Body composition and wages

Economics and Human Biology 8 (2010) 242–254

Contents lists available at ScienceDirect

Economics and Human Biology journal homepage: http://www.elsevier.com/locate/ehb

Body composition and wages Roy Wada a,*, Erdal Tekin b a

UCLA School of Public Health, Box 951772, Los Angeles, CA 90095-1772, United States Georgia State University, NBER, and IZA, Department of Economics, Andrew Young School of Policy Studies, Georgia State University, P.O. Box 3992, Atlanta, GA 30302-3992, United States b

A R T I C L E I N F O

A B S T R A C T

JEL classification: I1 J3

This paper examines the relationship between body composition and wages in the United States. We develop measures of body composition – body fat (BF) and fat-free mass (FFM) – using data on bioelectrical impedance analysis (BIA) that are available in the National Health and Nutrition Examination Survey III and estimate wage models for respondents in the National Longitudinal Survey of Youth 1979. Previous research uses body size or BMI as measures of obesity despite a growing concern that they do not distinguish between body fat and fat-free body mass or adequately control for non-homogeneity inside the human body. Therefore, measures presented in this paper represent a useful alternative to BMIbased proxies of obesity. Our results indicate that BF is associated with decreased wages for both males and females among whites and blacks. We also present evidence suggesting that FFM is associated with increased wages. We show that these results are not the artifacts of unobserved heterogeneity. Finally, our findings are robust to numerous specification checks and to a large number of alternative BIA prediction equations from which the body composition measures are derived. ß 2010 Elsevier B.V. All rights reserved.

Keywords: BMI Body composition Body fat Fat-free mass Bodyweight Obesity Wages

1. Introduction Obesity is defined as the presence of excessive body fat (Bjorntorp, 2002; World Health Organization, 1998). A person is classified as obese based on having an excessive level of body fat and it is the excessive levels of fat that have been shown to be associated with a number of adverse health conditions.1 Therefore, an appropriate measure of obesity should be one that closely measures the level of body fat inside a human’s body if one wishes to better understand the potentially harmful effects of obesity on an array of economic and social outcomes, ranging from labor market performance to self-esteem, discrimination, and marriage

* Corresponding author. E-mail addresses: [email protected] (R. Wada), [email protected] (E. Tekin). 1 Obesity-related diseases include coronary heart disease, type 2 diabetes, hypertensions, stroke, cancers, and liver and gallbladder diseases (Centers for Disease Control and Prevention, 2007). 1570-677X/$ – see front matter ß 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.ehb.2010.02.001

problems. However, body fat is not directly observable to researchers in most research settings. This is because most individuals do not know how much body fat they possess and methods to measure it typically require clinical instruments that are hardly available or expensive, especially for survey purposes. Most social surveys have instead relied on a number of proxies for measuring obesity. The most commonly used one is the body mass index (BMI), which is calculated as weight in kilograms divided by height in meters squared. One clear advantage of BMI is that it is easily calculated from height and weight, which are readily available in most surveys. Despite its convenience, BMI is increasingly known as an imperfect surrogate for body fat (Smalley et al., 1990; Gallagher et al., 1996; Romero-Corral et al., 2006). BMI alone explains only 26% of the variations in body fat and a wide a range of conditions exists in which BMI provides misleading information about the levels of body fat (Gallagher et al., 1996; Prentice and Jebb, 2001). In a recent review of the medical literature on the association between BMI and total

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

mortality for patients with coronary artery disease, RomeroCorral et al. (2006) found that overweight patients actually had a better survival rate and lower cardiovascular events than did underweight patients and that obese patients had no increased risk for mortality. Also referred to as ‘‘the obesity paradox’’ (Curtis et al., 2005), an absence of association (or an inverse association) between obesity and mortality in patients with cardiovascular disease has puzzled many medical researchers.2 Economists who have recently studied the effect of obesity on the labor market outcomes using BMI have also obtained mixed findings.3 While BMI is widely used by social scientists, a number of researchers have recently suggested that it might be the inability of BMI to properly distinguish body fat from nonfat body components that is responsible for some of the puzzling findings described above (e.g., Allison et al., 2002; Romero-Corral et al., 2006; Wada, 2007; Burkhauser and Cawley, 2008). Gallagher et al. (1996) and De Lorenzo et al. (2003) suggested that the reliability of BMI for measuring body fat was questionable, and that direct measurements of body fat could provide a significant improvement towards detection and diagnosis of obesity. Furthermore, a consensus report by World Health Organization (1995) warned researchers that BMI must be interpreted carefully to avoid confusing muscularity with obesity. Direct measurement of body fat has long been available through the analyses of body composition, in which a human body is analytically broken down into its various components (Heyward and Wagner, 2004). One popular approach is the two-compartment model of body fat (BF) and fat-free mass (FFM).4 In this model, BF is the smaller component consisting mostly of fat tissues, while FFM is the larger component that includes everything else, including muscles and skeletons. One advantage of the two-compartment model is that body fat can be tracked independently from the rest of human body. Using body composition, it has been clinically shown that BF is associated with the ill effects of obesity, while FFM is associated with health and physical fitness (e.g., Heitmann et al., 2000; Allison et al., 2002). Therefore, the marginal effects of BF can be interpreted as the incremental effect of obesity, while the marginal effect

2 Similar associations are also reported by several other studies that examine the association between BMI and mortality in patients without evidence of a cardiovascular disease (Flegal et al., 2005; McGee, 2005). 3 There is some consensus on a negative association between obesity and wages for white females. However, there is no agreement on the existence of a wage penalty for males or other female population groups. Averett and Korenman (1996), Baum and Ford (2004), and Cawley (2004) have all found a negative association between BMI and wages for white females but not for males or non-white females. For white males, the effect of BMI on wages was found to be non-linear, with overweight workers earning more than underweight or obese workers. Cawley (2004) found that BMI is positively and significantly associated with the wages of black males in some specifications. Baum and Ford (2004) report a weak penalty for male obesity, but the result becomes mixed when the sample is further divided by ethnicity. 4 The two-compartment models were first proposed by Siri (1961) and Brozek et al. (1963). FFM is sometimes referred to as lean body mass (Heyward and Wagner, 2004). Technically, lean body mass contains a small amount of lipids, while FFM does not any lipids at all. In males, about 97% of lean body mass is FFM, while it is about 92% in females (Lohman, 1992).

243

of FFM can be considered as the effect of healthy body growth. We argue that a single index like BMI may not capture the complex influence that BF and FFM can exert on the economic and social outcomes. Because the expected effects of BF and FFM are likely to be opposite to each other, they may result in a situation where these opposing effects cancel each other out. In this paper, we draw data from the National Health and Nutrition Examination Survey 1988–1994 (the NHANES III) and the National Longitudinal Survey of Youth 1979 (NLSY) to estimate the relationship between body composition and the wages. We aim to make three contributions to the literature. First, we develop and propose using body composition measures as an alternative to the BMI-based proxies of obesity. Second, we empirically demonstrate that the calculated body fat is associated with decreases in wages for both males and females among whites and blacks. Third, we show that calculated fat-free mass is associated with increased wages. This result draws attention to the beneficial effect of healthy body growth on wages. Since health is one of the conduits through which body size is hypothesized to influence worker productivity, it should be the healthy body component or the fat-free mass that is associated with worker productivity. This finding also lends support to the nutrition hypothesis, which posits a positive relationship between body size and worker productivity. In his Nobel address, Fogel (1994) stated that increased body size should be associated with increased worker productivity.5 This assumption of ‘‘bigger-is-better’’ has been questioned in light of the obesity epidemic.6 2. Body composition and empirical strategy Body composition has long been used by nutritionists and physiologists for the purpose of studying nutrition, physical growth, and physical performance (Forbes, 1999). However, the difficulty of obtaining body composition compared to BMI has been an obstacle to its adoption by social science researchers. However, this obstacle has been mitigated by the development of bioelectrical impedance analysis (BIA).7 In the BIA, the electrical resistance of an individual’s body is measured and then converted into BF and FFM using a predetermined prediction equation based on the principle that FFM (with higher water content) registers a lower electrical resistance than BF (National Institutes of Health, 1994). In an NIH conference in 1994, it

5 See Fogel (1994) and Steckel (1995) for a summary of the nutrition hypothesis. 6 Behrman and Rosenzweig (2001) explore the possibility that the negative effect of obesity is due to unobserved heterogeneity and not necessarily due to increased body size. Fogel (1994) presents his theory that the beneficial effect of body size is not properly captured by the observed relationship between BMI and mortality, which is U-shaped. 7 See Kushner et al. (1990), Roubenoff et al. (1995), Sun et al. (2003), and Chumlea et al. (2002) for more on the BIA. Other alternatives to body composition include skinfold thickness, underwater weighting, dual Xray absorptiometry, magnetic resonance imaging (Heshka et al., 1994; Heymsfield et al., 1998). However, compared to BIA, most of these methods are prohibitively expensive or intrusive for use in large-scale epidemiological studies.

244

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

was stated that BIA was a useful technique for determining body composition in healthy individuals (National Institutes of Health, 1994).8 The construction of body composition from BIA has recently caught the attention of economists. For example, Wada (2005, 2007) took a departure from the other economic studies of obesity by using BIA information available in the National Health and Nutrition Examination Survey (the NHANES IIII) to estimate the effect of body composition on labor market outcomes. Burkhauser and Cawley (2008) also utilized the BIA information available in the NHANES III to study the effect of body composition on employment. Recently, Johansson et al. (2009) used a Finnish data set containing information on body composition to examine the relationship between obesity and wages. In this paper, we also use BIA information available in the NHANES III, which is a nationally representative crosssectional survey conducted between 1988 and 1994. In the NHANES III, trained technicians in mobile laboratories obtained the necessary BIA information from respondents over the age of 12 who were not known to be physically handicapped or pregnant at the time.9 The BIA information from the NHANES III can be converted into body composition using predictive equations developed by a number of clinical researchers. We start our analysis by using those equations developed by Sun et al. (2003). It is important to note that we start with these particular equations for illustrative purposes as well as for the following reasons. First, these equations were developed and published in anticipation of others using them with data from the NHANES III. Second, these equations come from one of the most recently published studies on the subject. Third, most of the other studies using the BIA information (e.g., Wada, 2005, 2007; Burkhauser and Cawley, 2008) also relied on these equations.10 However, we recognize that relying only on a single set of prediction equations may raise the legitimate question as to whether our results are driven by the choice of these particular equations. To address this concern, we gathered a comprehensive set of prediction equations from published sources developed by other clinical researchers. In this paper, we also present results derived from these equations as a robustness check. Sun et al. (2003) developed their equation using a sample containing 1474 whites and 355 blacks aged 12– 94. Predictive equations developed using mostly white samples are known to be significantly less accurate for

8 The conference also stressed that the National Health and Nutrition Examination Survey (NHANES IIII), which contains measurements of BIA for a nationally representative sample, is promising for examining the relationship between body composition and clinical risk factors. 9 For more information on the sample design of NHANES III, see U.S. Department of Health and Human Services (1996). 10 Sun et al. (2003) used data from five research centers to establish the models that predict fat-free mass as a function of electrical resistance as well as height and weight. They obtained their measure of body fat by subtracting fat-free mass from total bodyweight. Fat-free mass is calculated from a deterministic formula based on bone mineral content, total body water, body volume, and bodyweight using a multi-component molecular model derived particularly for body composition analysis (Heymsfield et al., 1996).

members of minority groups (National Institutes of Health, 1994; Heyward and Wagner, 2004). Nevertheless, we will present results for both whites and blacks for comparison purposes. The predictive equations provided by Sun et al. (2003) are in the following forms for males: FFM ¼ 10:678 þ 0:262 weight þ 0:652

stature2 resistance

þ 0:015 resistance;

(1)

and for females: FFM ¼ 9:529 þ 0:168 weight þ 0:696 þ 0:016 resistance;

stature2 resistance (2)

where weight is clinically measured weight in kilograms and stature is clinically measured height in centimeters. The resistance, which is a measure of resistance to electric current, is measured in ohms. As reported by Sun et al., these equations have very high predictive power with the Rsquared values of 0.90 for males and 0.83 for females (Sun et al., 2003). Once the FFM is obtained from above equations, BF is calculated by subtracting FFM from total weight.11 Unfortunately, the NHANES III does not contain information necessary to compute hourly wages. To overcome this problem, we use the information on body composition available in the NHANES III to impute measures of body composition for respondents in the National Longitudinal Survey of Youths 1979 (NLSY), which provides data on hourly wages.12 The imputed body composition is then used to estimate the wage equation for the respondents in the NLSY. The parameters needed to predict body composition are extracted from the NHANES by separately regressing FFM and BF on selfreported characteristics that are available in both data sets. These characteristics include age, age2, age3, weight, weight2, weight3, height, height2, height3, and an interaction between height and weight.13 The results from the prediction equations for FFM and BF are presented in Table A1. To help account for differences across gender and race, they are estimated separately for each group. Age, weight, and height as well as their polynomials and interactions between height and weight appear to be important determinants of FFM. The adjusted R-squared values for FFM are about 0.82 for both males and females. For BF, the adjusted R-squared values are 0.76 and 0.90 for white males and white females, respectively and 0.77 and 0.87 for black males and black females. These high R-squared values imply that a large proportion of the variation in the FFM and BF can be

11 Sun et al. mentioned that their final equations over-predicted the FFM for white males by 0.4 kg and the FFM for white females by 0.3 kg. Our observations have been adjusted accordingly by adding or subtracting the average errors from each gender group. 12 Both the NHANES and the NLSY are nationally representative samples of the U.S. population, making them comparable. 13 The height and weight information provided in the NLSY is selfreported. To correct for possible reporting errors in the NLSY, we use selfreported height and weight in estimating clinical measures of FFM and BF in the NHANES.

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

245

3. Data

thereafter. The NLSY provides detailed information on the labor market outcomes of respondents along with a rich set of personal and family characteristics. Although the NLSY does not provide a direct measure of body composition, it is one of the few economic surveys with longitudinal information on the body measurements, such as height and weight. We pooled all the NLSY between years 1981 and 2004 – the period during which self-reported height and weight information are available.14 The age range for our NLSY sample for the wage models is 18 and 49. To avoid changes in body composition during pregnancy, females who were determined to be pregnant at the time of an interview are dropped from the analysis sample.15 After applying these exclusion criteria, we have 26,730 pooled observations for white males, 22,479 for white females, 9919 for black males, and 8802 for black females with non-missing information for the variables of interest. The NLSY asks about the hourly wage of respondents at their primary jobs. We deflated the hourly wages to 1991 dollars using the Consumer Price Index.16 The other variables included in the analyses are age, years of education, years of job tenure, an indicator for marital status, an indicator for urban residence, region indicators, the highest grade completed by the mother, the highest grade completed by the father, the score from Armed Forces Qualification Test (AFQT) as a proxy measure of intelligence, years of employment experience, and year dummies. We also include county unemployment rate as a control for labor market conditions. A dummy variable indicating whether the individual has any health problems limiting the kind or amount of work one can perform is also included. Finally, we also constructed a binary indicator indicating blue-collar workers. Table 1 presents the descriptive statistics for the NLSY sample including the predicted FFM and BF along with the definitions of the variables. The descriptive statistics is presented separately by race and gender. As illustrated in Table 1, males of both races are taller and heavier than their female counterparts. Females generally possess a slightly higher level of BF than males. White males have the highest average hourly wages. The proportion of sample having a health problem limiting work and other activities is slightly higher for females than males. The average education of parents is higher among whites than blacks though there is little difference between males and females. Similarly, average AFQT scores are higher among whites than blacks, but again the difference is little between males and females. Finally, males are much more likely to work in blue-collar occupations than females.

The NHANES III, which is described in the previous section, is an ideal data set for studying body composition because it contains information on both self-reported and measured height and weight, and more crucially the BIA readings. The availability of BIA readings in the NHANES III is critical for the purpose of this paper because they allow us to construct measures of FFM and BF. Our sample for the analysis of wage rate is drawn from the NLSY. The NLSY started in 1979 with a cohort of males and females between ages 14 and 21. These individuals have been followed annually until 1993 and biannually

14 Height was collected in 1981, 1982 and 1985. The last measurement of height was interpreted as the adult height for the rest of years. The weight was collected at the same time with height as well as the rest of survey years. 15 Reports of current pregnancies and past pregnancies were not collected at every interview. To overcome this problem, a data set was constructed from the birth dates of women’s biological children and the interview dates. Women were identified as pregnant if the interview occurred between 9 months before or after the birth of a biological child. 16 Following Cawley (2004), hourly wages were top- and bottom-coded to be between 1 and 500 in 1991 dollars.

explained by the variations in the covariates included in these regressions. Taken together, results in Table A1 suggest that these models accurately predict the FFM and BF in the NHANES III and that the estimated coefficients can reliably be used to construct FFM and BF in other data sets such as the NLSY. Using the parameter estimates obtained from these prediction equations, we imputed measures of FFM and BF d it and for respondents in the NLSY. The imputed values, FFM c it , are then included in the hourly wage model estimated BF with data from the NLSY. The wage equation has the following form: d it þ a2 BF c it þ eit ; lnðwit Þ ¼ bX it þ a1 FFM

(5)

where lnðwit Þ is the logarithm of the hourly wage rate for individual i in year t, X it is a vector of the observed determinants of wages, b and a’s are the parameters and eit is an idiosyncratic disturbance term. We exclude both height and weight from the wage d it and BF c it . equations, since they are already included in FFM It is important to note this is similar to the estimation strategy using BMI, where height and weight are also excluded from the wage equations. Instead of arbitrarily dividing weight by the square of height, we allow the data to determine how height and weight should interact with each other and other characteristics to define someone as obese. However, height can also influence earnings through higher social capital accumulated during high school years (Persico et al., 2004). To account for the possibility that height can have an independent effect on wages, we control for a set of variables that could serve as a proxy for the sociability of individuals, such as participation in high school sports and clubs, as measured in the NLSY. Alternatively, we use height during childhood (prior to age 18) as an additional specification check. Our models also include education, AFQT test scores, and parents’ education. These variables could further control for the individual’s social skills. We also estimate models with the method of 2-Stage Least Squares (2SLS) to address the potential endogeneity of the body composition measures. In the 2SLS models, we use the body composition of the respondents’ sibling as the instruments following Behrman and Rosenzweig (2001) and Cawley (2004). Finally, we estimate all of our models with individual fixed effects accounting for any timeinvariant unobserved heterogeneity that could be correlated with both body composition and wages.

246

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

Table 1 Descriptive statistics (mean and standard error) of NLSY 1979. Variables

Definitions

White males

White females

Black males

Black females

Hourly wage

Hourly wage rate in 1991 dollars (adjusted by CPI)

13.19 (18.51)

10.11 (16.83)

10.37 (15.77)

8.754 (10.00)

FFM

Estimated fat-free mass in kilograms

62.70 (8.641)

43.54 (5.600)

64.79 (8.926)

47.64 (6.708)

BF

Estimated body fat in kilograms

20.16 (7.467)

22.63 (9.830)

18.23 (7.853)

27.24 (12.14)

BMIa

Weight/height2

26.00 (4.458)

24.51 (5.355)

26.32 (4.673)

28.07 (6.688)

Underweight

Dummy variable = 1 if BMI < 18.5

0.00677 (0.0820)

0.0340 (0.181)

0.00454 (0.0672)

0.0137 (0.116)

Healthy

Dummy variable = 1 if 18.5  BMI < 25

0.465 (0.499)

0.629 (0.483)

0.452 (0.498)

0.377 (0.485)

Overweight

Dummy variable = 1 if 25  BMI < 30

0.370 (0.483)

0.201 (0.401)

0.354 (0.478)

0.286 (0.452)

Obese

Dummy variable = 1 if 30  BMI

0.158 (0.365)

0.136 (0.343)

0.190 (0.392)

0.323 (0.468)

Weightb

Kilograms

82.85 (15.83)

66.43 (15.15)

83.02 (16.39)

75.16 (18.61)

Heightb

Meters

1.783 (0.0635)

1.645 (0.0569)

1.774 (0.0614)

1.635 (0.0532)

Health limitation

Dummy variable = 1 if health limits kind or amount of work

0.0343 (0.182)

0.0482 (0.214)

0.0332 (0.179)

0.0502 (0.218)

AFQT 1980

Armed Forces Qualification Test from 1980 to 1981

54.15 (28.40)

54.38 (25.66)

24.74 (22.68)

26.33 (19.92)

Mother’s education

Years of education completed by mother

11.92 (2.453)

11.86 (2.433)

11.09 (2.558)

11.02 (2.603)

Father’s education

Years of education completed by father

12.12 (3.421)

12.07 (3.288)

10.27 (3.396)

10.25 (3.603)

Children

# of biological/step/adopted children in the household

0.792 (1.099)

0.943 (1.120)

0.662 (1.092)

1.243 (1.224)

Attend

Dummy variable = 1 if currently attending school

0.0847 (0.278)

0.103 (0.303)

0.0612 (0.240)

0.0891 (0.285)

Married

Dummy variable = 1 if married

0.529 (0.499)

0.535 (0.499)

0.324 (0.468)

0.312 (0.463)

Education

Years of education

13.25 (2.455)

13.43 (2.236)

12.72 (2.069)

13.31 (1.974)

Age

Age in years (to the closest month)

30.68 (6.800)

30.92 (7.026)

31.48 (6.824)

32.10 (6.882)

Tenure

Years of tenure (50 weeks/year)

4.392 (4.917)

3.796 (4.455)

3.528 (4.264)

4.127 (4.867)

Experience

Years of work experience (50 weeks/year)

11.37 (6.751)

10.54 (6.545)

10.45 (6.436)

9.980 (6.453)

Low unemploymentc

Dummy variable = 1 if unemployment rate is less than 5.9%

0.452 (0.498)

0.459 (0.498)

0.522 (0.500)

0.544 (0.498)

Medium unemployment

Dummy variable = 1 if unemployment rate is between 6% and 8.9%

0.357 (0.479)

0.354 (0.478)

0.353 (0.478)

0.335 (0.472)

High unemployment

Dummy variable = 1 if unemployment rate is between 9% and 11.9%

0.126 (0.331)

0.124 (0.330)

0.0955 (0.294)

0.0880 (0.283)

Very high unemployment

Dummy variable = 1 if unemployment rate is higher than 12%

0.0647 (0.246)

0.0636 (0.244)

0.0291 (0.168)

0.0331 (0.179)

Urban

Dummy variable = 1 if urban

0.720 (0.449)

0.721 (0.449)

0.840 (0.367)

0.863 (0.344)

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

247

Table 1 (Continued ) Variables

Definitions

White males

White females

Black males

Black females

Northeast

Dummy variable = 1 if Northeast region

0.189 (0.392)

0.194 (0.395)

0.167 (0.373)

0.139 (0.346)

West

Dummy variable = 1 if West region

0.164 (0.371)

0.166 (0.372)

0.0853 (0.279)

0.0723 (0.259)

Midwest

Dummy variable = 1 if Midwest region

0.345 (0.475)

0.311 (0.463)

0.177 (0.382)

0.171 (0.377)

Southc

Dummy variable = 1 if South region

0.301 (0.459)

0.329 (0.470)

0.570 (0.495)

0.618 (0.486)

Blue-collar

Dummy variable = 1 if blue-collar occupationb

0.568 (0.495)

0.312 (0.463)

0.688 (0.463)

0.413 (0.492)

Year 1981

Dummy variable = 1 if year = 1981

0.0545 (0.227)

0.0585 (0.235)

0.0411 (0.199)

0.0376 (0.190)

Year 1982

Dummy variable = 1 if year = 1982

0.0846 (0.278)

0.0862 (0.281)

0.0657 (0.248)

0.0571 (0.232)

Year 1985

Dummy variable = 1 if year = 1985

0.0765 (0.266)

0.0761 (0.265)

0.0681 (0.252)

0.0610 (0.239)

Year 1986

Dummy variable = 1 if year = 1986

0.0744 (0.262)

0.0748 (0.263)

0.0702 (0.255)

0.0653 (0.247)

Year 1988

Dummy variable = 1 if year = 1988

0.0803 (0.272)

0.0770 (0.267)

0.0734 (0.261)

0.0706 (0.256)

Year 1989

Dummy variable = 1 if year = 1989

0.0816 (0.274)

0.0779 (0.268)

0.0743 (0.262)

0.0711 (0.257)

Year 1990

Dummy variable = 1 if year = 1990

0.0789 (0.270)

0.0767 (0.266)

0.0711 (0.257)

0.0717 (0.258)

Year 1992

Dummy variable = 1 if year = 1992

0.0616 (0.240)

0.0582 (0.234)

0.0707 (0.256)

0.0687 (0.253)

Year 1993

Dummy variable = 1 if year = 1993

0.0629 (0.243)

0.0588 (0.235)

0.0708 (0.256)

0.0679 (0.252)

Year 1994

Dummy variable = 1 if year = 1994

0.0607 (0.239)

0.0570 (0.232)

0.0673 (0.251)

0.0653 (0.247)

Year 1996

Dummy variable = 1 if year = 1996

0.0623 (0.242)

0.0621 (0.241)

0.0730 (0.260)

0.0744 (0.262)

Year 1998

Dummy variable = 1 if year = 1998

0.0601 (0.238)

0.0608 (0.239)

0.0683 (0.252)

0.0746 (0.263)

Year 2000c

Dummy variable = 1 if year = 2000

0.0567 (0.231)

0.0613 (0.240)

0.0661 (0.249)

0.0728 (0.260)

Year 2002

Dummy variable = 1 if year = 2002

0.0530 (0.224)

0.0580 (0.234)

0.0624 (0.242)

0.0716 (0.258)

Year 2004

Dummy variable = 1 if year = 2004

0.0519 (0.222)

0.0567 (0.231)

0.0576 (0.233)

0.0701 (0.255)

26,730

22,479

9,919

8,802

Observations Note: Standard deviations are in parentheses. a Calculated from adjusted height and weight (see the text for explanations). b Adjusted height and weight. c Omitted category.

4. Results Studies using conventional measures of obesity (BMI, weight, binary indicators of overweight and obese) typically implement the empirical analyses in a numbers of steps (see e.g., Averett and Korenman, 1996; Baum and Ford, 2004; Cawley, 2004). First, they usually estimate models using contemporaneous measures of obesity. Second, they carry out their analyses using lagged measures of obesity (7 years) to guard against the

possibility of reverse causality from wages to obesity. Third, some use sibling BMI as an instrument to control for the possible endogeneity of own BMI. Fourth, they estimate models controlling for fixed effects to account for time-invariant unobserved heterogeneity that could potentially bias the results.17 Note that we also estimate

17 See Cawley (2004) for a more detailed discussion of the empirical specifications used in studies with conventional measures.

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

248

Table 2A OLS results from the models using contemporaneous fat-free mass and body-fat. Variable

White males ***

White females ***

***

Black males ***

Black females

Fat-free mass (kg)

0.00711 (0.00194)

0.00662 (0.00192)

0.0120 (0.00311)

0.0117 (0.00308)

0.00478 (0.00307)

0.00478 (0.00307)

0.00766 (0.00500)

0.00744 (0.00497)

Body fat (kg)

0.00858*** (0.00231)

0.00843*** (0.00227)

0.0109*** (0.00182)

0.0105*** (0.00180)

0.00398 (0.00340)

0.00398 (0.00340)

0.00546* (0.00279)

0.00528* (0.00277)

Sociability indicators

No

Yes

No

Yes

No

Yes

No

Yes

Observations

26,730

26,730

22,479

22,479

9919

9919

8802

8802

Note: Bootstrapped, robust standard errors clustered around individuals are in parentheses. * Statistical significance at 10%. *** Statistical significance at 1%.

**

Statistical significance at 5%.

Table 2B OLS results from the models using 7-year lagged fat-free mass and body-fat. Variable

White males

Fat-free mass (kg)

0.00731*** (0.00231)

0.00682*** (0.00232)

White females 0.0151*** (0.00383)

0.0149*** (0.00380)

Black males 0.00326 (0.00340)

0.00233 (0.00336)

Black females 0.0114* (0.00638)

0.0111* (0.00641)

Body fat (kg)

0.00996*** (0.00300)

0.00999*** (0.00298)

0.0130*** (0.00227)

0.0127*** (0.00226)

0.00210 (0.00385)

0.00195 (0.00385)

0.00888** (0.00363)

0.00865** (0.00367)

Sociability indicators

No

Yes

No

Yes

No

Yes

Yes

No

Observations

10,602

10,602

8727

8727

4135

4135

3656

3656

Note: Bootstrapped, robust standard errors clustered around individuals are in parentheses. * Statistical significance at 10%. ** Statistical significance at 5%. *** Statistical significance at 1%.

our models using these conventional measures of obesity to confirm the findings of the previous studies. Our results are largely consistent with those of previous studies, but they are not discussed here to economize on space.18 Table 2A presents the results from the wage models with two contemporaneous measures of body composition – FFM in kilograms and BF in kilograms. Tables 2B, 2C and 2D display the results from the models with the lagged measures, the instrumental variable analysis, and the fixed effects analysis, respectively.19 As discussed in Section 2, we also control for variables that could serve as a proxy for the sociability of the individual in some specifications. Specifically, the sociability variables that we include in the models are nine binary indicators for the most active high school club participation, such as athletics or marching band.20 To assess the effect of these variables on the impact of body

18 These results are available in a longer version of this paper (Wada and Tekin, 2007). 19 Since these models use measures of body composition constructed from the regressions coefficients that are transferred from the NHANES III to the NLSY, the standard errors will be underestimated. Therefore, we present bootstrapped standard errors in all these tables. We implemented bootstrapping with 399 replications. The implications of the results remained the same when we repeated higher values of replications. 20 It is possible that the contemporaneous sociability variables are endogenous to wages because higher wages are likely to raise sociability. In order to avoid bias due to potential reverse causality from wages to sociability, we use sociability indicators from high school years rather than current indicators of sociability. Nevertheless, models that also included current indicators of sociability such as measures of selfassessed ‘‘shyness’’ did not change the results.

composition measures, we present results in Tables 2A, 2B and 2C with and without these indicators. Note that the sociability variables are not included in Table 2D because their effects are gauged by the individual fixed effects.21 In Table 2A, we find that the coefficients on the measures of body composition have the signs consistent with our expectations for both genders. That is, the FFM and the BF are associated with an increase and a decrease in wages, respectively, regardless of the gender.22 Furthermore, including nine variables for high school clubs that capture the social skills of the individual does not cause appreciable changes to the coefficient estimates. Looking at the coefficients in the results with social indicators in Table 2A, a 1 kg increase in the BF reduces wages by about 1% for both white males and females. When the FFM is raised

21 In order to guard against the possibility that sociability indicators do not fully capture this channel, we experimented with models controlling for height during childhood. Specifically, we estimated models with earliest available height before age 18 in addition to the sociability indicators. In order to implement this, we had to restrict our sample such that earliest available height of an individual is from a point in life before age 18. As a result of this, our sample sizes went down to less than 1/4th of the original sizes. These results are presented in Table A2. Despite dramatic reductions in the sample sizes and high collinearity between adult height and childhood height, all of the FFM and BF coefficients are in the expected sign and the effects are largely consistent with those in previous tables. Note that sociability indicators are also controlled for in these models. 22 We also estimated our models with both linear and quadratic terms of body composition measures. The quadratic terms are mostly small and insignificant. Also note that the inclusion of quadratic terms complicate the instrumental variable analysis. Therefore, we only report results from the linear specifications here.

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

249

Table 2C Sibling 2SLS results from the models using contemporaneous fat-free mass and body-fat. Variable

White males *

White females

Black males

Black females

Fat-free mass (kg)

0.0203 (0.0111)

0.0176 (0.0112)

0.0222 (0.0223)

0.0243 (0.0214)

0.0229 (0.0195)

0.0227 (0.0202)

0.00988 (0.0249)

0.0114 (0.0259)

Body fat (kg)

0.0261* (0.0151)

0.0227 (0.0151)

0.0205 (0.0144)

0.0219 (0.0140)

0.0297 (0.0257)

0.0290 (0.0266)

0.00823 (0.0174)

0.00823 (0.0179)

Sociability indicators

No

Yes

No

Yes

No

Yes

No

Yes

Observations 1st stage F-statistics

10,833 16.61

10,833 17.21

8607 20.63

8607 20.71

4825 7.22

4825 7.10

4002 8.23

4002 8.23

Note: Bootstrapped, robust standard errors clustered around individuals are in parentheses. **Statistical significance at 5%; ***Statistical significance at 1%. * Statistical significance at 10%. Table 2D Fixed effects results from the models using fat-free mass and body-fat. Variable

White males

White females

Black males

Black females

Fat-free mass (kg)

0.0150*** (0.00494)

0.0252* (0.0131)

0.0109 (0.00662)

0.00982 (0.0174)

Body fat (kg)

0.0158*** (0.00486)

0.0150** (0.00669)

0.00647 (0.00648)

0.00253 (0.00901)

Observations Individuals

26,730 2909

22,479 2881

9919 1004

8802 999

Note: Bootstrapped, robust standard errors clustered around individuals are in parentheses. * Statistical significance at 10%. ** Statistical significance at 5%. *** Statistical significance at 1%.

by 1 kg, the wages increase by about 0.7% for white males and about 1.2% for white females. Note that a 1 kg (2.2 lb) increase in BF is equivalent to about 5% increase in the body fat of white males and about 4% increase in the body fat of white females. Similarly, a 1 kg (2.2 lb) increase in the FFM is equivalent to about 2% increase in the fat-free mass of white males and white females. Among blacks, a 1 kg increase in BF is associated with a 0.5 and 0.8% increase in wages for males and females, respectively, although the coefficients are not estimated with much precision. A 1 kg increase in BF is associated with about 0.5 and 0.7% decrease in the hourly wages of black males and black females, respectively, although only the coefficients for females are significant at conventional levels. The weaker results for blacks are likely to be due to considerably smaller sample sizes and the expected increase in measurement error for blacks as discussed earlier. These results indicate that, while an increase in body size due to BF would hurt hourly wages, an increase due to FFM is apparently beneficial. Table 2B presents the results for the models with lagged values of FFM and BF. We chose to lag FFM and BF by 7 years because larger lags would considerably reduce the number of available observations and 7 years is what other studies have used.23 Sample size is reduced by approximately half as a result of using 7-year lags. As shown in Table 2B, the results from the lagged estimations are largely similar to those using contemporaneous FFM and BF. Furthermore, both the coefficients for FFM and BF are significant for white males, white females, and black females.

Table 2C presents the 2SLS results using the FFM and BF of siblings as instruments.24 Because body composition is affected by age, the sibling observations were also matched by age over the years. For example, an observation at the age of 25 would be matched to a closest sibling observation when the sibling was between the ages of 24 and 26. This ensures that the instrument has the sufficient predictive power. A test of under-identification rejects the null hypothesis at <0.01% level for all gender and race groups. A test of weak instruments, however, indicates that the instruments are sufficiently strong for whites but only marginal for blacks.25 The validity of the sibling body composition measures as instruments, however, hinges on the supposition that sibling body composition is correlated with own body composition and uncorrelated with one’s own earnings except through body composition. These assumptions are extensively discussed by Cawley (2004), Gregory and Ruhm (2009), and Smith et al. (2009). As shown in Table 2C, the estimated coefficients for FFM are about twice as large as the OLS results. Furthermore, the results for white males are statistically significant at the 10% level. The results for blacks are more mixed. While the results for black males are very similar to whites, the signs on the coefficients for BF and FFM are opposite to what we would expect, but the estimates are not statistically significant. The BF coefficients are negative for all groups except for black females and they are marginally significant for white males. Although the evidence is not as strong as the

24

23

A lag of seven was also used by Averett and Korenman (1996) and Cawley (2004) in their studies using BMI.

Note that the 2SLS models also control for sibling gender. The Cragg–Donald Wald-F-Statistic ranges from 16 to 21 for whites and about 7–8 for blacks. 25

250

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

Table 3 Summary statistics for supplemental BIA equations with FFM and BF. Statistics

Variable

White males

White females

Black males

Black females

# of equations with significant coefficients

FFM BF

47 47

37 41

39 1

14 11

Median coefficient

FFM BF

0.0170 0.0135

0.0453 0.0270

0.0135 0.00565

0.0292 0.0146

results from the contemporaneous and lagged OLS models, the 2SLS results lend some support to the hypothesis that FFM and BF are positively and negatively associated with wages, respectively. Furthermore, significantly smaller sample sizes for black females may be another factor contributing to weaker results for blacks. Finally, although blacks are more likely than whites to have half-siblings and non-related ‘‘siblings’’ in the household, the NLSY unfortunately do not distinguish them. Table 2D presents the coefficient estimates from regressions with individual fixed effects. These models account for all of the time-invariant unobserved factors that might be correlated with both body composition and wages. Despite controlling for time-invariant heterogeneity, the results displayed in Table 2D suggest that FFM and BF are associated with an increase and a decrease in hourly wages of all groups and that the effects of BF and FFM are statistically significant for white males and white females. The magnitudes of the FFM and BF effects approximately double between Tables 2A and 2D. The persistent and significant effects of body composition even after controlling for individual fixed effects suggest that an increase in BF is indeed detrimental to the wages of both the white females and the white males. Once again, the coefficients are not estimated precisely for blacks possibly due to the smaller sample sizes and the fact that prediction equations are not generally well-suited for blacks. We further check for robustness of the result by repeating the specifications used in Table 2D for ages between 25 and 49. Our results are not sensitive to restricting the sample age in this way. 4.1. Alternative BIA conversion equations for FFM and BF It is possible that the findings discussed above are driven by the choice of a particular set of BIA conversion equations. In order to address this possibility, we gathered a comprehensive set of conversion equations estimated by other researchers and documented in published sources. This set includes 47 separate BIA equations.26 We believe

26 This is the same list identified in a recent study by Willett et al. (2006), minus one redundant equation due to replication. We combine four equations in that list because they were body-fat specific equations of Segal et al. (1988) that were originally meant to be combined (see Heyward and Wagner, 2004). All 4 equations based on percentage body fat (as opposed to level measures of FFM or bodyweight) are dropped from the list due to high degree of prediction errors. Many of the remaining prediction equations are actually less suitable for the purpose of this paper because they are derived for populations of different ages like children and older adults. Nevertheless, they are retained in our analysis to show that the final result is largely robust to such built-in errors. To this list, we further add four equations from other published resources.

that this set includes most of the well-known BIA prediction equations that exist in published sources. These equations are presented in Supplementary Table S1, which are available online from the journal’s website. We estimated our models using each of these alternative equations. Note that we use the same set of regressors in these models and also include individual fixed effects. As discussed below, these estimations produced FFM and BF coefficients that are highly consistent with those presented above. In Supplementary Tables S2 and S3 (which can be downloaded from the journal’s website at doi:10.1016/ j.ehb.2010.02.001), we present the coefficients on the FFM and BF for males and females among whites and blacks, respectively, for each of these 47 prediction equations. As summarized in Supplementary Table S2, in all 47 equations, all four FFM and BF coefficients – a total of 188 regression coefficients (47  2  2) – remarkably have the expected sign, that is, the effect of FFM is positive and the effect of BF is negative. As summarized in Table 3, the FFM coefficients are statistically significant in 47 out of 47 models and the BF coefficients are statistically significant in all 47 models for white males. This is consistent with our results presented in Table 2D. For white females, the FFM coefficients are significant in 41 models and the BF coefficients are statistically significant in 37 models. As shown in Deurenberg et al., 1989; Houtkooper et al., 1996; Kyle et al., 2001; Supplementary Table S3, the results are weaker for blacks. For black males, 37 of FFM are statistically significant. While the number of statistically significant estimates is fewer for blacks than for whites, this is to be expected given the less precision with which the body composition is predicted for blacks than for whites. Nevertheless, the FFM and BF coefficients are positive and negative, respectively, in all 47 models for both black males and black females. This further increases our confidence that the effects of FFM and BF on the wages of blacks are likely to be similar to those of whites. In order to summarize the information in Supplementary Tables S2 and S3, Table 3 also presents the medians of the 47 BF and 47 FFM coefficients estimates for each group. The medians for females are generally larger in magnitude than the estimated coefficients using the prediction equations of Sun et al. (2003). For males, they are similar for FFM, but the median for BF is about 20% larger. The fact that most of the median values move away from zero further support that our main results are not driven by the choice of a particular prediction equation. In summary, the persistent and significant effects of body composition on wages even after controlling for individual fixed effects suggest that an increase in BF is

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

indeed detrimental for the wages of not only white females as usually found in the studies using BMI but also for white males. Furthermore, FFM is consistently associated with increased hourly wages. The findings are more pronounced for whites than blacks possibly due to explanations suggested above. 5. Conclusion In this paper, we examine the effect of body composition on wages. A majority of the previous studies on this subject use BMI and bodyweight to measure obesity. However, there is increasing evidence to suggest they are not the best surrogates for obesity because of their inability to distinguish between fat body mass and fatfree mass. Since it is the body fat that classifies an individual as obese, the effects obtained in previous studies may be confounded by the impact of the fat-free component of body composition. We measure body composition by body fat (BF) and fat-free mass (FFM). We do so by utilizing the information on the bioelectrical impedance analysis that is available in NHANES III and implement a method for calculating body composition for use in regression analyses. By using measures of BF and FFM, we are able to distinguish between the effects of healthy physical growth (represented by an increase in FFM) and an unhealthy physical growth (represented by an increase in BF) on wages. The analysis using body composition also allows us to study the effect of FFM on wages. Because FFM consists mostly of muscles and skeletons, it presents a plausible way to estimate the effect of physical health on worker earnings. Our results suggest that a rise in BF is associated with decreases in the wages of both white males and white females, while a rise in the FFM is associated with an increase in the wages of both groups. These findings are in contrast to the previous studies that found strong evidence of a negative effect on white females but not always for white males. Given that a higher proportion of women’s body consists of fat than men due to demands for childbearing and other hormonal functions, BMI may serve as a better measure of excessive fatness for women than men. Such gender-dependent correlation could

251

particularly explain the previously mixed and unstable findings for men. These results also hold for black males and black females to lesser extent despite the smaller sample sizes for these groups and the fact that prediction equations from which body composition measures are derived are based on predominantly white samples. Our results indicate that individuals with high levels of FFM or lean body mass earn a wage premium. To the extent that FFM can be seen as a measure of healthy growth, our results provide evidence in favor of the nutrition hypothesis expounded by Fogel (1994). It has been long hypothesized that increased body size should be associated with increased worker productivity. Our result using fat-free mass demonstrates that it is the healthy growth of FFM that is beneficial. We also present evidence that these results are not artifacts of other characteristics of the individuals that are correlated with obesity. Finally, we show evidence that our findings are robust to the choice of prediction equation based on which the body composition measures are derived. Researchers have recently found that non-cognitive characteristics such as beauty, leadership, and tallness are positively related to earnings. By studying the effects of BF and FFM on wages, this study also contributes to the growing literature on the role of non-cognitive factors on wage determination.

Acknowledgements We would like to thank John Cawley, Paul Farnham, Michael Grossman, Julie Hotchkiss, Greg Lewis, Inas Rashad, and the seminar participants at University of Oregon, AHRQ, Rice University, Tinbergen Institute, CDC, IZA, the 2009 IHEA/AHEA Meetings, and the 2008 Latin American Meeting of the Econometric Society in Rio de Janeiro for helpful comments. We also thank Susan Averett for sharing her SAS codes. Alexander Brumlik and Jason Delaney provided excellent research assistance.

Appendix A See Tables A1 and A2.

Table A1 Determinants of (A) fat-free mass (FFM) and (B) body fat (BF) from NHANES III. (A) Fat-free mass (FFM) Variable

White males

White females

Black males

Black females

Age

0.0296 (0.0631) 0.0986 (0.135) 0.000340 (0.000881) 301.9 (250.4) 175.8 (147.3) 33.71 (28.98)

0.121*** (0.0366) 0.316*** (0.0800) 0.00194*** (0.000531) 181.3 (115.2) 108.8 (76.45) 18.86 (16.92)

0.0734 (0.0838) 0.0849 (0.196) 0.000577 (0.00140) 27.54 (251.0) 0.460 (146.3) 1.810 (28.35)

0.0603 (0.0535) 0.172 (0.129) 0.000942 (0.000947) 93.74 (71.27) 54.89 (48.16) 10.72 (10.75)

Age2 (1/100) Age3 (1/100) Height Height2 (1/100) Height3 (1/100)

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

252 Table A1 (Continued ) (A) Fat-free mass (FFM) Variable Weight Weight2 (1/100) Weight3 (1/100) Height  Weight Constant

Observations R-squared

White males *

White females ***

Black males

Black females

0.259 (0.152) 0.168 (0.113) 0.000265 (0.000347) 0.276*** (0.0870) 160.6 (142.3)

0.262 (0.0889) 0.187** (0.0784) 0.000757*** (0.000288) 0.0359 (0.0539) 111.2* (57.97)

0.146 (0.163) 0.534*** (0.125) 0.00204*** (0.000388) 0.130 (0.0879) 3.702 (143.2)

0.355*** (0.122) 0.413*** (0.119) 0.00186*** (0.000439) 0.262*** (0.0553) 80.52** (35.29)

3201 0.824

3539 0.816

2283 0.820

2513 0.807

White males

White females

Black males

Black females

(B) Body fat (BF) Variable Age 2

Age (1/100) Age3 (1/100) Height Height2 (1/100) Height3 (1/100) Weight Weight2 (1/100) Weight3 (1/100) Height  Weight Constant

Observations R-squared

***

0.0653 (0.0628) 0.241* (0.134) 0.00166* (0.000877) 258.2 (249.2) 147.9 (146.6) 29.30 (28.84) 0.327** (0.151) 0.402*** (0.112) 0.000964*** (0.000345) 0.162* (0.0866) 151.1 (141.6)

0.158 (0.0456) 0.350*** (0.0995) 0.00222*** (0.000660) 545.6*** (143.3) 373.9*** (95.14) 80.59*** (21.06) 0.685*** (0.111) 0.165* (0.0976) 0.000343 (0.000359) 0.122* (0.0671) 270.8*** (72.15)

0.135 (0.0842) 0.368* (0.197) 0.00269* (0.00141) 797.9*** (252.3) 487.2*** (147.0) 98.30*** (28.50) 0.709*** (0.164) 0.148 (0.126) 9.94e05 (0.000391) 0.260*** (0.0883) 443.5*** (143.9)

0.0778 (0.0760) 0.150 (0.184) 0.000771 (0.00135) 425.1*** (101.3) 274.7*** (68.45) 55.96*** (15.28) 0.360** (0.173) 0.420** (0.169) 0.00214*** (0.000624) 0.0651 (0.0787) 227.0*** (50.16)

3201 0.763

3539 0.897

2283 0.768

2513 0.871

Notes: Standard errors are in parentheses. Note that only the squared and cubed terms are adjusted by a factor of 1/100. * Statistical significance at 10%. ** Statistical significance at 5%. *** Statistical significance at 1%.

Table A2 OLS results from the models using contemporaneous fat-free mass and body-fat (restricted to earliest available height at age before 18). Variable

White males

White females

Black males

Black females

Fat-free mass (kg)

0.0136** (0.00547)

0.0210 (0.0173)

0.0282*** (0.00982)

0.000439 (0.0195)

Body fat (kg)

0.0112* (0.00580)

0.0126 (0.00894)

0.0276** (0.0110)

0.00245 0.000439

Sociability indicators Earliest available height

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Observations

5664

4253

2049

1712

Note: Bootstrapped, robust standard errors clustered around individuals are in parentheses. * Statistical significance at 10%. ** Statistical significance at 5%. *** Statistical significance at 1%.

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

Appendix B. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.ehb. 2010.02.001. References Allison, D.B., Zhu, S.K., Plankey, M., Faith, M.S., Heo, M., 2002. Differential associations of body mass index and adiposity with all-cause mortality among men in the first and second National Health and Nutrition Examination Surveys (NHANES I and NHANES II) follow-up studies. International Journal of Obesity 26 (3), 410–416. Averett, S., Korenman, S., 1996. The economic reality of the beauty myth. Journal of Human Resources 31 (2), 304–330. Baum, C.L., Ford, W.-F., 2004. The wage effects of obesity: a longitudinal study. Health Economics 13 (9), 885–899. Behrman, J.R., Rosenzweig, M.R., 2001. The returns to increasing bodyweight. PIER Working Paper 01-052. Bjorntorp, P., 2002. Definition and classification of obesity. In: Fairburn, C.G., Brownell, K.D. (Eds.), Eating Disorders and Obesity: A Comprehensive Handbook. Guilford Press, New York, pp. 377–381. Brozek, J., Grande, F., Anderson, J.T., Keys, A., 1963. Densitometric analysis of body composition: revision of some quantitative assumptions. Annals of the New York Academy of Sciences 110, 113–140. Burkhauser, R.V., Cawley, J., 2008. Beyond BMI: the value of more accurate measures of fatness and obesity in social science research. Journal of Health Economics 519–529. Cawley, J., 2004. The impact of obesity on wages. Journal of Human Resources 39 (2), 451–474. Centers for Disease Control and Prevention, 2007. Diseases and Risk Factors. , http://www.cdc.gov/steps/disease_risk/index.htm. Chumlea, W.C., Guo, S.S., Kuczmarski, R.J., Flegal, K.M., Johnson, C.L., Heymsfield, S.B., Lukaski, H.C., Friedl, K., Hubbard, V.S., 2002. Body composition estimates from NHANES III bioelectrical impedance data. International Journal of Obesity 26, 1596–1609. Curtis, J.P., Selter, J.G., Wang, Y., Rathore, S.S., Jovin, I.S., Jadbabaie, F., Kosiborod, M., Portnay, E.L., Sokol, S.I., Bader, F., et al., 2005. The obesity paradox: body mass index and outcomes in patients with heart failure. Archives of Internal Medicine 165 (1), 55–61. De Lorenzo, A., Deurenberg, P., Pietrantuono, M., Di Daniele, N., Cervelli, V., Andreoli, A., 2003. How fat is obese? Acta Diabetologica 40, S254– S257. Deurenberg, P., Weststrate, J.A., van der Kooy, K., 1989. Body composition changes assessed by bioelectrical impedance measurements. American Journal of Clinical Nutrition 49 (3), 401–403. Flegal, K.M., Graubard, B.I., Williamson, D.F., Gail, M.H., 2005. Excess deaths associated with underweight, overweight, and obesity. JAMA 293, 1861–1867. Fogel, R.W., 1994. Economic growth, population theory, and physiology: the bearing of long-term processes on the making of economic policy. American Economic Review 84 (3), 369–395. Forbes, G.B., 1999. Body composition: influence of nutrition, physical activity, growth, and aging. In: Shils, M.E. (Ed.), Modern Nutrition in Health and Disease. Williams & Wilkins, Baltimore, pp. 789–809. Gallagher, D., Visser, M., Sepulveda, D., et al., 1996. How useful is body mass index for comparison of body fatness across age, sex, and ethnic groups? American Journal of Epidemiology 143 (3), 228–239. Gregory, C.A., Ruhm, C., 2009. Where does the wage penalty bite? National Bureau of Economic Research Working Paper No. 14984, May. Heitmann, B.L., Erikson, H., Ellsinger, B.M., Mikkelsen, K.L., Larsson, B., 2000. Mortality associated with body fat, fat-free mass and body mass index among 60-year-old Swedish men—a 22-year follow-up. The study of men born in 1913. International Journal of Obesity and Related Metabolic Disorders 24 (1), 33–37. Heshka, S., Buhl, K., Heymsfield, S.B., 1994. Obesity: clinical evaluation of body composition and energy expenditure. In: Blackburn, G.L., Kanders, B.S. (Eds.), Obesity: Pathophysiology, Psychology, and Treatment. Chapman & Hall, New York, NY, pp. 39–79. Heymsfield, S.B., Wang, Z.M., Withers, R.T., 1996. Multicomponent molecular level models of body composition analysis. In: Roche, A.F., Heymsfield, S.B., Lohman, T.G. (Eds.), Human Body Composition: Methods and Findings. Human Kinetics, Champaign, IL, pp. 129–148. Heymsfield, S.B., Allison, D.B., Wang, Z.-M., Baumgartner, R.N., Ross, R., 1998. Body composition. In: Bray, G.A., Bouchard, C., James, W.P.T. (Eds.), Handbook of Obesity. Marcel Dekker, New York, pp. 41–77.

253

Heyward, Vivian H., Wagner, Dale R., 2004. Applied Body Composition Assessment, 2nd ed. Human Kinetics, Champaign, IL. Houtkooper, L.B., Lohman, T.G., Going, S.B., Howell, W.H., 1996. Why bioelectrical impedance analysis should be used for estimating adiposity. American Journal of Clinical Nutrition 64 (Suppl.), 436S–448S. Johansson, E., Bockerman, P., Kiiskinen, U., Heliovaara, M., 2009. Obesity and labour market success in Finland: the difference between having a high BMI and being fat. Economics and Human Biology 7, 36–45. Kushner, R.F., Kunigk, A., Alspaugh, M., Andronis, P.T., Leitch, C.A., Schoeller, D.A., 1990. Validation of bioelectrical-impedance analysis as a measurement of change in body composition in obesity. American Journal of Clinical Nutrition 52 (2), 219–223. Kyle, U.G., Genton, L., Karsegard, L., Slosman, D.O., Pichard, C., 2001. Single prediction equation for bioelectrical impedance analysis in adults aged 20–94 years. Nutrition 17 (3), 248–253. Lohman S T.G., 1992. Advances in Body Composition Assessment. Current Issues in Exercise Science, Monograph No. 3. Human Kinetics, Champaign, IL. Macias, N., Aleman-Mateo, H., Esparza-Romero, J., Valencia, M.E., 2007. Body fat measurement by bioelectrical impedance and air displacement plethysmography: a cross-validation study to design bioelectrical impedance equations in Mexican adults. Nutrition Journal 6, 18. McGee, D.L., 2005. Body mass index and mortality: a meta-analysis based on person-level data from twenty-six observational studies. Annals of Epidemiology 15, 87–97. National Institutes of Health, 1994. Bioelectrical impedance analysis in body composition measurement. In: National Institutes of Health Technology Assessment Conference Statement, December 12–14, http://consensus.nih.gov/1994/1994BioelectricImpedanceBodyta015PDF.pdf. Persico, N., Postlewaite, A., Silverman, D., 2004. The effect of adolescent experience on labor market outcomes: the case of height. Journal of Political Economy 112 (5), 1019–1053. Prentice, A.M., Jebb, S.A., 2001. Beyond body mass index. Obesity Reviews 2 (3), 141–147. Romero-Corral, A., Montori, V.M., Somers, V.K., Korinek, J., Thomas, R.J., Allison, T.G., Mookadam, F., Lopez-Jimenez, F., 2006. Association of bodyweight with total mortality and with cardiovascular events in coronary artery disease: a systematic review of cohort studies. Lancet 368 (August (9536)), 666–678. Roubenoff, R., Dallal, G.E., Wilson, P.W., 1995. Predicting body fatness: the body mass index vs estimation by bioelectrical impedance. American Journal of Public Health 85 (5), 726–728. Segal, K.R., Van Loan, M., Fitzgerald, P.I., Hodgdon, J.A., Van Itallie, T.B., 1988. Lean body mass estimation by bioelectrical impedance analysis: a four-site cross-validation study. American Journal of Clinical Nutrition 47 (1), 7–14. Siri, W.E., 1961. Body composition from fluid space and density. In: Brozek, J., Hanschel, A. (Eds.), Techniques for Measuring Body Composition. National Academy of Science, Washington, DC, pp. 223–244. Smalley, K.J., Knerr, A.N., Kendrick, Z.V., Colliver, J.A., Owen, O.E., 1990. Reassessment of body mass indices. American Journal of Clinical Nutrition 52 (3), 405–408. Smith, G.D., Sterne, J.A.C., Fraser, A., Tynelius, P., Lawlor, D.A., Rasmussen, F., 2009. The association between BMI and mortality using offspring BMI as an indicator of own BMI: large intergenerational mortality study. BMJ 339 (December), b5043. Steckel, R.H., 1995. Stature and the standard of living. Journal of Economic Literature 33 (4), 1903–1940. Sun, Shumei S., Cameron Chumlea, W., Heymsfield, S.B., Lukaski, H.C., Schoeller, D., Friedl, K., Kuczmarski, R.J., Flegal, K.M., Johnson, C.L., Hubbard, V.S., 2003. Development of bioelectrical impedance analysis prediction equations for body composition with the use of a multicomponent model for use in epidemiologic surveys. American Journal of Clinical Nutrition 77, 331–340. U.S. Department of Health and Human Services, 1996. National Center for Health Statistics. The Third National Health and Nutrition Examination Survey (NHANES III, 1988–1994) Center for Disease Control, Washington, DC (CDROM). Wada, R., March 2005. Obesity, muscularity, and body composition: the puzzle of gender-specific penalty and between-ethnic outcomes. Andrew Young School of Policy Studies Working Paper, Georgia State University. Wada, R., May 2007. Obesity and physical fitness in the labor market. Unpublished Dissertation. Andrew Young School of Policy Studies, Georgia State University. Wada, R., Tekin, E., 2007. Body composition and wages. National Bureau of Economic Research Working Paper No. 13595, November.

254

R. Wada, E. Tekin / Economics and Human Biology 8 (2010) 242–254

Willett, K., Jiang, R., Lenart, E., Spiegelman, D., Willett, W., 2006. Comparison of bioelectrical impedance and BMI in predicting obesity-related medical conditions. Obesity 14 (March (3)), 480–490. World Health Organization, 1995. Physical status: the use and interpretation of anthropometry. World Health Organization Technical Report Series, 854.

World Health Organization, 1998. Obesity: preventing and managing the global epidemic: Report of a WHO consultation on obesity, Geneva, 3– 5 June 1997. World Health Organization Division of Noncommunicable Disease Programme of Nutrition Family and Reproductive Health, Geneva.