A Comparison of the Relationship Between Obesity and Earnings in the U.S. and Germany

Cawley, John; Grabka, Markus M.; Lillard, Dean R. Article A Comparison of the Relationship Between Obesity and Earnings in the U.S. and Germany Schmollers Jahrbuch – Zeitschrift für Wirtschafts- und Sozialwissenschaften. Journal of Applied Social Science Studies Provided in Cooperation with: Duncker & Humblot, Berlin Suggested Citation: Cawley, John; Grabka, Markus M.; Lillard, Dean R. (2005) : A Comparison of the Relationship Between Obesity and Earnings in the U.S. and Germany, Schmollers Jahrbuch – Zeitschrift für Wirtschafts- und Sozialwissenschaften. Journal of Applied Social Science Studies, ISSN 1865-5742, Duncker & Humblot, Berlin, Vol. 125, Iss. 1, pp. 119-129, https://doi.org/10.3790/schm.125.1.119 This Version is available at: https://hdl.handle.net/10419/292114 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Schmollers Jahrbuch 125 (2005), 119-129 Duncker & Humblot, Berlin A Comparison of the Relationship Between Obesity and Earnings in the U.S. and Germany By John Cawley, Markus M. Grabka, and Dean R. Lillard* Abstract This paper investigates and compares the relationship between obesity and earnings in the U.S. and Germany. Using data from the Panel Study of Income Dynarnics (U.S.) and the German Socio-Economic Panel, instrumental variables models are estimated that account for the endogeneity of body weight. We find that, in both countries, hea vier women tend to eam less. For example, obesity is associated with almost 20 percent lower earnings for U.S. and German women. We test for causality using IV models; these models suggest that weight may lower labor earnings for U.S. women. However, our IV results yield no evidence of a causal impact of weight on earnings for women in Germany or for men in either country. lEL Classifications: 171, 131, JIO, JJO lntroduction The prevalence of obesity has risen drarnatically in the last several decades in all developed countries. In the U.S., it rose from 15 percent of the popula tion during 1976-1980 to more than 30 percent during 1999-2002 (Flegal et al. 2002; Hedley et al. 2004). Data are more scarce in Germany, but the frac tion of obese men in West Germany rose from 17.4 percent during the period 1990-1992 to 19.4 percent in 1998 (Bergmann and Mensik 1999) and rose an additional percentage point between 1999 and 2003 (Statisisches Bundesamt 2004). With the rise in obesity has come interest in better understanding the impli cations and consequences of the condition. There are large literatures devoted to the health consequences of obesity (e.g. Field et al. 2002) as well as the social or emotional impact (Sobal 2004). Several studies have documented a * Acknowledgements: The authors thank participants at the 2004 SOEP Data Users Conference in Berlin for their helpful comments. They thank Shuaizhang Feng for re search assistance. Schmollers Jahrbuch 125 (2005) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12 120 John Cawley, Markus M. Grabka, and Dean R. Lillard negative correlation between obesity and wages or eamings among women (Averett and Korenman 1996; Pagan and Davila 1997; Cawley 2004). Cawley (2004) finds evidence that this relationship is causal (that is, that weight low ers wages) for white, but not black or Hispanic, females in the U.S. He also finds that observed negative correlations between wages and weight for other groups are likely the result of unobserved heterogeneity. Part of the relationship between weight and eamings may be determined by culture. For example, in societies in which obesity is more stigmatized, obese individuals may suffer lower self-esteem, leading to worse labor market per formance. Altemately, in such societies employers may be more likely to dis criminate against obese job applicants or obese employees. For these reasons, the relationship between weight and earnings is likely to differ across coun tries. This paper is the first to compare the relationship between weight and eam ings across countries, in particular, Germany and the United States. We pre sent a basic model that links weight to eamings. We use data from the German Socio-Economic Panel (SOEP) and the Panel Study of Income Dynamics (PSID) components of the Cross-National Equivalent File (CNEF) (Burkhau ser et al. 2001) to estimate regression models and compare the association between obesity and eamings in the two countries. Neither of these data sets has been previously used to study the relationship between weight and eam ings. 1. Conceptual Framework and Methods In this section we sketch a conceptual framework for understanding the relationship between body weight and earnings. Tue logic of the model ap plies as well to other outcomes ( employment, labor market eamings, and hours worked) that we examine in Cawley, Grabka, and Lillard (2004). Assume that log earnings are a function of body mass index (BMI) and other variables such as human capital. Research in behavioral genetics sug gests that roughly half of the variation in body mass index is due to non-genet ic factors such as individual choices and environment (Comuzzie and Allison 1998). In addition, obesity may be influenced by eamings, especially for adult females (Sobal and Stunkard 1989). For these reasons BMI is endogenous, and an OLS estimate of the coefficient on weight will not consistently esti mate the true effect of BMI on log eamings. We use the method of instrumental variables to address the endogeneity of weight. For IV to be successful, one must identify at least one instrument that is highly correlated with weight but not with the error term in the log eamings regression. The challenge of the instrumental variables method is to find valid instruments. Schmollers Jahrbuch 125 (2005) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12 Obesity and Earnings in the U.S. and Germany 121 We adopt the approach of Cawley (2004) and use as our instrument the weight of a family member; for the PSID, a child or parent, and in the SOEP, a parent. The weight of a child or parent is a powerful instrument for the weight of a respondent because each parent shares on average half of her genes with each of her biological children, and about half of the variation in weight is genetic (Comuzzie and Allison 1998). Our identifying assumption requires that the weight of a child or parent be uncorrelated with the respondent's earn ings residual. One rnight be concerned that they are correlated if they both depend on habits learned in the farnily household. However, studies have been unable to detect any effect of common household environment on body weight in samples comparing adopted and biological children (Comuzzie and Allison 1998; Grilo and Pogue-Geile 1991) or in samples of twins reared apart compared to twins reared together (Maes et al. 1997). lt is not possible to prove the null hypothesis of no effect of household environment on body weight; the repeated failure to reject the null hypothesis is the strongest evi dence that will ever be available. The weight of a child or parent would be correlated with the respondent's earnings residual if the genes that deterrnine weight and those that determine earnings are bundled in transmission from parent to child; the extent to which this occurs is unknown. Most estimates of the correlation between earnings of sons and fathers lie in the range .2 to .35 (Corak 2004). Note that this is the total intergenerational correlation, not an estimate of the genetic variation in earnings that is bundled with the genetic variation in weight, which is the only part that would be troubling for our IV method. 1 2. Data This study uses data from the PSID and the SOEP. Respondents report height and weight in survey years 1986, 1999, and 2001 in the PSID and in 2002 in the SOEP. When people report their weight and height, they often err (intentionally or unintentionally), and reporting error may bias coefficient esti mates (Judge et al. 1985). To correct for reporting error in the PSID we predict true height and weight with information on the relationship between true and reported values in the Third National Health and Nutrition Examination Survey (NHANES III) using the method outlined in Lee and Sepanski (1995) and Bound et al. (2002); an appendix detailing this procedure is available upon request. No parallel study was available to investigate or correct for 1 In Cawley, Grabka, and Lillard (2004), we estimate models that control for indivi dual fixed effects (FE), a specification which eliminates the influence of time-invariant unobserved heterogeneity. This is possible for the PSID because it contains up to three measures of weight for each respondent. Because the SOEP data (up to wave 2003) include only one measure of weight, we cannot estimate the FE model using those data. Schmollers Jahrbuch 125 (2005) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12 f f 1 � Table 1: Descriptive statistics of variables in the OLS and IV regression models: Mean and (standard deviation) Men Women U.S. (PSID) Germany (SOEP) U.S. (PSID) Germany (SOEP) OLS IV OLS IV OLS IV OLS IV Weight 83.79 (14.86) 82.39 (14.68) 83.91 (12.99) 82.64 (12.93) 77.37 (16.68) 75.89 (16.89) 67.17 (12.31) 66.90 (13.03) Height 1,75 (0,07) 1,75 (0,07) 1,79 (0,07) 1,79 (0,07) 1,60 (0,06) 1,61 (0,06) 1,66 (0,06) 1,68 (0,06) Underweight 0.00 (0.06) 0.01 (0.08) 0.01 (0.06) 0.01 (0.08) 0.01 (0.11) 0.01 (0.11) 0.03 (0.17) 0.04 (0.20) Overweight 0.40 (0.49) 0.38 (0.49) 0.45 (0.50) 0.41 (0.49) 0.31 (0.46) 0.32 (0.47) 0.26 (0.44) 0.21 (0.41) Obese 0.17 (0.38) 0.15 (0.35) 0.14 (0.34) 0.11 (0.31) 0.41 (0.49) 0.36 (0.48) 0.10 (0.30) 0.10 (0.30) BMl 26.57 (4.28) 26.09 (4.22) 26.30 (3.67) 25.71 (3.66) 29.11 (6.35) 28.48 (6.20) 24.40 ( 4.33) 23.84 (4.56) Age 36.68 (9.37) 31.80 (5.37) 43.52 (9.89) 33.51 (6.16) 36.74 (9.06) 32.28 (5.81) 43.03 (9.41) 32.23 (6.13) WestGerman resident - - 0.80 (0.40) 0.81 (0.39) - - 0.78 (0.42) 0.80 (0.40) Foreigner - - 0.13 (0.33) 0.14 (0.35) - - 0.11 (0.31) 0.09 (0.29) Black 0.22 (0.42) 0.25 (0.43) - - 0.29 (0.45) 0.36 (0.48) - - Hispanic 0.01 (0.11) 0.00 (0.06) - - 0.01 (0.12) 0.01 (0.08) - - Married 0.79 (0.41) 0.70 (0.46) 0.72 (0.45) 0.41 (0.49) 0.67 (0.47) 0.58 (0.49) 0.70 (0.46) 0.37 (0.48) No. of Children 1.14 (1.21) 1.08 (1.18) 0.82 (1.03) 0.66 (0.96) 1.12 (1.17) 1.20 (1.15) 0.71 (0.92) 0.59 (0.87) Lower education 0.35 (0.48) 0.35 (0.48) 0.08 (0.28) 0.10 (0.30) 0.37 (0.48) 0.35 (0.48) 0.10 (0.30) 0.10 (0.30) � i � i 8. � r. OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12 J 1 j Middle education 0.24 (0.43) 0.22 (0.42) 0.48 (0.50) Higher education 0.31 (0.46) 0.36 (0.48) 0.42 (0.49) Experience in years 17.21 (9.38) 12.38 (5.37) 12.16 (10.34) Full time employed - - 0.96 (0.20) Job tenure 8.32 (8.09) 5.80 (5.29) - Lor rearnings 10.15 (0.66) 10.07 (0.66) 10.45 (0.70) Interviewer present - - 0.59 (0.49) Weight of father -83.36 (14.68) - Height of father -1.74 (0.07) - Age of father -39 .46 (34.54) - Weight of mother -75.47 (13.38) - Height of mother -1.61 (0.07) - Age of mother -50.62 (29.77) - Weight of child -50.98 (20.56) - Height of child -1.48 (0.21) - Age of child -3.37 (4.64) - Sex of child -0.58 (0.80) - N 1716 833 6649 Source: SOEP 2002, PSID 1986, 1999, and 2001; authors' calculations. 0.58 (0.49) 0.27 (0.44) 0.28 (0.45) 0.28 (0.45) 0.27 (0.44) 0.30 (0.46) 7.13 (6.44) 14.32 (7 .80) 11.83 (6.01) 0.94 (0.24) - - -6.14 (6.15) 4.72 (4.83) 10.16 (0.70) 9.51 (0.90) 9.56 (0.90) 0.57 (0.49) - - 64.72 (36.50) -83.99 (14.29) 1.35 (0.73) -1.75 (0.07) 47.77 (26.47) -39.62 (34.08) 66.33 (22.51) -77.07 (13.70) 1.51 (0.43) -1.61 (0.07) 55.58 (17.79) -50.33 (28.01) - - 47.04 (18.72) - - 1.42 (0.20) - - 3.45 (4.43) - - 0.69 (0.83) 930 1698 906 0.53 (0.50) 0.54 (0.50) 0.34 (0.47) 0.31 (0.46) 9.86 (9.05) 5.94 (5.99) 0.53 (0.50) 0.63 (0.48) - - 9.68 (0.94) 9.58 (0.92) 0.59 (0.49) 0.53 (0.50) -66.61 (36.33) -1.38 (0.72) -47.29 (25.46) -69.11 (20.00) -1.56 (0.36) -55.65 (15.20) - - - - - - - - 5410 553 q· [ �- :: {lq s· g" C 0 [ j OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12 124 John Cawley, Markus M. Grabka, and Dean R. Lillard differences between true and reported weight among Germans. Kroh (2004) presents some evidence that the presence of an interviewer may affect the reporting error for height and weight. We therefore include a variable in our regressions that indicates whether an interviewer was present when the SOEP questionnaire was filled out. Our outcome of interest in this paper is the natural log of annual earnings. To characterize body weight we use three alternative measures: 1) weight in kilograms controlling for height in meters; 2) BMI; and 3) three indicator variables for clinical weight classification. These indicate when a person is underweight, overweight, and obese. BMI is calculated as weight in kilo grams divided by height in meters squared. Underweight is defined as a BMI of less than 18.5, overweight is defined as a BMI between 25 and 30, and obese is defined as a BMI of 30 or more. (The omitted category is healthy weight, which is defined as a BMI between 18.5 and 25.) Other regressors include: age, age squared, total work experience, number of children, job tenure with current employer, and indicator variables for educational attain ment, race (PSID), marital status, number of children (SOEP), student status (PSID), resident of West Germany (SOEP), non-German citizen (SOEP), em ployment status (SOEP) and whether the interviewer was present (SOEP).2 Finally, we restrict our sample to individuals aged 25 to 65 years with posi tive earnings in the previous year, including the self-employed and those working in the public sector. In Table 1 we present basic descriptive statis tics for the analysis samples of U.S. and German men and women. Due to missing data on the body weight and height of parents or children, fewer observations are available to estimate the IV models. However, the means of variables in the OLS and IV samples differ significantly only for age and marital status. 3 We use weight of each respondent's parents (both datasets) and/ or child (PSID) to instrument for respondent's weight.4 Results from the first-stage regression confirm that these are strong instruments for the weight of U.S. and German men and U.S. women but much less so for the weight of German women. The F-statistic associated with the hypothesis that the first-stage coef- 2 We do not include other covariates used in typical wage regressions like occupa tion or industry indicators because they are endogenous. Consequently, our estimates should be considered as estimates of the net correlation of weight with earnings after all behavioral choices influenced by weight have been made. 3 In case of the SOEP this sub-sample mainly consists of respondents who are living together with at least one parent. Therefore the result of the IV models for Germany are not fully comparable with those of the OLS estimates which are based on all respon dents with positive labor earnings. 4 In all those cases, a value of zero is assigned when weight is not observed for a given relative; such missing status is controlled for by means of a dummy variable. When weight is missing for all relatives, the observation is dropped. Schmollers Jahrbuch 125 (2005) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12 Obesity and Earnings in the U.S. and Germany 125 ficients on the instruments are jointly equal to zero was 46.31 and 11.15 for U.S. and German men respectively, 29.75 for U.S. women but only 4.65 for German women. The F-statistics for U.S. and German men and U.S. women exceed, but that for German women falls short of, the minimum F statistic of 10 suggested by Staiger and Stock (1997). The partial R-squared contributed by the instruments in the first stage was .05 (U.S. men), .09 (German men), .03 (U.S. women) and .06 (German women). 3. Empirical Results Table 2 presents results for OLS and IV models for both the U.S. and Ger man samples. Outcomes for the three measures of body weight are presented in separate vertical panels. Results for men are presented on the left-hand-side of Table 2, and those for women on the right-hand-side. Within each panel, each column represents a separate regression. We do not estimate IV models for the models reported in the bottom panel that use the indicator variables for clinical weight classification because we lack sufficient instruments to identi fy all three indicator variables for clinical weight classification. In the first column of Table 2, the OLS coefficients from all three specifica tions for men indicate that labor earnings of U.S. men are positively correlated with weight. For example, men who are 10 kg heavier than the mean tend to have 1.2 percent higher earnings. Men who are overweight or obese have roughly .09 log points higher earnings than men of normal weight while men who are underweight earn .43 log points less. This is consistent with McLean and Moon (1980), who find that weight is positively correlated with hourly earnings among men aged 51-65 in the National Longitudinal Survey ofMa ture Men in 1973. The authors attribute their finding to a "portly banker" effect -that for middle-aged American men, a large body size is a non-verbal signal of power that commands respect. In contrast, the OLS results for German men, which are presented in col umn 3, offer no evidence of a correlation between weight and earnings. We do find evidence of a strong and highly significant relationship between height and labor earnings, which is consistent with earlier research (Heineck 2004). Columns 2 and 4 present IV results for the PSID and SOEP samples of men.5 In all cases the coefficients on kilograms and BMI are not statistically different from zero. We are unable to reject the hypothesis that the labor earn ings of men in the U.S. or Germany are uncorrelated with body weight. s Some of the difference between the OLS and IV results arises because the samples used differ. Table 1 shows that the IV samples are younger and less overweight than the füll sample. Schmollers Jahrbuch 125 (2005) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12 f f 1 � Table 2: Coefficients and t-Statistics from Log annual labor earnings regressions Men Women U.S. (PSID) Germany (SOEP) U.S. (PSID) Germany (SOEP) OLS IV OLS IV OLS IV OLS IV Weight (kilograms) .0012** .0044 -.0009 -.0028 -.0021 ** -.0132** -.0046** -.0031 (.0005) (.0030) (.0006) (.0055) (.0006) (.0047) (.0008) (.0098) Height (meters) -.0030 -.0247 .9787** .7555 .0041 .0292 .6503** .4159 (.0063) (.0182) (.1140) (.7426) (.0082) (.0202) (.1688) (1.2826) Instrument statistics Change in adjusted R2 .045 .103 .034 .084 F-statistic 46.31 14.77 29.75 7.44 p-value .00 .00 .00 .00 Body mass index (BMI) .0081 ** .0310 -.0033 -.0119 -.0118** -.0802** -.0133** -.0022 (.0034) (.0213) (.0019) (.0179) (.0033) (.0277) (.0023) (.0276) Instrument statistics Change in adjusted R2 .054 .0722 .034 .050 F-statistic 44.42 9.6 28.94 4.77 p-value .00 .00 .00 .00 Underweight (BMI < 18.5) -.4327** -.0524 --.1134 --.0796 (.2208) (.1044) (.1718) (.0571) Overweight (25 < BMI < 30) .0902** --.Qlll --.1487** --.0859** - (.0305) (.0148) (.0518) (.0228) Obese (BMI > 30) .0930** --.0352# --.1986** --.1951 ** - (.0409) (.0213) (.0519) (.0327) N 1716 833 6649 930 1698 906 5410 553 Note: Numbers are rounded. # stands for significance at 10 percent level, * stands for significance at 5 percent level, ** stands for significance at I percent level. Numbers in parentheses are standard errors. Other variables that are included in the regressions are: age, age squared, education dummies, race dummies (PSID), marriage status, total work experience, job tenure with current employer (PSID), student dummy (PSID), number of children, resident of West Germany (SOEP), non German citizen (SOEP), interviewer present dummy (SOEP). Instruments used in the PSID N estimation are: first child's age, sex and BMI, whether first child's information is missing, father's age and BMI, whether father's information is missing, mother's age and BMI, whether mother's information is missing. Instruments used in the SOEP IV estimation are: father's age and BMI, whether father's information is missing, mother's age and BMI, whether mother's information is missing. Source: SOEP 2002, PSID 1986, 1999, and 2001; authors' calculations. � i � i 8. � r. OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.125.1.119 | Generated on 2023-04-04 12:34:12