Big and Beautiful? Evidence of Racial Differencesin the Perceived Attractiveness of Obese Female Adolescents
Frank W. Heiland[1]
Mir M. Ali[2]
April 26, 2010
Abstract
This paper investigates the relationship between body weight, race, and interviewer-rated attractiveness using large representative samples of female adolescents. Accounting for interviewer fixed effects and an extensive list of controls for individual and parental background characteristics, we find that obese white female adolescents are 36% less likely to be perceived as physically attractive compared to their non-obese counterparts. Obese black teenagers, on the other hand, are only 28% less likely to be rated physically attractive compared to non-obese black female adolescents. Similarly, we observe that the penalty associated with high BMI is 30% smaller for black female adolescents compared to whites. The findings indicate that the aesthetic judgment of female body size depends importantly on the race of the subject but varies little by rater characteristics in our sample. We also find that obesity and high BMI are more strongly negatively associated with having an attractive personality among white girls than among black girls. On the other hand, we also provide some evidence that elevated BMI is more strongly associated with physical maturity among black girls. We discuss these findings in the context of the literatures on race-specific obesity penalties and socio-cultural body size norms.
Keywords:Obesity, Body Mass Index (BMI), Beauty, Female Body Image, Race
JEL Classification: J7, Z1
1. Introduction
A person’s weight (given height)is a prominent aspect of appearance, and as such it is subject to aesthetic assessment.Physical attractiveness is an asset that is valuable in many situations of human interactions such as in the labormarket (Hamermesh and Biddle, 1994; Mobius and Rosenblat, 2006; Fletcher, 2009),in (romantic)relationships(Carmalt et al., 2008; Cawley et al., 2006; Pearce et al. 2002), and more broadly in situations of face-to-face interaction and exchange (Mulford et al., 1998).How individuals’ aesthetic judgementsare linked to weight statusmay depend on a complex set of influences on preferences, includingsocio-cultural factors such as ideal body sizes (“beauty norms”)and the stigmatization of overweight(“obesity penalty”) or underweight.
Discrimination on the basis of elevated body weight is widely documented among women and takes various forms, including social exclusion, mistreatment, and public ridicule.[3]The literaturesuggests that discriminationcontributes to the observed wage, job-status, and marriage-market penalties associated with obesityamong women (see, for example, Averett and Korenman, 1996 and 1999; Cawley, 2004; Baum and Ford, 2004; Conley and Glauber, 2005; Morris, 2007; and Mukhopadhyay, 2008). Adolescents are particularly preoccupied with body image and thus are especially vulnerable to discrimination on the basis of appearance. Ideals of thinness presented in the media and peer pressureshave been linked to eating disorders among girls(Field et al. 1999, Lieberman et al. 2001, Wang et al. 2006, Salvy et al. 2007).[4]
A number of findings from recent studies suggest thatfemale obesity and high BMI are penalizeddifferently across racial groups.For example, Cawley (2004) documents significant wage penalties associated with obesity among white women, but not among black women. Averett and Korenman (1996, 1999) report significant negative effects of obesity on white women’s marriage prospects and spousal income, but little to no effect of obesity on black women’s wages or marriage prospects. Heiland and Ali (2010) find that obese white girls, but not obese black girls, are significantly less likely to be in a romantic relationship and have sexual experiences compared to their non-obese counterparts.One explanationprovided for these patterns is the presence of socio-cultural differences in ideal body sizes.
This paper examines the relationship between female adolescents’ obesity (high BMI) and their attractiveness in appearance and personality as judged by others.Our investigationis guided by four(nested) hypotheses regarding therelationship between body size and attractiveness rating: (i) it is universal, (ii) it differs between raters but is universal for a given rater, (iii) it is universal among raters but depends on characteristics of the person rated, and (vi) it differs between raters and by characteristics of the person rated.We conduct a series of empirical tests of the relationship between obesity, Body Mass Index (BMI), race, and interviewer-rated attractiveness empirically, using large nationally representative samples of white and African American (henceforth “black”) female adolescentsfrom the National Longitudinal Survey of Adolescent Health (henceforth “Add Health”). We examine the possibility that the relationship may be interviewer-specific - as hypothesized in (ii) and (vi). Obese (high BMI) girls are found to be less likely to be perceived asphysically attractive compared to their non-obese counterparts. The BMI/obesity penalty is greater for white girls than for black girls and robust to interviewer fixed effects (and mediating factors), consistent with hypothesis (iii) that the aesthetic judgment of female body size depends importantly on the race of the subject but varies little by rater characteristics in our sample.
2. Data
Study Samples
Add Health surveyed adolescents in 132 schools in grades 7 to 12 nationwide, oversampling disadvantaged children. The in-school portion of wave I (1994)contains about 90,000 adolescents. A subset of the initial sample (20,745 respondents in 1994) was also interviewed in their homes with follow-up surveys in wave II (1996) (about 15,000 respondents). One parent (mostly mothers) was interviewed as part of the in-home survey in 1994. Our sample is drawn from wave II female respondents in grade 7 to 12 under the age of 19, with records linked to parental information from wave I. While Add Health did not collect interviewer characteristics in the adolescent surveys (wave I and II), the ID of the interviewer is available for almost all respondents (it is missing for 4 respondents). Our final sample size is 5,947, comprised of 4,348 non-Hispanic white females and 1,599 non-Hispanic black females.These respondents were interviewed by 338 different interviewers. Among those interviewers 193 (57.7%) interviewed both black and white girls, covering 3,908 respondents or about two thirds of our sample. In other words the majority of respondents (2,522 white and 1,386 black girls) were rated by interviewers who faced a racially diverse group of respondents.
Interviewer Assessments
Add Health required the interviewers to describe the respondent after leaving the individual's home. Interviewers werefirst asked to grade attractiveness of appearance then attractiveness of personality on a scale ranging from “very unattractive,”“unattractive,”“about average,”“attractive” to“very attractive.” We created dummies of attractive physique and personality based on the responses from the top two categories. The interviewer was also asked to comment on how well groomed the respondent was. We coded “well groomed” and “very well groomed” to indicate that the respondent appeared neat. Interviewersalso graded physical maturityof the respondent compared with other adolescents of her age.We generated a dummy coding the respondent as physically mature if the interviewerrated“mature” or “very mature.” Interviewers were also asked to rate how candid the respondent was and we coded the respondent as being candid if the interviewer's response was “very candid” and “moderately candid.”
Explanatory Variables
The main explanatory variables of interest are BMI and whether the person is obese. We use interviewer measured height and weight to construct BMI [weight(kg)/height(m2)]. Obesityis defined according to the CDC classification system for adolescent girls.It is important to note that BMI/obesity are general measures of “largeness”that do not directly speak to the distribution of weight and tissue (waist-to-hip ratios and body fat are unavailable in Add Health).
We analyze robustness with respect to: (i) respondent demographics(age dummies, first-born, siblings, US-born),(ii) parental/family characteristics (education, family on welfare, respondent lives with biological parents, age moved to residence, moved for school district), and (iii) respondent physical/mental health and development(self reported in good health,exercised at least three or more times in past 7 days, plays an active sport,experienced menarche, RSE self-esteem scale (see Rosenberg, 1965), CES-D depression scale (see Radloff, 1977)).
Sample Descriptives
Table 1 provides sample averagesof the measures used in the analysis by race and obesity status. The black and the white subsamples are comparable in many aspects (e.g., age distribution, menarche).White girls are also less likely to be classified as obese (9.4% white vs. 15.8% black) and have lower BMI values (22.6 vs. 24.0). Consistent with this white girls are more likely to participate in an active sport (35.5% vs. 22.6%). Black girls are more likely to reside in an economically disadvantaged household without both biological parents. Important differences by weight status in the raw data will be discussed as part of the results in Section 3.
3. Main Results
Table 1 shows that 24.3% of obese white girls are rated physically attractive compared to 61.9% of non-obese white girls, suggesting an obesity penalty of 37.6 percentage points for whites. Among black girls this difference is only 28.6 percentage points, as 26.1% of black obese girls are rated attractive compared to 54.6% among non-obese black girls. If all interviewers appliedthe same weight-attractiveness assessment to all respondents regardless of respondents’ racewe would expect lower average attractiveness ratings among the black obese girls in our sample compared to the white obese girls since the former have higher BMI, on average. The fact that average attractiveness is greater among the non-obese white girls compared to the non-obese black girls could be taken as evidence of a thinness premium among white girls, however, we note that the non-obese white girls also tend to be thinner, on average, than their black counterparts.
This simple comparison of means provides evidence against the presence of a universal range of attractive body sizesamong the interviewers and supports the idea that there are race-specific obesity penalties. However, the pattern in the raw data may be an artifact of interviewer-specificityin the way body weight is linked to attractiveness combined with non-random assignment of interviewers to respondents. For example, the observed pattern could emerge if black interviewers perceive a wider range of female body sizes as attractive, on average, regardless of the race (or other characteristics) of the girls, and are also more likely to interview black respondents.While no interviewer characteristics were collected in the Add Health for wave I and II, we know which interviewer was assigned to a respondent. We estimate multivariate models with interviewer fixed effects, individual and parental controls, and school fixed effects in pooled (Tables 2 and 3) and split samples by race (Table 4). If interviewers differed in their aesthetic judgementof BMI/obesity based on their own characteristics but not based on respondent characteristics, and are not randomly assigned, we would expect any race-specificity in the estimated effect of BMI/obesity to disappear in the split samples by race when interviewer effects are controlled for.[5]
The analysis of the pooled sample shown in Tables2 and 3 uses an interaction between black and BMI/obesity which provides a familiar setup to test for race-specific effects of body size on aesthetic and personality assessment. The drawback of this empirical design is that the interviewer fixed effects here do not capture differences in interviewer judgments as it pertains to the race. Table 4 provides estimates from the black and white subsamples. While it is more difficult to see race-specificity of effects in this setup, it is more general in the sense that it allows al marginal effects to be race-specific. Most importantly, this design facilitates testing for the role of interviewer characteristics in assessing attractiveness among black vs. white women.
Model 1, reports the BMI/obesity coefficients estimated from linear models regressing each of the five interviewer-assessments on BMI/obesity and the set of age dummies. Model 2, reports the BMI/obesity coefficients estimated from linear models regressing each of the five interviewer-assessments on BMI/obesity, age, and interviewer fixed effects. Comparing the results from Model 2 to Model 1 in Table 2, we find the estimated negative effect of being obese on being perceived as physically attractive is significantly smaller for black girls.Similar robust results are provided by the split sample analysis. This rejects the idea that interviewers assess beauty independently of the race of the respondent. The results confirm the pattern found when comparing means in the raw data, that obese girls are less likely to be rated physically attractive and thatthis obesity penalty is significantly smaller for black girls than for white girls.
According to Model 3 in Table 2, which also includes extensive controls for individual and parental characteristics,white obese girls are 37.1% less likely to be considered physically attractive compared to white non-obese girls. For black girls this figure is only 30%. As shown in Table 4, similar differences are found using BMI, consistent with the idea that the race-specificity found for obesity is not the result of higher BMI values among obese white girls.
Model 4 regresses physical attractiveness on the full set of controls and the four interviewer assessments. The estimated associations are smaller (by about 8 percentage points for obesity) but remain highly significant and race-specific, suggesting that personality (likeability) and attention to appearance may be mediating the impact of BMI/obesity on beauty perception in general, but, together with a potentially race-specific interpretation of high body mass as physical maturity, do not fully explain the race-specificity of the relationship.
Tables 2,3, and 4also documentBMI/obesity penalties for being perceived as “well groomed” and having an “attractive personality.”The estimated coefficients of BMI/obesity tend to be smalleramong black girlsbut the differences are typically not statistically significant. The split sample results (Table 4) suggest that BMI/obesity is associated with physical maturity among black girls (controlling for menarche). No relationship is found between BMI/obesity and being candid for white girls, but there is some evidence that obese (high BMI) black girls are less likely to be viewed as candid.
4.Discussion and Conclusion
Using nationally representative data on adolescents from Add Health, we document that obesity (high BMI) is associated with lower physical and personality attractiveness ratings for female adolescents. We present evidence of the race-specificity of these penalties such that the estimated attractiveness penaltiesare lower for black female adolescents than for white female adolescents. Our findings contribute to an emerging multi-disciplinary literature on obesity stigma, cultural differences in ideal body size, and related identity prescriptions that suggests less stigmatization of overweight and a higher ideal body size among black females. This evidence suggests that discrimination based on appearance may explain the race-specificity of the penalties associated with high BMI/obesity in labor, marriage, and adolescent relationship marketsdocumented in a number of recent studies (Cawley, 2004; Averett and Korenman, 1996, 1999;Heiland and Ali,2010; among others).
The evidence presented here is consistent with the presence of a broader range of body sizes considered attractive for black femalescompared to white females. While we do not observe interviewer characteristics, the fact that the estimated impact of BMI/obesity on beauty assessment is remarkably robust to the inclusion of interviewer fixed effects and other controls suggests that the influence of interviewer characteristics on beauty assessment may be very limited.In other words, our evidence is consistent with the idea that weight generally plays less of a role in female beauty among blacks than whites.This may reflecta wider range of body sizes considered attractive among blacks and/or a larger black female body ideal. Black women have been found to be less likely than women in other racial/ethnic groups to perceive themselves as overweight, even after controlling for objective weight status (see Dawson, 1988; Molloy and Herzberger, 1998; Burke and Heiland, 2008; Burke et al., 2010).[6] Cusumano and Thompson’s (1997) analysis of the body sizes of photographic images in women’s magazines found that Essence, a magazine targeting black readers, had the largest average image size (“body shape rating”).In addition, using weight self-perception data from a nationally representative sample of women,Burke and Heiland (2008) infer an ideal BMI for black women that is three points higher than forwhitewomen.
To the extent that obese black girls face weaker incentives to desire weight loss, the effectiveness of public health campaigns aimed at weight reduction may be more limited for black girls (see also Stevens et al., 1994). Further, identity prescriptions within the black community traditionally cast women as selfless nurturers, prescriptions that have resulted in black women’s neglecting their own health needs (see hooks 1981, Beauboeuf-LaFontant 2003, and Lynch et al. 2007). On the other hand, there may be health benefits associated with improved body image, such as higher self-esteem and a decline in the incidence of eating disorders (e.g., Rucker and Cash, 1992). More fundamentally, recent research questions the usefulness of the BMI classification system, especially in the context of race/ethnicity, and recommends alternative criteria for obesity that rely on the distribution of weight and body fat (see e.g., Cawley and Burkhauser, 2006; Burke and Heiland, 2008).