first panel of Table 4 for women and
introduced an interaction term between obesity in 1981 and black racial identification. The coefficient
of this interaction term in the income/needs equation was (positive) 0.30, and was significant at the
0.01 level. The coefficient of the obesity variable itself was -0.38 (p < 0.01), about the size of the
corresponding coefficients for white and Hispanic women presented in the first panel of Table 7. The
corresponding wage equation coefficients exhibited the same pattern: the coefficient of the black *
obese (in 1981) interaction was 0.13 (p < 0.10). In fact, since the coefficient of the black variable
itself was minus 0.05, obese black women earned on average about 8 percent more per hour than
obese white women (although this difference is not statistically significant; p = 0.20), controlling for
labor market experience, education, and so forth.
Results for men in the bottom panel of Table 7 suggest an obesity wage penalty for white men
(last two columns) only, according to either the contemporaneous or the early BMI measure. Family
income deficits among underweight men appear for all three groups.

V. SUMMARY AND DISCUSSION


We have presented a range of evidence about the association between obesity and economic
status for men and women. There are substantial deficits in family income at ages 23 to 31 among
obese women compared to women with BMIs in the recommended range. There is evidence that
disadvantages in both the marriage and labor markets contribute to these deficits, although marriage
markets seem to play a more important role in determining family income deficits among obese
women. We find similar results when we compare same-sex siblings as a way to control for family
background (for example, social class) differences.
For men, there is less evidence of an effect of body mass (overweight or underweight) on family
income or marriage-market outcomes. There is some evidence of wage penalties associated with both
underweight and obesity among men, although this evidence is less consistent across samples and
model specifications than the evidence for an obesity effect among women.
We took advantage of the longitudinal nature of the data to argue that social and economic
differentials are most likely not the result of adverse labor market or marriage-market outcomes
causing weight gain among women. In particular, using an earlier weight measure strengthens the
adverse association between obesity and economic status.
Finally, the link between empirical analyses and recommendations for protection of obese persons
from labor market discrimination under the Americans with Disabilities Act (for example, by
Gortmaker et al. 1993) requires further clarification. Such recommendations are most properly based
on differences in labor market outcomes, not family income. But, as noted, the great majority (as
much as 96 percent) of the economic deficit associated with obesity among women in our sample
results from differences in the marriage market (especially lower probabilities of marriage), not the
labor market. Labor market remedies such as the Americans with Disabilities Act, therefore, have
limited ability to close the gap in family income between obese women and those of recommended
weight.
Added material
Susan Averett is an assistant professor of economics and business at Lafayette College (Penn.);
Sanders Korenman is an associate professor of public affairs at the University of Minnesota. The
authors thank Dennis Ahlburg, Ted Joyce, Robert Kaestner, Felicia LeClere, Samuel L. Myers, Jr.,
Cordelia Reimers, and seminar participants at Lafayette College, Princeton University, and the
University of Minnesota for their comments. Deborah Campbell provided outstanding research
assistance. The authors take responsibility for all errors. The data used in this article can be obtained
beginning in August 1996 through July 1999 from Susan Averett, Department of Economics and
Business, Lafayette College, Easton, PA 18042.
[Submitted May 1994; accepted February 1995]
Table 1 Distributions of 1988 Body Mass Index, by 1981 Body Mass Index, Full Sample

1988 Body Mass Index


Women <19 19-23 24-29 30+ All N
1981 BMI
<19 35.9 60.3 3.8 0.0 14.1 711
19-23 3.7 63.0 30.3 3.1 62.2 3,139
24-29 0.1 11.3 52.1 36.5 19.1 962
30+ 0.0 3.0 12.7 84.3 4.7 236
All 7.4 49.9 29.9 12.8 100.00
N 371 2,521 1,510 646 5,048
1988 Body Mass Index
Men <20 20-24 25-29 30+ All N
1981 BMI
<20 26.7 66.6 6.7 0.0 10.3 506
20-24 1.4 60.5 35.3 2.8 65.1 3,196
25-29 0.2 11.0 60.5 28.3 20.4 1,003
30+ 0.0 3.4 14.6 82.0 4.2 205
All 3.7 48.7 36.6 11.0 100.0
N 183 2,389 1,798 540 4,910

Table 2 1988 Sample Means and Frequencies, by Body Mass Index in 1981, Women (figures are
percents unless indicated)

Body Mass Index in 1981


All <19 19-23 24-
29 30+
Family income ($) 26,387 29,423 27,747
21,829 16,978
Income/needs 2.9 3.2 3.0
2.2 1.8
Poor 20.5 15.2 18.6
26.4 39.3
Income missing 17.0 14.9 16.9
19.2 16.9
Married 51.7 55.6 53.5
47.3 34.7
Divorced/separated 5.5 4.5 5.0
6.9 9.7
Schooling (years) 12.8 13.0 12.9
12.3 11.8
Age (years) 26.7 26.2 26.6
26.9 27.3
Any kids? 61.5 56.5 61.1
67.0 57.6
Any * age youngest 2.1 1.8 2.0
2.5 2.6
Black 25.9 15.2 24.6
33.2 44.9
Hispanic 15.6 12.9 15.1
20.5 11.0
Body Mass Index, 1988 24.4 19.9 23.2
28.9 36.5
Body Mass Index, 1981 22.3 18.1 21.3
26.2 34.0
Employed 79.2 79.5 80.4
76.7 72.0
Hourly wage ($)(FNa) 7.52 7.56 7.77
6.99 5.84
Actual experience (years)(FNa) 6.1 6.1 6.3
5.7 5.3
AFQT, 1980 39.8 42.6 41.8
34.0 27.8
Low self-esteem, 1980 0.0 0.1 -0.0
0.1 0.2
Low self-esteem, 1987 0.0 -0.1 -0.0
0.1 0.3
Health limits, 1988 8.4 7.5 8.1
8.8 14.1
Sex discrimination, 1983 11.0 10.2 10.3
10.8 17.1
Occupation(FNa)
Managerial/professional 21.6 25.6 22.8
17.3 11.0
Sales 9.4 9.1 9.7
9.0 7.2
Clerical 22.7 21.7 23.5
21.9 18.2
Services 16.5 15.0 15.9
18.4 21.2
Manufacturing, unskilled 8.2 7.9 7.6
10.0 10.2
Manufacturing, skilled 2.0 1.5 2.1
1.8 2.1
Other 19.6 19.1 18.3
21.6 30.1
Spouse's earnings(FNb) 23,482 26,314 24,171
19,331 16,776
Unmarried, no kids, 1982 52.8 57.4 54.5
45.0 47.5
Sample size 5,048 711 3,139
962 236

FOOTNOTES
a. If employed.
b. Annual earnings of spouse, if sample person is married and spouse employed.
Table 3 1988 Sample Means and Frequencies, by Body Mass Index in 1981, Men (figures are
percents unless indicated)

Body Mass Index in 1981


All <20 20-24 25-
29 30+
Family income ($) 27,997 23,638 28,128
29,766 27,267
Income/needs 3.2 2.7 3.3
3.3 3.1
Poor 11.1 14.8 11.1
9.0 12.3
Income missing 21.4 25.3 21.2
19.4 24.4
Married 44.2 36.8 43.7
49.8 43.9
Divorced/separated 4.1 5.9 4.0
4.0 2.0
Schooling (years) 12.6 12.4 12.7
12.6 12.5
Age (years) 26.5 25.5 26.5
27.0 27.2
Any kids? 35.1 31.0 34.0
41.1 32.7
Any * age youngest 0.9 0.7 0.8
1.2 0.9
Black 26.8 25.7 28.8
22.4 20.5
Hispanic 16.3 13.4 15.3
18.8 25.4
Body Mass Index, 1988 25.4 21.3 24.5
28.5 34.3
Body Mass Index, 1981 23.4 19.0 22.4
26.8 33.0
Employed 87.9 82.6 88.0
89.5 91.2
Hourly wage ($)(FNa) 9.37 8.23 9.50
9.59 8.93
Actual experience (years)(FNa) 7.0 6.2 6.9
7.5 7.8
AFQT, 1980 39.7 35.1 39.9
41.3 40.4
Low self-esteem, 1980 0.0 0.1 0.0
-0.0 0.1
Low self-esteem, 1987 0.0 0.1 0.0
0.0 0.1
Health limits, 1988 5.7 6.2 5.1
6.2 9.8
Sex discrimination, 1983 4.0 4.6 3.9
4.2 1.5
Occupation(FNa)
Managerial/professional 18.3 15.8 19.2
16.8 17.6
Sales 6.5 5.5 6.7
6.6 6.8
Clerical 7.6 8.1 7.7
7.2 7.3
Services 10.6 10.1 10.6
11.3 9.8
Manufacturing, unskilled 24.4 22.3 24.3
24.8 29.3
Manufacturing, skilled 19.0 22.1 18.3
19.5 18.5
Other 13.5 16.0 13.2
13.8 10.7
Spouse's earnings(FNb) 12,538 11,355 12,751
12,580 11,068
Unmarried, no kids, 1982 79.7 87.2 81.3
71.4 78.5
Sample size 4,910 506 3,196
1,003 205

FOOTNOTES
a. If employed.
b. Annual earnings of spouse, if sample person is married and spouse employed.
Table 4 Regression-Adjusted Percent or Percentage Point Differences in Marriage and Labor Market
Outcomes in 1988 According to Body Mass Index Category in 1988 or 1981

Income/Needs(FNa) Pr(Married)(FNb) Spouse's


Earnings(FNc) Hourly Wage(FNd)
SIB SIB
SIB SIB
XSEC DIFFS XSEC DIFFS XSEC
DIFFS XSEC DIFFS
Women
Full sample
1981 BMI(FNe)
30+ -0.25(FN*) -0.33(FN*) -0.19(FN*) -0.23(FN*)
-0.23(FN*) -0.25 -0.15(FN*) -0.12
24-29 -0.14(FN*) -0.12 -0.05(FN*) -0.10(FN*)
-0.17(FN*) 0.03 -0.01 -0.03
<19 0.03 0.02 0.01 0.02
0.09(FN*) 0.14 -0.00 -0.07
N 4,188 609 5,048 299 2,137
198 3,996 593
Full sample
1988 BMI(FNe)
30+ -0.14(FN*) -0.07 -0.01 0.02
-0.23(FN*) 0.27 -0.10(FN*) -0.04
24-29 -0.10(FN*) -0.03 0.02 0.02
-0.09(FN*) -0.14 -0.05(FN*) -0.04
<19 -0.09(FN#) 0.17 -0.03 -0.05 -0.01
0.22 -0.04 -0.04
N 4,188 609 5,048 299 2,137
198 3,996 593
Single & childless
1981 BMI(FNe)
30+ -0.33(FN*) -0.27 -0.31(FN*) -0.71(FN*)
-0.42(FN*) -1.06 -0.24(FN*) -0.15
24-29 -0.15(FN*) -0.01 -0.10(FN*) -0.17(FN#)
-0.21(FN*) 0.09 -0.03 0.04
<19 0.03 0.08 0.01 -0.19(FN#)
0.17(FN*) 0.14 0.00 -0.10
N 2,189 256 2,663 146 1,030
79 2,559 288
Men
Full sample
1981 BMI(FNe)
30+ -0.04 -0.22 -0.04 -0.14 -0.09
-0.10 -0.08(FN*) -0.09
25-29 0.05 -0.03 0.02 -0.02
-0.06 0.02 -0.01 -0.04
<20 -0.14(FN*) -0.04 -0.04(FN#) 0.02
-0.20(FN*) -0.32 -0.09(FN*) -0.03
N 3,859 615 4,910 356 1,493
105 4,317 781
Full sample
1988 BMI(FNe)
30+ 0.05 -0.22(FN*) 0.08(FN*) 0.12 -0.02
-0.44 -0.03 -0.06
25-29 0.12(FN*) 0.08 0.08(FN*) 0.09(FN*)
0.15(FN*) 0.19 0.02 0.01
<20 -0.29(FN*) 0.01 -0.09(FN*) 0.05 0.03
-0.00 -0.07(FN#) 0.03
N 3,859 615 4,910 356 1,493
105 4,317 781
Single & childless
1988 BMI(FNe)
30+ -0.02 -0.31 -0.07(FN#) -0.15 0.00
-0.37 -0.06 -0.13
25-29 0.06 -0.09 -0.00 -0.02 0.06
-0.11 -0.02 -0.08(FN#)
<20 -0.17(FN*) 0.03 -0.01 -0.00
-0.19(FN#) -0.35 -0.09(FN*) -0.02
N 3,015 430 3,915 264 1,054
60 3,442 570

Notes: N is the number of observations, or (approximately) the number of pairs for same-sex
sibling differences. See text for details.

FOOTNOTES
a. Dependent variable = In(income/needs) in 1987, where needs are defined as the poverty line
for the family unit. Regression controls include in addition to dummy variables for BMI category: age,
years of schooling, and dummy variables for region (3), racial/Hispanic identification (2), and
residence in an SMSA (2).
b. Dependent variable is dichotomous: "married in 1988." Controls entered in the logit models
include in addition to dummy variables for BMI category: age, years of schooling, and dummy
variables for region (3), racial/Hispanic identification (2), and residence in an SMSA (2). Numbers
shown are derivatives evaluated at the sample mean and may be interpreted as percentage point
differences in the probability of being married in 1988. Sibling differences are from fixed-effects logit
models.
c. Dependent variable = In(spouse's annual earnings) in 1987. The sample consists of sample
members who are married and whose spouses earned at least $100 in 1987. Regression controls
include in addition to dummy variables for BMI category: age, years of schooling, and dummy
variables for region (3), racial/Hispanic identification (2), and residence in an SMSA (2).
d. Dependent variable = In(sample member's hourly wage in survey week). The sample consists of
sample members who worked for pay during the survey week. Regression controls include in addition
to dummy variables for BMI category: actual labor market experience, years of schooling, and dummy
variables for region (3), racial/Hispanic identification (2), and residence in an SMSA (2).
e. 1981 BMI is based on an average of weights reported in 1981 and 1982. 1988 BMI is based on
an average of weights reported in 1988 and 1989.
* p < 0.05;
# 0.05 < p < 0.10.

BMI at Ages 16 to 24 (1981)


Women Men
19-23 BMI 20-24 BMI
Minus 30+ BMI Minus <20 BMI
Outcomes at
Ages 23 to 31 ($) (% ) ($) (% )
Labor market 1,050 17.0 3,373 85.2
Marriage market 4,957 80.4 1,426 36.0
Other 162 2.6 -839 -21.2
Total adjusted income 6,169 100.0 3,960 100.0
difference

Table 5 OLS Regressions of 1988 Wages on 1981 Body Mass Index Categories, Women (dependent
variable = ln(hourly wage))

Cross Section Sister Differences


(1) (2) (3) (1) (2)
(3)
Body Mass Index, 1981
30+ -0.15 -0.16 -0.14 -0.12
-0.10 -0.07
(0.03) (0.03) (0.03) (0.12)
(0.12) (0.12)
24-29 -0.01 -0.01 -0.01 -0.03
-0.02 -0.02
(0.02) (0.02) (0.02) (0.05)
(0.05) (0.05)
<19 0.00 0.00 -0.01 -0.07
-0.05 -0.07
(0.02) (0.02) (0.02) (0.06)
(0.06) (0.06)
Schooling (years) 0.07 0.04 0.03 0.06
0.04 0.03
(0.00) (0.00) (0.00) (0.01)
(0.01) (0.01)
Experience 0.06 0.05 0.04 0.05
0.04 0.04
(0.00) (0.00) (0.00) (0.01)
(0.01) (0.01)
Ill health -0.05 -0.05
-0.01 -0.00
(0.03) (0.03)
(0.08) (0.08)
AFQT/100 0.38 0.32
0.49 0.47
(0.04) (0.04)
(0.13) (0.12)
Low self-esteem -0.02 -0.02
-0.03 -0.02
(0.01) (0.01)
(0.02) (0.02)
Married -0.00 -0.01
-0.03 -0.03
(0.02) (0.01)
(0.04) (0.04)
Divorced or separated 0.05 0.05
0.04 0.05
(0.03) (0.03)
(0.09) (0.09)
Kids (any) -0.10 -0.08
-0.06 -0.06
(0.02) (0.02)
(0.05) (0.05)
Kids * age of youngest/10 0.04 0.03
0.06 0.05
(0.03) (0.03)
(0.09) (0.09)
Occupational dummies? No No Yes No No
Yes
Adjusted R[sup2 0.30 0.33 0.37 0.10
0.12 0.20
Sample size 3,996 3,996 3,996 593 593
593

Notes: Standard errors in parentheses. Cross-section models also include dummy variables for
black and Hispanic identification, urban residence (2), and region (3). Sister-differenced models also
include urban and region controls.
Table 6 OLS Regressions of 1988 Family Income/Needs on 1981 Body Mass Index Categories,
Women (dependent variable = ln(income/needs))

Cross Section Sister Differences


(1) (2) (3) (1) (2)
(3)
Body Mass Index, 1981
30+ -0.25 -0.22 -0.19 -0.33
-0.29 -0.10
(0.06) (0.06) (0.05) (0.16)
(0.16) (0.15)
24-29 -0.14 -0.13 -0.09 -0.12
-0.11 -0.06
(0.03) (0.03) (0.03) (0.08)
(0.08) (0.07)
<19 0.03 0.05 0.03 0.02
0.03 0.02
(0.04) (0.04) (0.03) (0.09)
(0.08) (0.07)
Schooling (years) 0.14 0.09 0.06 0.10
0.08 0.06
(0.01) (0.01) (0.01) (0.02)
(0.02) (0.02)
Age 0.02 0.01 0.01 0.01
0.00 0.00
(0.01) (0.01) (0.01) (0.02)
(0.02) (0.02)
Ill health -0.19 -0.19
0.03 -0.08
(0.04) (0.04)
(0.11) (0.10)
AFQT/100 0.72 0.59
0.58 0.64
(0.06) (0.06)
(0.19) (0.17)
Low self-esteem -0.08 -0.06
-0.03 -0.03
(0.01) (0.01)
(0.03) (0.03)
Married 0.70
0.71
(0.02)
(0.02)
Divorced or separated -0.22
-0.13
(0.05)
(0.11)
Kids (any) -0.64
-0.54
(0.03)
(0.07)
Kids * age of youngest 0.02
0.02
(0.00)
(0.01)
Adjusted R[sup2 0.25 0.29 0.45 0.06
0.06 0.27
Sample size 4,188 4,188 4,188 609 609
609

Notes: Standard errors in parentheses. Cross-section models also include dummy variables for
black and Hispanic identification, urban residence (2), and region (3). Sister-differenced models also
include urban and region controls.
Table 7 Regression-Adjusted Percent or Percentage Point Differences in Marriage and Labor Market
Outcomes in 1988 According to Body Mass Index Category in 1988 or 1981, by Race
[Table Omitted]
Table A1 Regression-Adjusted Percent or Percentage Point Differences in Marriage and Labor
Market Outcomes in 1988-1990 (pooled data) According to Body Mass Index Category in 1988 or
1981

Income/Needs(FNa) Pr(Married)(FNb) Spouse's


Earnings(FNc) Hourly Wage(FNd)
SIB SIB
SIB SIB
XSEC DIFFS XSEC DIFFS
XSEC DIFFS XSEC DIFFS
Women
Full sample
1981 BMI(FNe)
30 + -0.21(FN*) -0.25(FN*) -0.19(FN*) -0.50(FN*)
-0.36(FN*) 0.26 -0.14(FN*) -0.20(FN*)
24-29 -0.09(FN*) -0.11(FN#) -0.08(FN*) -0.13
-0.11(FN*) -0.01 -0.03(FN#) 0.01
<19 0.02 0.05 0.03 0.04
0.08(FN*) 0.00 -0.02
N 4,859 802 5,048 299
2,707 288 4,449 698
Full sample
1988 BMI(FNe)
30 + -0.14(FN*) -0.20(FN*) -0.07(FN*) -0.04
-0.21(FN*) -0.02 -0.011(FN*) -0.10(FN#)
24-29 -0.07(FN*) 0.01 0.01 0.07
-0.03 -0.11 -0.04(FN*) -0.02
<19 -0.00 0.11 -0.02 -0.00
0.03 0.14 -0.00 0.07
N 4,859 802 5,048 299
2,707 288 4,449 698
Single & childless in 1982
1981 BMI(FNe)
30 + -0.31(FN*) -0.14 -0.28(FN*) -0.63(FN*)
-0.57(FN*) -0.30 -0.23(FN*) -0.21(FN#)
24-29 -0.10(FN*) -0.01 -0.13(FN*) -0.19(FN#)
-0.18(FN*) -0.12 -0.05(FN*) 0.05
<19 0.01 0.13 0.03 0.17(FN#)
0.16(FN*) -0.01 0.01 -0.01
N 2,554 332 2,663 149
1,342 122 2,444 319
Men
Full sample
1981 BMI(FNe)
30 + -0.02 -0.30(FN*) -0.05 -0.16
-0.14 0.19 -0.10(FN*) -0.19(FN*)
25-29 0.05(FN#) 0.04 0.03 -0.04
0.02 0.14 -0.01 -0.03
<20 -0.14(FN*) -0.15(FN*) -0.03 0.06
-0.18(FN*) -0.08 -0.08(FN*) -0.05
N 4,607 866 4,910 348
2,103 219 4,583 875
Full sample
1988 BMI(FNe)
30 + 0.02 -0.10 0.06(FN*) 0.13(FN#)
-0.03 0.02 -0.03(FN#) -0.04
25-29 0.07(FN*) 0.03 0.08(FN*) 0.15(FN*)
0.07(FN#) 0.18 0.02 0.01
<20 -0.30(FN*) -0.21(FN#) -0.07(FN#) -0.08
-0.04 0.12 -0.13(FN*) -0.00
N 4,607 866 4,910 348
2,103 219 4,583 875
Single & childless in 1982
1981 BMI(FNe)
30 + -0.00 -0.42(FN*) -0.06 0.25(FN#)
0.01 -0.04 -0.09(FN*) -0.20(FN*)
25-29 0.05(FN*) -0.01 0.01 -0.03
0.04 -0.13 -0.02 -0.06(FN#)
<20 -0.16(FN*) -0.14 -0.03 0.01
-0.16(FN*) 0.22 -0.09(FN*) -0.04
N 3,650 631 3,915 258
1,508 129 3,671 641

Notes: N is the number of observations, or (approximately) the number of pairs for same-sex
sibling differences.

FOOTNOTES
a. Dependent variable = average of In(income/needs) in 1987, 1988, and 1989, where needs are
defined as the poverty line for the family unit in the interview year. Regression controls include in
addition to dummy variables for BMI category: age, years of schooling, and dummy variables for
region (3), racial/Hispanic identification (2), and residence in an SMSA (2), all based on averages of
values from 1988-1990 interviews.
b. Dependent variable is dichotomous: "married at the time of the 1988, 1989, or 1990 interview."
Controls entered in the logit models include in addition to dummy variables for BMI category: age,
years of schooling, and dummy variables for region (3), racial/Hispanic identification (2), and
residence in an SMSA (2), all based on averages of values from 1988-1990 interviews. Numbers
shown are derivatives evaluated at the sample mean and may be interpreted as percentage point
differences in the probability of being married in at least one of the 1988-1990 interviews. Sibling
differences are from fixed-effects logit models.
c. Dependent variable = average of In(spouse's annual earnings) over the years 1987, 1988, and
1989 in which the respondent was married. The sample consists of sample members who were
married as of the 1988, 1989, or 1990 interview, and whose spouses earned at least $100 in any year
in which they were married. Regression controls include in addition to dummy variables for BMI
category: age, years of schooling, and dummy variables for region (3), racial/Hispanic identification
(2), and residence in an SMSA (2), all based on averages of values taken over the 1988-1990
interviews at which the respondent was married.
d. Dependent variable = average of In(sample member's hourly wage in survey week) 1988, 1989,
and 1990. The sample consists of sample members who worked for pay during the survey week.
Regression controls include in addition to dummy variables for BMI category: actual labor market
experience, years of schooling, and dummy variables for region (3), racial/Hispanic identification (2),
and residence in an SMSA (2), all based on averages of values taken over the 1988-1990 interviews.
e. 1981 BMI is based on an average of weights reported in 1981 and 1982. 1988 BMI is based on
an average of weights reported in 1988, 1989, and 1990.
* p < 0.05;
# 0.05 < p < 0.10.

FOOTNOTES
1. In their review, Autry et al. (p. 537) report a prevalence of 5 percent of college and high
school females and 1.5 percent of males but even these rates may be too high. Drewnoski, Hopkins,
and Kessler (1988) report a prevalence of 1 percent of college females and 0.2 percent of college
males in a nationally representative sample of college students. The prevalence among female
undergraduates living in group quarters on campus, the group at highest risk, was 2.2 percent.
2. There are Development Economics and Economic History literatures on weight and stature. Short
stature and low weight-for-height are indicators of nutritional deficits; height and weight, therefore,
may be interpreted as indicators of the economic status of populations. For an excellent review, see
Steckel (1991).
3. For example, Autry et al. call for research to address "Psychological factors that influence the
development and maintenance of anorexia nervosa and bulimia" and "Genetic, environmental, and
psychosocial studies that might elucidate why the phenomena are more prevalent among females" (p.
541).
4. For example, Sciacca et al. (1991) find in a survey of university students that, although 17
percent of women and 20 percent of men were above "normal" body weight, 40 percent of women and
24 percent of men considered themselves overweight. In addition, 53 percent of women and 20
percent of men reported experiencing a fair amount or great deal of discomfort from being
overweight. Sciacca et al. recommend that efforts should be made to help students who are not over
recommended weight-for-height, but who consider themselves overweight, to change perceptions or
expectations about their bodies (p. 167). While changing self-perception may be an important step in
combating eating disorders, we must recognize the possibility that such students do not have
"distorted body images," but rather, the discrepancy between recommended weight and self-perceived
overweight may reflect an accurate perception of the social norms surrounding body weight. Put
differently, there is no reason to think that social norms should conform to recommended weights that
are, after all, based on mortality risks.
5. They also note the need to repeat the analysis for an older sample (p. 139), presumably because
the wages of young workers are highly variable (for example, 5 percent of males and 40 percent of
females in their sample were enrolled in school). Other indications of the need for reanalysis are the
small magnitude and lack of significance of several standard wage equation coefficients. For example,
in the male wage equation, coefficients of black racial identification, union status, education, age,
health status, marital status, and labor market experience are small and insignificant. Whether this
result is due to the selectivity correction cannot be determined from the information presented.
6. We use the term "obese" to refer to persons with Body Mass Indexes of 30 or more, and the
term "overweight" for persons with BMIs between 24 and 29 for women, or 25 to 29 for men.
Gortmaker et al. use the term "overweight" to refer to persons above the 95th percentile of NCHS
standards of weight for height, age, and sex. Since this group corresponds closely to the group we
refer to as "obese,"
7. Of 11,602 potential respondents, 10,465 or 90.2 percent were interviewed in 1988 (CHRR 1992).
Of these, we dropped 72 due to missing wages or hourly wages less than one dollar or greater than
100 dollars per hour, and 352 due to missing height information.
8. Among those who appeared to lose more than two inches in height, we identify two types: those
who lost more than two inches between 1982 and 1985 (Group A) and those who lost more than two
inches between 1981 and 1982 (Group B). For Group A we imputed heights as follows. For those who
lost less than two inches from 1981 to 1985, height in 1981, 1982, and 1985 equals the average of
the 1981 and 1985 heights. For those who lost less than two inches from 1981 to 1982, all three
heights were set to the average of the 1981 and 1982 heights. Although everyone in Group A lost
more than two inches in height from 1982 to 1985, if they grew at all between 1981 and 1985, then
1982 height was set to a weighted average of the 1981 and 1985 heights. For Group A individuals who
lost more than two inches from 1981 to 1985, all three heights were set equal to the mean of the
1981 and 1982 heights.
Case B individuals were handled in the following manner. First, if they lost less than two inches or
grew between 1982 and 1985, then we extrapolated linearly to impute the 1981 value. If they lost
more than two inches between 1981 and 1982, then all three heights were set to the mean of the
three of the reported height measures.
9. We get similar results when we use family income as the dependent variable rather than the
income/needs ratio.
10. The distributions (not shown) by BMI in 1982 and 1988 for the subsample of persons who were
single and childless in 1982 are remarkably similar to those for the entire sample.
11. Same-sex sibling differences are computed as differences from within-family means. This
procedure is equivalent to entering a dummy variable for each family of origin (for example, Greene
[1993], pp. 466-69). One differenced observation per family is dropped.
12. For fixed-effects logits, sister pairs contribute to the likelihood function only if sisters differ with
respect to the outcome (namely, one is married in 1988 and one is not).
13. Estimates of obesity differences from cross-sectional models for the pooled sibling subsample
are similar to (or smaller than) those for the full sample.
14. Appendix 1 presents a parallel analysis to Table 4, where we have averaged data from all
available interview years 1988-1990 for each sample member interviewed in 1988. The cross-
sectional results are similar to those in Table 4, whereas results from sibling comparisons are mostly
strengthened. For example, the difference in marriage probabilities between an obese woman and her
sister of recommended weight increases from - .23 to - .50; the difference in wages rises from - .12
to - .20. One exception to this pattern is that the sister difference of spouse's earnings conditional on
being married changes signs from - .25 to + .26; however, this coefficient is not precisely estimated
in either model. The standard error of the coefficient estimate in Appendix 1 is 0.30. For men, cross-
sectional estimates are similar to those in Table 4; however, there is more evidence of an obesity
penalty for men in sibling comparisons when the three years of data are pooled.
15. These figures were computed as follows. For each respondent, total family income is partitioned
into three components: own earnings, spouse's earnings, and other income. Unmarried persons were
assigned 0 for spouse's earnings. Persons with no annual "own" earnings were assigned 0 for own
earnings. We ran four regressions each for men and women (one of the four is redundant); dependent
variables are total family income, own earnings, spouse's earnings, and other income. Controls were
those included in the models summarized in Table 4, except we substituted age for actual experience
in the earnings models. The numbers reported in the text tables are the coefficients of the obese
dummy variables (for women) and the underweight dummy variables (for men).
16. Models for men are available from the authors.
17. One version of this hypothesis holds that obesity effects are confounded by unmeasured self-
esteem, amounting to a suggestion that heavier persons earn lower wages due to low self-esteem and
not weight per se. However, self-esteem may be affected by social treatment, income, or wages.
Therefore, a second version holds that the low economic status of obese women may be accounted for
by low self-esteem, but views low self-esteem to be primarily the result of obesity (for example, see
Gortmaker et al. [1993], who find no effects of obesity on an index of self-esteem).
18. Table 2 also suggests that obese women are slightly less likely to be employed than women of
recommended weight. However, we estimated logit models of labor force participation which included
BMI controls and demographic characteristics and found that BMI was not a significant determinant of
labor force participation.
19. There is also some evidence in Appendix Table A1 of wage discrimination against underweight
and overweight men.
20. For a more detailed (albeit speculative) discussion, see Averett and Korenman (1993).

REFERENCES
Abraham, S. 1983. "Obese and Overweight Adults in the United States." Vital and Health
Statistics. Series 11, Number 230. DHHS Pub. No. 83-1680. Public Health Service. National Center for
Health Statistics. Washington, D.C.: GPO.
Armstrong, Larry, and Maria Mallory. 1992. "The Diet Business Starts Sweating." Business Week
22:32-33.
Autry, Joseph H., Ellen S. Stover, Natalie Reatig, and Regina Casper. 1986. "Anorexia Nervosa and
Bulimia." Annual Review of Public Health 7:535-43.
Averett, Susan, and Sanders Korenman. 1993. "The Economic Reality of The Beauty Myth."
National Bureau of Economic Research Working Paper Number 4521. Cambridge, Mass.
Bray, George A. 1978. "Definition, Measurement, and Classification of the Syndromes of Obesity."
International Journal of Obesity 2(2):99-112.
Bray, George A. 1979. "Obesity in America." International Journal of Obesity 3(4):363-75.
Chamberlain, Gary. 1980. "Analysis of Covariance with Qualitative Data." Review of Economic
Studies 47(1):225-38.
CHRR (Center for Human Resource Research). 1992. NLS Handbook 1992. Columbus: The Ohio
State University.
Coleman, Jennifer A. 1993. "Discrimination at Large." Newsweek August 2: 9.
Dawson, Deborah A. 1988. "Ethnic Differences in Female Overweight: Data from the 1985 National
Health Interview Survey." American Journal of Public Health 78(10):1326-29.
Drewnowski, Adam, Stephen A. Hopkins, and Ronald C. Kessler. 1988. "The Prevalence of Bulimia
Nervosa in the U.S. College Student Population." American Journal of Public Health 78(10):1322-25.
Filer, Randall. 1993. "The Usefulness of Predicted Values for Prior Work Experience in Analyzing
Labor Market Outcomes for Women." Journal of Human Resources 28(3):518-37.
Furnham, Adrian, and Naznin Alibhai. 1983. "Cross-Cultural Differences in the Perception of Female
Body Shapes." Psychological Medicine 13(4):829-37.
Gortmaker, Steven L., Aviva Must, James M. Perrin, Arthur M. Sobol, and William H. Dietz. 1993.
"Social and Economic Consequences of Overweight in Adolescence and Young Adulthood." New
England Journal of Medicine 329(14):1008-12.
Greene, William. 1993. Econometric Analysis. 2nd ed. New York: Macmillan.
Greenwood, M. R. C. 1983. Obesity. New York: Churchill Livingstone.
Hamermesh, Daniel S., and Jeff E. Biddle. 1993. "Beauty and the Labor Market." American
Economic Review 84(5):1174-94.
Johnson, William G., and James Lambrinos. 1985. "Wage Discrimination Against Handicapped
Men and Women." Journal of Human Resources 20(2):264-77.
Kalmijn, Matthijs. 1993. "Trends in Black/White Intermarriage." Social Forces 72(1):119-46.
Kannel, William B. 1983. "Health and Obesity: An Overview." In Health and Obesity, ed. H. L.
Conn, Jr., E. A. DeFelice and P. Kuo, 1-19. New York: Raven Press.
Kolata, Gina, Jane E. Brody, and Elisabeth Rosenthal. 1992. "Fat in America." The New York Times
November 22-24, v. 142.
Lampert, Leslie. 1993. "Fat Like Me." Ladies Home Journal, May, pp. 154-57.
Loh, Eng Seng. 1993. The Economic Effects of Physical Appearance." Social Science Quarterly
74(2):420-38.
Metropolitan Life Insurance Company. 1983. "1983 Metropolitan Height and Weight Tables."
Statistical Bulletin 64(1):1-9.
Mincer, Jacob and Solomon Polachek. 1974. "Family Investments in Human Capital: Earnings of
Women." In Economics of the Family, ed. T. W. Schultz, 397-429. Chicago: The University of Chicago
Press.
Register, Charles A., and David R. Williams. 1990. "Wage Effects of Obesity among Young
Workers." Social Science Quarterly 71(1):130-41.
Sargent, James D., and David G. Blanchflower. 1994. "Obesity and Stature in Adolescence and
Earnings in Young Adulthood." Archives of Pediatrics and Adolescent Medicine. Forthcoming.
Sciacca, John P., Christopher L. Melby, Gerald C. Hyner, Amy C. Brown, and Paul L. Femea. 1991.
"Body Mass Index and Perceived Weight Status in Young Adults." Journal of Community Health
16(3):159-68.
Steckel, Richard H. 1991. "Stature and Living Standards in the United States." National Bureau of
Economic Research Working Paper on Historical Factors in Long-Run Growth, Number 24, April.
Stunkard, Albert J., and T. I. A. Sorensen. 1993. "Obesity and Socioeconomic Status: A Complex
Relation." New England Journal of Medicine 329(14):1036-37.
Wolf, Naomi. 1991. The Beauty Myth. New York: Doubleday.
World Bank, 1992. World Development Report. Oxford: Oxford University Press.
Zuckerman, Diana, Anne Colby, Norma C. Ware, and Judith S. Lazerson. 1986. "The Prevalence of
Bulimia among College Students." American Journal of Public Health 76(9):1135-37.

WBN: 9610602323002