Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Do employers discriminate against obese employees: evidence from individuals who
simultaneously work in self-employment and paid-employment?
Sankar Mukhopadhyay
Associate Professor
Department of Economics (MS – 030),
University of Nevada, Reno, NV, 89557.
Phone: 775-784-8017, Fax: 775-784-4728, Email: [email protected]
JEL Codes: I10, I12, I13, J7
Keywords: Self-employment, paid-employment, BMI, obesity, discrimination, wage,
compensation
Do employers discriminate against obese employees: evidence from individuals
who simultaneously work in self-employment and paid-employment?
Abstract: We test whether employers discriminate based on body-weight, by analyzing
wages of individuals who simultaneously work in paid-employment and self-employment (in the
same occupation). We explore whether difference in the wages between these two types of jobs is
related to body weight (BMI). While lower productivity and/or customer discrimination against
higher BMI individuals may affect wages in both types of jobs, employer discrimination cannot
affect wages of solo-entrepreneurs.
Our non-parametric estimates suggest that the wage differential (paid-employment wage –
self-employment wage) is negative and small for White women with BMI in low 20’s. This
differential decreases with BMI, and the differential is minimum for White women with BMI
around 30, suggesting employer discrimination. The wage differential starts narrow as the BMI
increases further. Somewhat surprisingly, White women with BMI 40 or more do not face
employer discrimination. This is true for both hourly wages and hourly compensation, and holds
even after controlling for large number of covariates, including whether they have employer
provided health insurance. Our results suggest that protection the category-II and category-III
obese White women may be coming at the cost of overweight and category-I obese White women.
Results for other demographic groups are less consistent.
JEL Codes: I10, I12, I13, J7
Keywords: Self-employment, paid-employment, BMI, obesity, discrimination, wage,
compensation
1. Introduction There has been dramatic increase in the obesity rates throughout the developed world in
the last few decades. For example, the obesity rate in the United States has doubled in the last 30
years from less than 15% in 1980 to 34.9% in 2011 (Ogden et. al. 2014). There is a large economics
literature estimating the impact of body-weight on earnings. A number of studies (Register &
Williams, 1980; Averett & Korenman, 1996; Pagan & Davilla, 1997; Behrman and Rosenzweig,
2001; Cawley, 2004; Atella et al., 2008; Johar & Katayama, 2012; Cawley & Meyerhoefer, 2012;
Sabia and Rees, 2012; Larose et al, 2016 among others) have reported that higher body weight is
associated with lower earnings. This result is most consistent for White women; however, some
studies found that the result holds for other race-gender categories. Several hypotheses have been
proposed to explain this relationship (Cawley, 2015). One possible reason is that body weight
affects marginal productivity (Baum and Ford 2004). Han, Norton, and Powell (2011) found that
obese individuals invest less in human capital and/or choose occupations that require less
interaction with public, which in-turn results in an indirect wage penalty. It is also possible that
body weight affects healthcare expenditures that employer’s pay, and in turn employers reduce the
wages of overweight and obese workers to compensate for extra healthcare spending
(Bhattacharya and Bundorf, 2009). Another explanation is that lower wages for heavier individuals
is a result of discrimination (Atella, Pace, and Vuri, 2008). Kline and Tobias (2008) and Gregory
and Rhum (2011) report that earnings of White women decline with BMI, even when they have
BMI in normal or healthy (18.5-25) range. Kline and Tobias (2008) report that the slope of BMI-
wage locus is highest when BMI is between 23 and 28. The decline in women’s wage with BMI,
even in the “healthy” BMI range, suggests that the negative relation between BMI and wages may
be a result of discrimination. These results (Cawley 2004, Kline and Tobias 2008, Gregory and
Rhum 2011 among others) also suggest that it is a body-weight penalty as opposed to an obesity
penalty. Several studies report that there is a stronger weight penalty for women in occupations
with significant interpersonal tasks (Pagan and Davila 1997; Baum and Ford 2004; Han, Norton,
and Stearns 2009), suggesting possible customer discrimination.
A number of epidemiological studies find that overweight and obese employees are more
likely to report employer discrimination (Puhl and Brownell 2006; Puhl, Andreyeva, and Brownell
2008; Roehling et al., 2008). However, these studies focus on the perception of the employees and
not on objective measures. Another strand of research has used laboratory experiments to
document weight-based discrimination in employment related settings. See Roehling et al. (2008)
for a meta-analysis on this topic. See Puhl and Heuer (2009) for a comprehensive review of weight
discrimination in the workplace.
These studies have prompted discussion about the legal protection against such
discrimination (Pomeranz, 2008). Body weight is not a protected category in the U.S. under the
Civil Rights Act of 1964. However, in some cases aggrieved individuals have used the Americans
with Disabilities Act of 1990 (ADA) to sue employers (Pomeranz and Puhl 2013). Prior to 2009,
most of these lawsuits were not successful since neither the Equal Employment Opportunity
Commission (EEOC) nor the courts considered obesity a protected category under the ADA
(Pomeranz and Puhl 2013). However, in 2008, Congress amended the ADA with the Americans
with Disabilities Act Amendments Act of 2008 (ADAAA) to include severe obesity1 as a protected
category. This leaves anyone who is not severely obese without any legal protection from weight-
based discrimination. Individuals who are obese (but not severely obese) or even overweight may
face discrimination in the labor market. As we discussed, there is a substantial amount of evidence
that higher body weight is associated with worse labor market outcomes, not only in severely
obese, but also in overweight, and in some cases, normal weight White women . However, from
the current literature, it is not clear whether employer discrimination is responsible for such
association. Nonetheless, given the volume of evidence about negative association between BMI
and wages, an act modeled after the Age Discrimination in Employment Act of 1967 (ADEA) has
been proposed to prevent weight discrimination (Pomeranz, 2008). This act, the Weight
Discrimination in Employment Act (WDEA) is hypothetical at this stage, but public polling
suggests considerable support in favor of such a law (Puhl and Heuer, 2011).
In this paper, we take a novel approach to test whether employers discriminate based on
weight. We use data from The National Longitudinal Survey of Youth 1997 cohort, which
interviewed respondents about multiple job holdings. In each survey, each respondent can report
up to eight distinct jobs held during the previous calendar year. We focus on individuals who
simultaneously work in paid-employment and self-employment. We restrict our attention to
1 Defined as a body weight of more than 100 percent over the norm. Since then Equal Employment Opportunity
Commission (EEOC) has treated category-III obesity (BMI>40) as a disability and have successfully sued
employers for discriminating against obese employees. However, in a recent case (Morriss v. BNSF Railway Co,
2016) the Eighth Circuit Court ruled that even category-III obesity is not a protected category under the ADAAA
unless it is also associated with an underlying psychological condition.
individuals who are solo entrepreneurs (i.e. they do not have any employees or business
associates). This restriction makes in more likely that wage from self-employment can be taken as
a proxy for productivity. We explore whether wage differential between these two types of jobs is
related to body weight (BMI). While lower productivity and/or customer discrimination against
higher BMI individuals may affect wages in both types of jobs, employer discrimination cannot
affect wages of solo-entrepreneurs. Since the same person is earning two different wages from two
different jobs at the same time, in the same geographic location, and in the same occupation, we
are differencing out the effect of observable and unobservable components of human capital and
other factors that may affect wages. An exception may be employer provided health insurance and
we control for it in our analysis. We acknowledge that wage (or compensation) in paid-
employment may not reflect current productivity in the presence of deferred compensation
practices. However, in that case, there should not be any systematic relation between the wage
differential and BMI. Therefore, the relation between BMI and the difference from these two types
of wages can provide us with important information about employer discrimination. We are not
aware of any previous studies that have compared wages from self-employment and paid-
employment to infer about weight discrimination.
Our non-parametric estimates suggest that there is a systematic relation between BMI and
wage differential (defined as paid-employment wage – self-employment wage). This wage
differential is small but negative for White women with BMI in low 20’s. The absolute value of
this wage differential increase with BMI and the differential is maximum (i.e. the differential is
minimum) for White women with BMI in their low 30’s. The wage differential starts narrow as
the BMI increases further and is positive for White women with BMI 40 or more. This is true for
both hourly wages and hourly compensation and holds even after controlling for whether they
have employer provided health insurance or not. Since this wage differential cannot be explained
by either productivity based explanation or even customer discrimination, we conclude that this
suggests weight discrimination by employers. However, our results suggest that not all obese
women face discrimination. While White women who are overweight (25<=BMI<30) or
category-I obese (30<=BMI<35) seems to face most discrimination, category-III obese
(BMI>40) women do not face discrimination. If anything, employers may be subsidizing the
wages of category-III obese White women. This may suggest employers do not want to appear
discriminatory when such discrimination is most visible. It may also imply employer believe that
that women who are severely obese have a better chance of winning employment discrimination
lawsuits, especially after the ADAAA. We find suggestive evidence to that effect. We do not
find any evidence of employer discrimination against Black or Hispanic women based on weight.
However, we do find some evidence of weight discrimination against Black and Hispanic men,
which are similar to White women. In white men, the under-weight individuals seem to face
most discrimination. However, the results for groups other than White women are not consistent.
Rest of the paper is organized in the following way. Section 2 describes the data, Section
3 presents empirical results, and Section 4 concludes.
2. Data We use data from The National Longitudinal Survey of Youth 1997 cohort (NLSY97). The
NLSY97 is a nationally representative sample of 9,022 youths, aged 12-16 as of December 31,
1996. Interviews were conducted annually until 2011, and bi-annually thereafter. In the first three
waves (1997, 1998 and 1999) relevant questions about multiple job holding were not asked
presumably because respondents were too young. We use all of the waves of data where questions
about multiple job holding and other relevant variables were asked (2000 to 2015). The most recent
round of interviews (2015) included 6795 of the original respondents.
In each wave, respondents were asked about up to eight jobs they held in the year
preceding an interview. We focus on respondents with multiple jobs. Although multiple job
holding2 is not a rare phenomenon, (about 5% of all workers in the U.S. hold multiple jobs), why
individuals hold multiple jobs is not well understood. Common explanations suggest that
employees may be constrained by hours (Paxson and Sicherman, 1996) or they want a task or
skill diversification (Conway and Kimmel, 1998; Panos et al., 2014) or they are insecure about
their primary job (Bell et al., 1997; Panos et al., 2009). While overall multiple job holding rates
have remained stable, female multiple job holding rates have steadily increased from 2.2 percent
in 1970 to 5.7 percent in 2007 (Amuedo-Dorantes and Kimmel 2009). We focus on a particular
type of multiple-job holders: those who are simultaneously working in paid-employment and
self-employment. The primary job may be either in self-employment or in paid-employment. If
2 Sometimes referred to as moonlighting
they worked in more than one paid-employment type job then we keep the job they reported first
during interviews.
We use self-reported height and weight to compute BMI. We calculate the BMI as the
weight measured in pounds divided by the height measured in inches multiplied by the factor
703.0696. In our data, BMI varies from 12.01 to 71.09. The 1st, 95th, and 99th percentile of the
distribution are 17.43, 39.53, and 49.01 respectively. These numbers are consistent with Strum
and Hattori (2013) who reported that 6.6% of U.S. adults have a BMI of 40 or above. We restrict
our attention to BMI between 17.43 (the first percentile of the BMI distribution) and 49.01 (the
99th percentile of the BMI distribution). Our outcome variables are hourly wage rate and total
compensation per hour for each job. These variables were created by NLSY staff from raw data3
. In the documentation, users are advised that sometimes these variables can take “extremely low
or extremely high” values. To reduce the influence of outliers we exclude wage observation that
are below $1 or above $100 per hour (roughly equivalent to 1st and 99th percentile of wage
distribution).
Since we are implicitly assuming that wage from self-employment is a measure of
productivity; we exclude self-employed individuals with paid employees or business associates.
In other words, we include only solo-entrepreneurs. After imposing all of these restrictions, we
have 1761 (1806) person-year observations from 899 (973) women (men). In Table 1, first three
columns present the summary for women and the last three are for men. In the text, we discuss
the descriptive statistics for White women only. Column 1 shows that among White women, the
average real hourly wage in paid-employment (self-employment) is $10.50 ($15.45) per hour (in
2003 prices). Average real hourly compensation wage in paid-employment (self-employment) is
$12.20 ($16.13) per hour (in 2003 prices). Average BMI is about 25. Average age is 24.6 years
and they have 14.7 years of education. About 28% are married and 23% have children. About
39.2% have employer provided health insurance.
3 The variable name (in NLSY97) for hourly wage variable is “CV_HRLY_PAY_”. For the compensation variable it
is “CV_HRLY_COMPENSATION_”. The primary difference between these variables is that “CV_HRLY_PAY_”
does not include overtime, tips, and bonuses but “CV_HRLY_COMPENSATION_” does. For more details on
construction of these variables please see http://nlsinfo.org/content/cohorts/nlsy97/topical-guide/employment/wages
Table 1: Summary statistics
Women Men
White Black Hispanic White Black Hispanic
mean/sd mean/sd mean/sd mean/sd mean/sd mean/sd
Wage in paid emp. 10.499 9.487 10.648 12.581 10.339 11.565
(6.594) (5.338) (7.872) (7.975) (7.264) (6.746)
Wage in self emp. 15.450 17.094 13.310 21.442 20.410 22.035
(16.400) (17.876) (13.964) (18.552) (20.053) (19.127)
Comp. in paid emp. 12.202 16.061 12.276 13.503 11.577 12.730
(12.235) (104.026) (11.990) (8.673) (8.439) (8.005)
Comp. in self emp. 16.127 18.293 14.287 21.942 21.246 22.709
(17.429) (20.462) (15.655) (18.912) (20.559) (19.695)
BMI 24.985 28.228 27.492 26.457 27.341 27.545
(5.363) (6.733) (6.149) (4.989) (5.453) (5.464)
Age 24.596 24.901 24.776 25.127 25.023 25.094
(4.466) (4.518) (4.605) (4.325) (4.316) (4.329)
Yrs. Of Educ. 14.698 13.049 13.000 13.748 12.666 12.537
(2.437) (2.339) (2.537) (2.607) (2.100) (2.258)
Work exp. 8.222 8.896 7.780 9.067 8.916 8.409
(6.126) (5.956) (6.335) (5.934) (5.929) (6.348)
Employer health ins. 0.392 0.384 0.336 0.384 0.378 0.406
(0.488) (0.487) (0.473) (0.487) (0.485) (0.492)
Married 0.277 0.130 0.295 0.234 0.124 0.220
(0.448) (0.337) (0.457) (0.423) (0.330) (0.415)
Have children 0.231 0.462 0.422 0.156 0.184 0.294
(0.422) (0.499) (0.495) (0.363) (0.388) (0.456)
Observations 1116 380 265 1109 347 350
As we have discussed before, wage penalty due to weight may vary across occupations.
Since our identification strategy relies on comparing wages from self-employment and paid-
employment, we create samples with the restriction than a woman must be working in “same” or
“comparable” occupations in self-employment and paid-employment. In NLSY, occupation for
each job is reported in 2002 Census occupational codes. We restrict our sample to individuals
who work in the same four-digit occupation in paid-employment and self-employment. We refer
to this sample as OCC-4D sample. We prefer this specification since it is the most detailed
occupational information available and therefore theoretically most appropriate. One problem
with this restriction is that it severely restricts sample size. For example, after imposing this
restriction we have 103 observations in our White women sample.
As an alternative, following Acemoglu and Autor (2011), we use Occupational
Information Network (O*NET) to define occupational similarity. Since a number of previous
papers have reported that the BMI penalty is higher in jobs with significant social interaction
component, we use O*NET work activity importance scale to determine how important social
tasks are in an occupation. O*NET assigns an importance value to 42 work activities. Since
NLSY97 occupations are reported in 2002 Census four digit codes, we convert them into Census
1990 code (using Autor and Dorn, 2013). Then we follow Acemoglu and Autor (2011) and
assign a value between one and five for each activity. For each occupation, we compute
importance of “non-routine-cognitive-interpersonal” (NRCI) tasks in that occupation. We take
this as a proxy for how social an occupation is. We compute the difference in the importance of
NRCI tasks between two occupations in paid-employment and self-employment. Then we take
the absolute value of this difference. Median of this variable (for all race-gender categories
combined) in 0.75, and 25th percentile is 0.31. One advantage of this definition is that the
underlying running variable (difference in the importance of NRCI tasks) is continuous.
Therefore the results can be checked for various cut-offs. We tried various cut-offs between
bottom 25th and 50th percentile. We report results from two samples using two cut-offs for this
variable. First, we define two occupations as comparable if the absolute value of the difference in
the importance of NRCI tasks is in bottom quartile (267 observations for White women). We
refer to this sample as NRCI-1 sample. Next, we define median as the cut-off (NRCI-2 sample).
The sample size for White women sample increases to 512. The qualitative results are not
sensitive to the choice of cut-off in this range.
3. Results In this section, we empirically investigate whether employers discriminate based on
weight of employees. In Section 3.1, we estimate (first using OLS and then semi-parametrically)
the association between BMI and wages in paid-employment, and in self-employment. Since, we
use observational data, and we do not use instruments these results represent association between
BMI and wages. Next in Section 3.2, we explore how BMI affects the wage differential between
paid-employment and self-employment. Since the same individual is earning two different wages
at the same time, in the same location, in the same occupation, we are differencing out the effect
of observable and unobservable components of human capital and other factors that may affect
wages. Thus, these estimates may be interpreted as the causal relation between BMI and wage
differential between two sectors. In Section 3.3, we check the robustness of our empirical results.
In Section 3.4, we explore the role of the ADAAA.
3.1. BMI and wages in paid-employment and self-employment
Table 2 presents the OLS estimates where BMI enters the wage equation in a linear way.
This specification has been widely used in the literature. Panel A presents the results for White
women. Column 1 shows that in paid-employment, one unit increase in BMI is associated with
1.02 percent reduction in hourly wage among White women. This is similar to the estimate
reported in Cawley (2004). Second column of Panel A shows that in self-employment, one unit
increase in BMI is associated with 1.50 percent reduction in hourly wage among White women.
Thus, the linear model suggests that the negative association between wage and BMI is present
among White women in both paid-employment and self-employment. Columns 3 and 4 presents
the results for hourly compensation. The results are similar. Among white women, one unit
increase in BMI is associated with 1.45 percent decline in total compensation in paid-
employment sector, and 1.59 percent reduction in total compensation in self-employment sector.
Panel B (C) present the results for Black (Hispanic) women. Estimates suggest that the
association between BMI and wage is not statistically significant for either Black or Hispanic
women. However, the association between hourly compensation and BMI is negative and
significant for Black women in paid-employment (column 3 of Panel B).
Panels D, E, and F present the results for White, Black, and Hispanic men respectively.
Estimates suggest that the association between BMI and wage is not statistically significant for
either White or Black men. However, the association between hourly compensation and BMI is
negative and significant for Hispanic men in self-employment (columns 2 and 4 of Panel F). The
association between BMI and hourly compensation is positive and significant for Hispanic men.
Cawley (2004) reported a similar result for wages of Black men.
Table 2: Association between BMI and hourly wage or compensation (linear model)
Hourly wage Hourly compensation
Paid-emp. Self-emp. Paid-emp. Self-emp.
Panel A: White women
BMI -0.0102*** -0.0150** -0.0145*** -0.0159**
(-2.941) (-2.090) (-4.636) (-2.219)
Observations 1,116 1,116 1,116 1,116
Panel B: Black women
BMI -0.00560 -0.00624 -0.00946*** -0.00605
(-1.488) (-0.661) (-2.719) (-0.621)
Observations 380 380 380 380
Panel C: Hispanic women
BMI -0.000745 -0.00139 -0.00438 -0.00283
(-0.110) (-0.123) (-0.671) (-0.246)
Observations 265 265 265 265
Panel D: White men
BMI -0.00515 0.000303 -0.00351 -0.000180
(-1.220) (0.0455) (-0.834) (-0.0270)
Observations 1,109 1,109 1,109 1,109
Panel E: Black men
BMI 0.00444 0.00416 0.00332 0.00575
(0.940) (0.349) (0.607) (0.483)
Observations 347 347 347 347
Panel F: Hispanic men
BMI 0.00387 -0.0201* 0.0130** -0.0209*
(0.654) (-1.879) (1.973) (-1.933)
Observations 350 350 350 350
Note 1: control variables included but shown in tables include age (quadratic), years of
education, work experience (quadratic), marital status, dummy for whether they have children,
and year dummies.
Note 2: cluster robust t-stat in parenthesis. *** Signifies statistically different from zero at the
1% level, **signifies statistically different from zero at the 5% level and *signifies statistically
different from zero at the 10% level.
Next, we relax the linearity assumption and estimate a semi-parametric model where
BMI enters the wage (compensation) equation non-parametrically. We estimate the following
equation
𝑙𝑛𝑤𝑖 = 𝛾𝑋𝑖 + 𝑓(𝐵𝑀𝐼𝑖) + 𝜀𝑖
Where 𝑋𝑖 is a set of control variables (such as age, education, work experience, marital
status, whether they have children etc.). BMI can affect wage in non-linear way, which is
represented by the function 𝑓(. ). We still treat BMI as an exogenous variable. We use
Robinson’s (1988) double residual estimator.4 We estimate this equation separately for wage
from paid-employment and wage from self-employment. Figure 1 presents the results from
semiparametric regressions for six race-gender combinations.
Results for White women (Panel A) suggest that relation between BMI and wage is
inverse and approximately linear in paid-employment until BMI reaches mid 30’s, which is
consistent with Kline and Tobias (2008). It also shows that relation between BMI and wage in
self-employment is similar to that in the paid-employment until BMI reaches mid 30’s. After
that, wage declines with BMI at a higher rate in self-employment compared to paid-employment.
Our results for other race-gender categories are mixed. The semi-parametric components are
relatively stable with BMI for Black (Panel B) and Hispanic (Panel C) women. However, results
for Black men (Panel E) suggests steep drop in wages for self-employed Black men with BMI
above 40.
4 For estimation we use the Stata command “semipar” developed by Verardi and Debarsy (2012)
Figure 1: Association between BMI and hourly wage
Our semiparametric results show that BMI is associated with paid-employment wages for
all levels of BMI in White women. This is consistent with previous literature. Since it is hard to
conceive that an increase in BMI in the healthy range (for example from 21 to 22), would have a
detrimental effect on productivity or would increase expected healthcare cost, the pre-dominant
explanation for this result is discrimination. However, it is not clear whether it is customer
discrimination or employer discrimination. To investigate this further, we explore the relation
between BMI and the differential between the wages from paid employment and self-
employment.
3.2 BMI and differential in wages between paid-employment and self-employment
Under the assumptions that i) individuals are paid according to productivity, and ii) wage
from self-employment also reflects the productivity in paid-employment sector, the differential
between the wages from paid employment and self-employment may reflect whether paid-
employees face discrimination (if the differential is negative) or favorable treatment (if the
differential is positive). We note that wage in paid-employment may be lower than productivity
because of reasons other than discrimination (such as deferred pay or delayed compensation
contract). However, in those scenarios, there should not be any systematic relation between wage
differential and BMI. Our focus is whether there is a systematic relation between wage
differential and BMI.
In Figure 2, we present results from kernel weighted local polynomial regressions
between BMI and wage differential. We should note that there are 103 (145) observations in
White women (men) OCC-4D sample. Sample sizes for Black and Hispanic OCC-4D samples
even smaller. There are 24 (15) observations in Black (Hispanic) female OCC-4D sample and 27
(39) observations in Black (Hispanic) male OCC-4D sample. Therefore, we also report the
results for NRCI-1 sample, which not only has a bigger sample size, but also uses an alternative
definition to control for occupational differences across sectors. There are 267, 88, and 67
observations respectively in White, Black, and Hispanic women NRCI-1 sample. There are 386,
126, and 129 observations respectively in White, Black, and Hispanic men NRCI-1 sample.
Therefore, even though we report the results for Black (Hispanic) samples they should be treated
with caution.
In these regressions, we do not include any control variables since their effects should be
differenced out. Nonetheless, in Section 3.3.4 we report results after including controls. Figure 2
has six panels for six race-gender categories. In each panel, the bold line represents the results
for OCC-4D sample (same four-digit occupation) and the dashed line represents NRCI-1
(absolute value of difference in importance of NRCI tasks across occupations is in the bottom
quartile) sample. Panel A presents the results for White women. In both OCC-4D and NRCI-1
samples, the wage differential is negative for low-BMI women. It declines until BMI reaches low
30’s, and then it increases with BMI. The wage differential becomes positive after BMI reaches
low 40’s. Results for Black (Panel B) and Hispanic (Panel C) women suggest no relation
between BMI and wage differential in OCC-4D samples. The nature of relationship between
BMI and wage differential among Black (Panel E) and Hispanic (Panel F) men is somewhat
similar to that of White women (Panel A). On the other hand, among White men, the relation
between BMI and wage differential is bell-shaped in OCC-4D sample but “U-shaped” in NRCI-1
sample. We also plotted these lines for various other samples. We do not include all of them in
this graph to reduce clutter, but discuss some of them below. Only in the White women sample,
the relation between BMI and wage differential is consistent across different samples. This is
consistent with previous literature.
Figure 3 presents the results from kernel weighted local polynomial regressions along
with 95% confidence intervals for our preferred (same four-digit occupation or OCC-4D sample)
sample i.e. the bold lines in figure 2. The results suggest that for White women (Panel A) the
differential in wage is “U-shaped”. The differential is maximum around a BMI of 30. It is
negative and significant (at 95%) until BMI reaches high 30’s. We find similar results for Black
and Hispanic men. On the other hand, among White men, the wage differential is highest for
underweight White men (BMI<18.5).
Figure 2: BMI and difference in hourly wage between paid-employment and self-employment:
controlling for occupational differences
Figure 3: BMI and difference in hourly wage between paid-employment and self-employment
(same four-digit occupation)
3.3. Robustness checks
3.3.1 Alternative ways of controlling occupational differences
Figure 4 shows the results along with the 95% confidence intervals for White women for
various samples. To economize on space, we only present the estimates for White women.
Results for other race-gender categories are available upon request. Once we take into account
95% confidence interval, we do see a small but significantly negative difference in the wages
earned in paid-employment and self-employment (Panel A, i.e. no control for differences in
occupations; sample size 1116), when BMI is in 20’s. Panel B is a reproduction of Panel A of
Figure 3 (sample size 103). Panels C and D presents the results from NRCI-1 (267 observations)
and NRCI-2 (512 observations) samples respectively, and they are similar to Panel B. Our results
suggest that while employers may be discriminating against overweight and category-I obese
White women, they may be subsidizing the wages of category-III obese White women.
3.3.2 Allowing for skill prices to vary across sectors
The assumption that wage from self-employment reflects productivity in that sector is
plausible, especially given the fact that these are solo-entrepreneurs. The assumption that it also
reflects the productivity in paid-employment is more debatable. It assumes that human capital is
valued equally in both sectors. Clearly, it is more plausible when respondents are working in
same occupation. Nonetheless, if human capital were valued differently in paid-employment and
self-employment then the differential in wage would include differences in prices of human
capital, along with possible discrimination. However, results presented in Figure 2 do not
account for that. Therefore, we compute the difference between the two non-parametric
components presented in Panel A of Figure 1. These two come from two different regressions
(𝑓(𝐵𝑀𝐼)) from the paid-employment wage regression - (𝑓(𝐵𝑀𝐼) from the self-employment
regression) which allow for all coefficients of control variables (including those pertaining to
human capital of an individual) to differ across paid-employment and self-employment. Figure 5
shows that these two differentials are similar and they vary with BMI in a similar way in all four
panels. Again, to economize on space, we only present the estimates for White women. Results
for other race-gender categories are available upon request.
Figure 4: BMI and difference in hourly wage between paid-employment and self-employment
(White women)
Figure 5: Comparing raw wage differential and differential between semi-parametric components
(White women)
3.3.3. Using hourly compensation differential as outcome
As we have discussed before, hourly wage do not include overtime and bonus pay, which
may be systematically different in paid-employment and self-employment. Therefore, we check
whether the relation between compensation differential and BMI is similar to that between wage
differential and BMI. Figure 6 presents the results for OCC-4D sample for all race-gender
categories. Figure 6 is similar to Figure 3, which is the corresponding figure with wage
differential as outcome.
3.3.4 Parametric regressions
We also performed parametric regressions to check the relationship between wage
differential and BMI. Given the kernel weighted local polynomial regressions, it is clear that the
relation is non-linear in most cases. We approximate it using a quadratic function of BMI in the
parametric regressions.
∆𝑙𝑛𝑤𝑖 = 𝛼𝐵𝑀𝐼𝑖 + 𝛽𝐵𝑀𝐼𝑖2 + 𝛾𝑋𝑖
Where ∆𝑙𝑛𝑤𝑖 is the difference in (log) wage between paid-employment and self-employment
and 𝑋 is a set of control variables (such as age, work experience, education etc.). Theoretically,
in most cases, effects of 𝑋 should be differenced out. A potential exception is employer provided
health insurance. Since obese employees are likely to have higher healthcare cost, therefore
employers may reduce the wages of obese employees to compensate for added healthcare cost.
This would create a negative association between BMI and wages (Bhattacharya and Bundorf,
2009). While Cawley (2015) and others have argued that such a scenario is unlikely, it may be
important to control for whether they have employer provided health insurance.
Table 3 presents the results for OCC-4D sample. Panel A is for White women. Columns
1 and 2 present the results for hourly wage and columns 3 and 4 for hourly compensation.
Columns 1 and 3 do not include any control variables (𝑋), but columns 2 and 4 do. Results in
column 1 show that the coefficient of BMI is negative and coefficient of BMI squared is
positive, i.e., relation between BMI and differential in wages is ‘U’-shaped. This is consistent
with Panel A of Figure 4. Results in columns 2 suggest that this result is robust to adding the full
set of controls (including whether they have employer provided health insurance). Columns 3
and 4 suggest that results are similar when we use hourly compensation instead of hourly wage.
Panels B (Black women) and C (Hispanic women) suggest that there is no relation
between BMI and wage differential for these race-gender categories. Among White men (Panel
D), coefficient of BMI is positive and BMI squared is negative, although the estimates lose
significance when we include full set of controls. Among Black men, the relation between BMI
and wage-differential is U-shaped and similar to that in White women. Again, the estimates lose
significance once we add full set of controls. The estimates in columns 2 and 4 should be treated
with caution. The sample sizes in Black and Hispanic samples are too small to estimate the
model with full set of parameters (22 parameters). That is the reason columns 2 and 4 of Panel C
do not have t-stats below the point estimates.
Figure 6: BMI and compensation differential
Table 3: BMI and wage/compensation differential: controlling for observables
Wage diff. Wage diff. Comp. diff. Comp. diff.
Controls No Yes No Yes
Panel A: White women
BMI -0.243*** -0.230* -0.218*** -0.219*
(-2.770) (-1.845) (-2.627) (-1.954)
BMI sq. 0.00415*** 0.00403** 0.00372*** 0.00379**
(3.022) (2.004) (2.855) (2.106)
Observations 103 103 103 103
Panel B: Black women
BMI -0.0689 -0.671 -0.0586 -0.654
(-0.359) (-1.660) (-0.319) (-2.398)
BMI sq. 0.00121 0.0105 0.000941 0.0103
(0.377) (1.488) (0.308) (2.173)
Observations 24 24 24 24
Panel C: Hispanic women
BMI -0.0273 0.885 0.0619 -0.0287
(-0.163) - (0.331) -
BMI sq. 0.000773 -0.0149 -0.000988 -0.000731
(0.243) - (-0.280) -
Observations 15 15 15 15
Panel D: White men
BMI 0.228* 0.154 0.256** 0.170
(1.832) (1.066) (2.063) (1.208)
BMI sq. -0.00440* -0.00298 -0.00489** -0.00326
(-1.947) (-1.152) (-2.172) (-1.304)
Observations 145 145 145 145
Panel E: Black men
BMI -0.954*** -0.531 -0.970*** -0.514
(-2.836) (-0.844) (-2.977) (-0.895)
BMI sq. 0.0148*** 0.00980 0.0150*** 0.00973
(2.836) (1.074) (2.967) (1.180)
Observations 27 27 27 27
Panel F: Hispanic men
BMI -0.169 -0.138 -0.124 -0.0595
(-0.570) (-0.380) (-0.414) (-0.149)
BMI sq. 0.00349 0.00334 0.00278 0.00209
(0.741) (0.584) (0.585) (0.332)
Observations 39 39 39 39
Note 1: control variables included: age (quadratic), years of education, work experience
(quadratic), marital status, whether they have children, whether they have employer provided
health insurance, and year dummies.
Note 2: cluster robust t-stat in parenthesis. *** Signifies statistically different from zero at the
1% level, **signifies statistically different from zero at the 5% level and *signifies statistically
different from zero at the 10% level.
3.4 Role of the ADAAA
In this section, we explore whether the relationship between BMI and wage differential
changed after the ADAAA of 2008. It is important to note that since 2008 Equal Employment
Opportunity Commission (EEOC) has treated category-III obesity (BMI>40)5 as a disability.
However, in a recent case (Morriss v. BNSF Railway Co, 2016) the Eighth Circuit Court ruled
that even category-III obesity is not a protected category under the ADAAA unless it is also
associated with an underlying psychological condition. Therefore, the legal implication of
discriminating against an obese employee is unclear at this point. However, it is important to
point out that apart from the federal laws; some states (such as New York, New Jersey,
Michigan) have state laws protecting obese employees. We report the results from kernel
weighted local polynomial regressions between BMI and wage differential in Figure 7 for White
women. In each panel, the dashed line represent the results for before 2008 (2000-2008, both
included) period, and the bold line represents the results for after 2008 (2009-2015, both
included) period. About 45% of our observations are from post-2008 years. Once we account for
occupational differences, there is some suggestive evidence that the wage differential at the top
end of the BMI distribution have increased. At the same time it seems that the differential has
decreased for overweight and category-I obese women. This may suggest that protection the
category-II and category-III obese women is coming at the cost of overweight and category-I
obese women. However, the patterns are not consistent across samples.
5 https://www.eeoc.gov/eeoc/newsroom/release/4-10-12a.cfm
Figure 7: Did the ADAAA change the nature of employer discrimination?
4. Discussion and Conclusion
We analyze wages (compensation) of individuals who are simultaneously working in
paid-employment and self-employment. We focus on solo-entrepreneurs, who are working in
same four-digit 2002 Census occupation in both paid-employment and self-employment. Since
the same individuals are working in same occupation, at the same time, in the same geographic
location, we may expect that productivity (and therefore wage) would be similar in both sectors.
Under the assumption that wage from self-employment reflects productivity the differential
between the two wages may provide information regarding discrimination (or favoritism). We
acknowledge that wage (or even compensation) in paid-employment may not reflect current
productivity in the presence of deferred compensation practices. However, in that case there
should not be any systematic relation between the wage differential and BMI. An exception may
be employer provided health insurance, and we control for it in our analysis.
Our results suggest that for White women, there is a systematic relation between BMI
and wage differential (paid-employment wage – self-employment wage). This wage differential
is small but negative when BMI is in the healthy region. The absolute value of this wage
differential increase with BMI, and the absolute differential is maximum for White women with
BMI in their low-30’s. The wage differential starts to narrow as the BMI increases further and is
positive for women with BMI 40 or more. This is true for both hourly wages and hourly
compensation, and holds even after controlling for full set of control including whether they have
employer provided health insurance. Since this wage differential cannot be explained by either
productivity based explanation or even customer discrimination, we conclude that this suggests
weight discrimination by employers. White women who are overweight (25<=BMI<30) or
category-I obese (30<=BMI<35) seems to face most discrimination. On the other hand, women
are category-III obese (BMI>40) do not face discrimination. If anything, employers may be
subsidizing the wages of category-III obese White women. This may happen if employers do not
want to pay women with very high BMI a significantly lower amount compared to their normal-
weight counterpart, even when the former group has relatively low productivity, possibly
because employers do not want to appear discriminatory. This may also suggest that employers
believe severely obese individuals may have a better chance of winning if they sue an employer
in a court of law, especially in the post-2008 era when severe obesity became a “protected”
category under the ADA. Our results suggests that the ADAAA may have helped category-III
obese women, possibly at the expense of overweight and category-I obese women. If true, this
would be an example of unintended consequence of the ADAAA. However, our results on this
issue are not conclusive and it may be an important future research topic. We do not find any
evidence of employer discrimination against Black or Hispanic women based on weight.
However, we do find some evidence of weight discrimination against Black and Hispanic men,
which are similar to White women. In white men, the under-weight individuals face most
discrimination.
These results have significant policy implication. While the details of a potential
Weight Discrimination in Employment Act (WDEA) have not yet been proposed, a critical
question is whether there should be a minimum cut-off weight (or BMI) to be covered under
WDEA. If so, what would that be? The ADAAA sought to provide protection for individuals
with BMI of 40 or above. The Age Discrimination in Employment Act (ADEA) of 1967 treats
workers above the age6 of 40 as a protected category. Our results suggest that in order for the law
to be effective in mitigating weight discrimination, it must cover not only obese individuals, but
also overweight, and possibly healthy weight individuals.
6 https://www.eeoc.gov/facts/age.html
REFERENCES
Acemoglu, D., & Autor, D. (2011). Skills, tasks and technologies: Implications for employment
and earnings. Handbook of labor economics, 4, 1043-1171.
Amuedo-Dorantes, C., & Kimmel, J. (2009). Moonlighting over the business cycle. Economic
Inquiry, 47(4), 754-765.
David Autor and David Dorn. (2013) "The Growth of Low Skill Service Jobs and the Polarization
of the U.S. Labor Market." American Economic Review, 103(5), 1553-1597
Atella, V., Pace, N., & Vuri, D. (2008). Are employers discriminating with respect to weight?
European evidence using quantile regression. Economics and Human Biology, 6(3), 305-329.
Averett, S. & Korenman, S. (1996). The economic reality of the beauty myth. Journal of Human
Resources, 31(2), 304-330.
Baum, C. & Ford, W. (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 body weight. Working
Paper, Department of Economics, University of Pennsylvania, Philadelphia, PA.
Bell, David N F, Robert A Hart and Robert E Wright (1997), ‘Multiple job-holding as a ‘hedge’
against unemployment’, CEPR discussion paper series No. 1626. Centre for Economic Policy
Research, London, UK.
Bhattacharya, J. & Bundorf, M. (2009). The incidence of the healthcare costs of obesity. Journal
of Health Economics, 28(3), 649-658.
Cawley, J. (2004). The impact of obesity on wages. The Journal of Human Resources, 39(2), 451-
474.
Cawley, J. (2015) "An economy of scales: A selective review of obesity's economic causes,
consequences, and solutions." Journal of health economics 43, 244-268.
Cawley, J., & Meyerhoefer, C. (2012). The medical care costs of obesity: An instrumental
variables approach. Journal of Health Economics, 31(1), 219-230.
Conway, Karen Smith, and Jean Kimmel. "Male labor supply estimates and the decision to
moonlight." Labour Economics 5.2 (1998): 135-166.
Gregory, C., & Ruhm, C. (2011). Where does the wage penalty bite? NBER Working Paper No.
14984, in Michael Grossman and Naci H. Mocan (eds), Economic Aspects of Obesity, Chicago, Il:
University of Chicago Press, 315-347.
Han, E., Norton, C., & Powell, L. (2011). Direct and indirect effects of body weight on adult
wages. Economics and Human Biology, 9(4), 381-392.
Han, E., Norton, C., & Stearns, S. (2009). Weight and wages: Fat versus lean paychecks. Health
Economics, 18(5), 535-548.
Johar, M. & Katayama, H. (2012). Quantile regression analysis of body mass and wages. Health
Economics, 21(5), 597-611.
Kline, B. & Tobias, J. (2008). The wages of BMI: Bayesian analysis of a skewed treatment-
response model with non-parametric endogeneity. Journal of Applied Econometrics, 23(6), 767-
793.
Larose, S. L., Kpelitse, K. A., Campbell, M. K., Zaric, G. S., & Sarma, S. (2016). Does obesity
influence labour market outcomes among working-age adults? Evidence from Canadian
longitudinal data. Economics & Human Biology, 20, 26-41.
Ogden, C., Carroll, M., Kit, B., & Flegal, K. (2014). Prevalence of Childhood and Adult Obesity
in the United States, 2011-2012. Journal of American Medicine, 311(8), 806-814.
Pagan, J. & Davila, A. (1997). Obesity, occupational attainment, and earnings. Social Science
Quarterly 8(3): 756-770.
Panos, Georgios A., Kostas Pouliakas, and Alexandros Zangelidis. (2009) "The inter-related
dynamics of dual job holding, human capital and occupational choice.
Panos, Georgios A., Konstantinos Pouliakas, and Alexandros Zangelidis. "Multiple job holding,
skill diversification, and mobility." Industrial Relations: A Journal of Economy and Society 53.2
(2014): 223-272.
Paxson, Christina H., and Nachum Sicherman. "The dynamics of dual job holding and job
mobility." Journal of labor economics 14.3 (1996): 357-393.
Pomeranz JL. (2008) A historical analysis of public health, the law, and stigmatized social
groups: the need for both obesity and weight bias legislation. Obesity 16(S2):S93-103.
Pomeranz JL, Puhl RM. (2013) New developments in the law for obesity discrimination
protection. Obesity 21(3):469-71.
Puhl RM, Heuer CA. (2009) The stigma of obesity: a review and update. Obesity 17(5):941-64.
Puhl RM, Andreyeva T, Brownell KD (2008) . Perceptions of weight discrimination: prevalence
and comparison to race and gender discrimination in America. International journal of obesity
32(6):992-1000.
Puhl RM, Brownell KD. (2006) Confronting and coping with weight stigma: an investigation of
overweight and obese adults. Obesity 14(10):1802-15.
Puhl RM, Heuer CA. (2011) Public opinion about laws to prohibit weight discrimination in the
United States. Obesity. 19(1):74-82.
Register, C. & Williams, D. (1980). Wage effects of obesity among young workers. Social Science
Quarterly, 71(1):130-141.
Robinson, P. M. (1988). Root-N-consistent semiparametric regression. Econometrica: Journal of
the Econometric Society, 931-954.
Roehling MV, Pilcher S, Oswald F, Bruce T. The effects of weight bias on job-related outcomes:
a meta-analysis of experimental studies. In Academy of Management Annual Meeting 2008.
Sabia, J. & Rees, D. (2012) Body weight and wages: Evidence from Add Health. Economics
and Human Biology, 10(1), 14-19.
Sturm R, Hattori A (2013). Morbid obesity rates continue to rise rapidly in the US. International
Journal of Obesity 37(6): 889.
Verardi V, Debarsy N (2012). Robinson's square root of N consistent semiparametric regression
estimator in Stata. Stata Journal 12(4): 726-735.