labor market consequences of race differences in health

Transcription

John Bound, Timothy Waidmann, Michael Schoenbaum, and
Jeffrey Bingenheimer
The Labor Market Consequences of Race Differences
in Health
PSC Research Report
Report No. 03-535
May 2003
PSC
P OPULATION S TUDIES C ENTER
A T THE I N S T I T U T E FOR
S OCIAL R ESEARCH
U NIVERSITY
OF
M ICHIGAN
The Population Studies Center (PSC) at the University of Michigan is one of the oldest population centers
in the United States. Established in 1961 with a grant from the Ford Foundation, the Center has a rich history
as the main workplace for an interdisciplinary community of scholars in the field of population studies.
Currently the Center is supported by a Population Research Infrastructure Program Grant (R24) from the
National Institute of Child Health and Human Development, and by a Demography of Aging Center Grant
(P30) from the National Institute on Aging, as well as by the University of Michigan, the Fogarty International
Center, the William and Flora Hewlett Foundation, and the Andrew W. Mellon Foundation.
PSC Research Reports are prepublication working papers that report on current demographic research
conducted by PSC-affiliated researchers. These papers are written for timely dissemination and are often
later submitted for publication in scholarly journals. The PSC Research Report Series was begun in 1981.
Copyrights for all Reports are held by the authors. Readers may quote from this work as long as they
properly acknowledge the authors and the Series and do not alter the original work.
PSC Publications
http://www.psc.isr.umich.edu/pubs/
Population Studies Center, University of Michigan
PO Box 1248, Ann Arbor, MI 48106-1248 USA
A revised final version of this paper will appear in
the September, 2003 issue of Milbank Quarterly.
The Labor Market Consequences of Race Differences in Health
John Bound
Professor of Economics University of Michigan and
Senior Research Scientist, Population Studies Center, University of Michigan
Timothy Waidmann
Senior Research Associate, Urban Institute and
Research Affiliate, Population Studies Center, University of Michigan
Michael Schoenbaum
Economist, RAND, and
Research Affiliate, Population Studies Center, University of Michigan
Jeffrey B. Bingenheimer
Doctoral Student, Department of Health Behavior and Health Economics,
University of Michigan
Acknowledgements: Financial support was provided through a contract from the National
Institutes of Health, and by the National Institute on Aging (R01 AG17579-01). We are grateful
to Seema Thomas for able research assistance; and to Raynard Kington, Arline Geronimus,
participants in the National Institute of Health Conference on Health Disparities, Washington,
D.C., 2001 and three anonymous referees for helpful comments on earlier drafts.
ABSTRACT
This paper examines the extent to which race and ethnicity disparities in health can account
for race and ethnicity disparities in employment status and other labor-related outcomes. Using
White Americans as the reference case, we identify population groups whose health is
systematically worse than that of Whites; in practice, this means Blacks and Native Americans.
We then document the distribution of labor-related outcomes – employment status, earnings,
participation in public transfer programs, and household income – for these three groups.
Finally, we examine the extent to which differences across groups in health status can account
for differences in labor-related outcomes across those groups. Data come from the 1990 US
Census. Our findings suggest that health disparities contribute to the substantial difference in
employment and in participation in public transfer programs observed between Whites and
Blacks and between Whites and Native Americans. At the same time, health disparities account
for relatively little of the (also substantial) differences in household income, or in earnings
among the employed, observed across race/ethnic groups.
Data Sets Used:
U.S. Census of Population and Housing, 1990: Public Used Data Sample: 5% Sample.
Page 1
INTRODUCTION
Differences in health status across different race and ethnic groups in the United States,
particularly between Black and White Americans, have been the subject of considerable medical
and social science research. For instance, numerous studies using a variety of health measures
have shown the health of Black men and women to be worse than that of Whites (e.g. Manton et
al., 1987). The health disadvantage, relative to Whites, of Native Americans has also been
documented extensively. Patterns for other major ethnic groups (e.g., Hispanics and Asians)
have been somewhat more variable, depending on the measure, the age of the study sample, and
other factors such as place of birth and acculturation (Hayward and Heron, 1999; Shalala et al.,
1999).
Systematic health disparities are likely to have profound – and self-reinforcing –
consequences for the relative well-being of different population groups. In this paper, we focus
one dimension of such consequences: the association between race differences in health status
and race differences in labor market outcomes, respectively. In particular, the same groups for
which health disparities relative to Whites are consistently large also have significantly lower
rates of employment, earnings, and individual and household income than Whites. It seems
likely that these patterns are related. Poor health reduces individuals’ labor supply and reduces
the productivity of those who work. In addition, poor health increases individuals’ demand for
medical care but reduces the probability that individuals have access to private health insurance
(particularly since private insurance is most affordable when offered as an employer-sponsored
benefit). At the same time, social policies such as the progressive income tax structure, income
transfer programs, and public health insurance are intended to offset some of these forces.
In prior work, we have documented the extent to which the poor health of Blacks, relative to
Whites, can account for the generally lower employment rates of Blacks nearing retirement age
(Bound et al., 1995, 1996). In this study, we extend those analyses to examine other outcomes
associated with labor market behavior, including earnings, household income, and participation
in public income transfer programs; a broader age range; and additional race/ethnicity groups.
Using White Americans as the reference case, we limit our analysis to the two racial/ethnic
groups – Blacks and Native Americans – whose health is systematically worse than Whites. We
then document the distribution of labor-related outcomes for these respective groups. Finally,
we examine the extent to which differences across groups in health status can account for
outcome differences across those groups.
DATA AND MEASURES
We sought microdata that contained information on respondents’ demographic and health
status, as well as their labor force behavior and labor-related outcomes such as earnings and
income. Given our focus on race/ethnicity disparities, we needed data with sufficiently large
samples to stratify analyses by race and ethnicity. Given the strong age gradients on health, we
also needed to be able to stratify by age. In our view, the only data source meeting all these
criteria was the US Census. Data in the current study thus come from the five percent public use
microdata sample (PUMS) of 1990 US Census, a nationally representative sample of the US
population. The PUMS sample is drawn from the random subset of US Census respondents who
completed the “long form” Census questionnaire in 1990.
A revised final version of this paper will appear in the September, 2003 issue of Milbank Quarterly.
Page 2
The PUMS contains data on the key domains of interest here, including respondents’ health
and labor force status and individual and household income from various sources. However, as
we discuss in detail below, the available measures for some domains, particularly health, are
fairly limited. In addition, the PUMS lacks certain kinds of information, such as data on health
insurance coverage, which would have contributed to this analysis.
There are a variety of other data sources with much more detailed information on health,
health insurance and medical expenditures. For example, the National Health Interview Surveys
(NHIS), the Medical Expenditure Panel Survey (MEPS), its predecessor the National Medical
Expenditure Survey (NMES), and the Health and Retirement Study (HRS) all contain detailed
measures of health status as well as employment and income data and include over-samples of
racial and ethnic minorities. While the HRS covers too narrow an age range, we did conduct
extensive analyses using data from MEPS. At best, however, the much smaller MEPS sample
permitted analyses only of Whites and Blacks, and it did not support stratification by age. For
the broad age group analyzed in the MEPS, findings were broadly consistent with those
presented here, but precision was substantially lower.
Labor-related Outcomes
No single definitive outcome captures all the possible labor market effects of health
disparities. We therefore examine a range of related outcomes.
Employment. – We calculate the effect of health disparities on the proportion of each
population that is employed. For the purposes of our analyses, we define employment as doing
any work in the week prior to April 1, 1990, the index date of the Census questionnaire. This
definition includes self-employed persons and members of the armed forces on active duty.
Average Weekly Earnings. – Respondents reported their earnings for 1989, and the number
of weeks worked during 1989. Analyses of weekly earnings are restricted to respondents who
had positive earnings in 1989. We calculate annual earnings as the sum of income from wages,
salary, commissions, tips and bonuses; and self-employment income from farm and non-farm
businesses, and then divide by weeks worked. For persons employed by others, we add to
earnings the taxes employers pay into the Social Security system (OASDI and Medicare),
calculated using the applicable formulae for 1989 earnings (SSA, 2000). We view this as
deferred compensation. For similar reasons, we would have liked to assess total compensation,
including the employer cost of worker's compensation and unemployment insurance, and health
insurance and retirement benefits; however, data on these elements of compensation are not
available in the Census and imputation based on industry and occupation of employment would
have been unreliable. Tabulations using the Health and Retirement Survey (Barsky et al., 2002a)
and MEPS, which do contain such information, suggest that ignoring these forms of
compensation will lead us to underestimate the effect of health on labor market outcomes by a
small amount.
Participation in Public Transfer Programs. – Respondents reported the amount of income
they received in 1989 from Railroad Retirement or from Social Security, and we examined
whether respondents received any such income. Since we focus primarily on people under age
65 in this analysis, most Social Security participation will be due to Social Security Disability
Insurance. In parallel analyses, we examined whether respondents received any income in 1989
from public assistance, including Aid to Families with Dependent Children, Supplemental
Page 3
Security Income, and other public welfare sources. Results are qualitatively similar and are
reported in an appendix 3.
Household Income. – We calculate respondents’ total household income as the sum of the
household’s earned income and retirement benefits in 1989, excluding asset income (i.e.,
interest, royalties, rental income and income from estates or trusts); plus income from Social
Security and public assistance. We sum across members of the same dwelling unit who are
related by blood, adoption, marriage, foster care, or other family association.
We then subtract income, property and Social Security taxes. To calculate households’ tax
burden, we use the data on all persons in the household to create tax-related variables (e.g., state,
filing status, number of dependents, and age exemptions) as well as non-asset income
disaggregated by category (i.e., own wage; spouse wage, if married filing jointly; farm and
business income, alimony and other cash income; pension and IRA distributions; gross Social
Security income, non-taxable transfer income; and unemployment compensation.) as inputs to
TAXSIM. TAXSIM, a microsimulation model supported by the National Bureau of Economic
Research, takes these data and calculates federal and state taxes for each filing unit. Couples
who were married with spouses present were assumed to file jointly. Children and step-children
of the householder ages 18 and under and living in the household were considered dependents of
the householder. Individuals over 19 and subfamilies in the household were considered separate
tax-filing units.
Demographic Characteristics
We assigned each individual to one race/ethnicity category using responses to the racial
identity and Hispanic ethnicity questions in the census. We used non-Hispanic Whites
(“Whites”) as the reference group for our analyses. Of the race/ethnicity groups we could
examine in the PUMS, only two had systematically worse health than Whites, and we limited our
analyses to these groups: non-Hispanic Blacks (“Blacks”); and Native Americans and Alaska
Natives (“Native Americans”). In this analysis, “Native Americans” include all those
responding “Indian(Amer.)”, “Eskimo” or “Aleut” to the race question, regardless of their
response to the question on Hispanic origin (see Snipp, 1989 and Harris, 1994) for a discussion
of the identification of the Native American population using census data). One alternative data
source we investigated (NMES) contains a special supplementary sample of Native Americans
obtained by oversampling persons living on or near reservations. Nationwide, however, only
about half of persons who identify themselves as Native American live in such areas. (Snipp,
1989) We also examined the health status of Asians and of several different Hispanic groups,
including residents of Puerto Rico, people of Puerto-Rican descent in the United States, and
Mexican-Americans. These groups showed no consistent health disadvantage relative to Whites,
particularly after accounting for levels of educational attainment, and were thus excluded from
this analysis.
Given strong age gradients in health status, and in the relationship between health and labor
market outcomes, we stratify respondents into age categories for all analyses (except as noted).
Within age categories, we account for age using single year dummy variables.
Except as noted, we account for education in our simulation models, because of the strong
relationship between education and health for all race/ethnicity groups and because there are
substantial differences in the distribution of educational attainment across race/ethnicity groups.
We measure education using five mutually-exclusive dummy variables, indicating completion of
Page 4
8th grade or less, completion of some high school, high school graduation, some college, and
college graduation. For some analyses, we stratified respondents by education, grouping the first
two categories (“high school dropouts”), the next two categories (“high school graduates”) and
the last category (“college graduates”), respectively.
For analyses of household income, all analyses account for marital status, defined by a single
variable indicating whether the individual is married and living with his/her spouse as of April 1,
1990. We note that for models of household income, demographic (and health) characteristics
used for standardization are measured only for the individual, e.g., the analyses do not explicitly
account for race differences in the joint distribution of health among couples.
Health Status
The Census includes only four health measures, each regarding physical disability: whether
the person has a physical, mental or other health condition that has lasted for six or more months
and that (a) limits the kind or amount of work the person can do at a job; (b) whether such a
condition prevents the person from working at a job; (c) whether such a condition makes it
difficult for the person to go outside the home; and (d) whether such a condition makes it
difficult for the person to take care of his/her own personal needs.. The census questions
regarding work, mobility and personal care limitations represent three distinct questions. As a
result, any possible pattern of responses to these questions is possible. In practice, those that
report mobility or personal care limitations usually also report limitations on the amount or kind
of work they can do.
Unless otherwise noted, we account for three out of four of these measures in our analyses,
excluding the indicator of whether the respondent is unable to work out of concern that this
measure may be particularly endogenous to labor force outcomes (Waidmann et al., 1995). In
practice, however, simulations that controlled for all four Census health measures yielded
substantively similar results.
We recognize that there are legitimate concerns about the validity of the Census disability
items, which are based on a mailback questionnaire rather than an interview; are sometimes
filled out by proxy respondents; and contain only general questions on functional limitation,
which have been shown not to correspond precisely to more specific and detailed measures such
as Activities of Daily Living (ADLs) or Instrumental Activities of Daily Living (IADLs)
assessed in other surveys (Andresen et al., 2000). To date, however, little empirical attention has
been paid to the validity of the Census disability items.
One key question for this analysis is whether differences across the populations in responses
to Census questions accurately reflect differences across these populations in terms of the agespecific functional capacities of the respective populations. Available evidence suggests that this
is plausibly the case. For instance, patterns of functional limitations by age and race in our
Census sample are consistent with evidence on disparities from other sources (e.g., Hayward and
Heron, 1999; Geronimus et. al 2001), lending “face validity” to the data. In addition,
comparisons between Census and National Health Interview Survey (NHIS) data show similar
patterns where comparisons can be made (e.g., Waidmann et al., 1995; Geronimus et al., 2001);
the same is true for comparisons between the Census and the National Long Term Care Survey
(e.g., Waidmann et al., 1995; Crimmins et al., 1997). Census-type health questions have been
shown to be correlated with clinical measures of morbidity across racial and ethnic groups
(McGee et al, 1999). Finally, despite a slightly higher rate of non-response to one or more of the
Page 5
Census disability items by Blacks (10%) than by Whites (7%), our results were insensitive to
whether we included or omitted respondents with imputed health measures.
A second key issue is whether the association between the Census disability measures and
labor market outcomes accurately reflects the association between “true” health and labor market
outcomes. Global self-reported measures of health, although subjective, are highly correlated
with clinical measures of morbidity and predictive of subsequent death, health care utilization,
and labor market behavior (Idler and Benyamini, 1997; Nagi, 1969; Maddox and Douglas, 1973;
LaRue et al., 1979; Ferraro, 1980; Mossey and Shapiro, 1982; Manning et al., 1982). Moreover,
this appears to be true across racial and ethnic groups (McGee et al, 1999). Our results, reported
below, mirror these findings, in that we observe a very strong association between the Census
functional limitation measure and labor market outcomes, overall and for different race/ethnicity
groups.
However, even if the Census health measures are reliable indicators of actual health, this
does not necessarily mean that using such measures as proxies for health when examining racial
differences in labor market outcomes will yield valid results. At issue is whether the Census
measures are systematically biased. There are a number of reasons to be suspicious of the use of
the kind of survey measures available on the Census for studying labor market outcomes
(Parsons, 1982; Anderson and Burkhauser, 1984, 1985; Bound, 1991; Waidmann et al., 1995).
For instance, respondents are being asked for subjective judgments, and there is no reason to
expect that these judgments will be entirely comparable across respondents. Second, responses
may not be independent of the labor market outcomes of interest. Third, since health may
represent one of the few “legitimate” reasons for a working aged man to be out of work, men out
of the labor force may mention health limitations to rationalize their behavior. Each of these
problems will lead to a different kind of bias. The lack of comparability across individuals
represents measurement error that is likely to lead us to underestimate the impact of health on
labor market outcomes, while the endogeneity of self-reported health is likely to lead us to our
exaggerate its impact. Biases in our estimation of health's impact on outcomes will also induce
biases on coefficients of any variables correlated with health.
In previous work, we have taken advantage of more detailed information on health status
available in other datasets (e.g., the Health and Retirement Survey) to address the potential
biases that may affect estimates using the kind of measures available in the Census (Bound,
1991; Bound et al, 1996, 1998). Overall, this work suggests that findings regarding the effects of
health on labor force outcomes have been consistent across different datasets, health measures,
and estimation methods. Indeed, if anything, the results from this previous work suggest that the
“naïve” use of Census-type health measures (or self-rated general health) tends to understate the
effects of health on labor market outcomes. These issues are discussed more formally in
appendix 2 and at greater length in Bound (1991), Bound et al. (1996).
METHODS
We exclude respondents with missing or imputed data on age, gender, race/ethnicity,
education, health status, and the respective labor-related outcomes examined here. All our
analyses were conducted using the person-level weights developed to make the PUMS sample
nationally representative. Our analysis sample includes 5,487,713 White, 554,471 Black, and
51,058 Native American respondents aged 25 and over. Because of possible age gradients in the
relationships being examined here, we stratified respondents into ten-year age groups. All
Page 6
analyses were conducted within these age strata.
We begin by calculating descriptive statistics on the levels of poor health, educational
attainment, and income for different race/ethnicity groups, by age and gender. We then use
simple multivariate regressions that estimate the association between poor health and labor force
outcomes and income, for different race/ethnicity groups. Finally, we use data on Whites,
Blacks and Native Americans to simulate what the distribution of outcomes would be among
Blacks/Native Americans if the health status of Blacks/Native Americans were to rise to the
levels actually observed among Whites.
To implement these simulations, we reweight the data on Blacks/Native Americans so that
the distribution of health characteristics in the reweighted sample matches the empirically
observed distribution of health characteristics among Whites with the same age and gender. We
illustrate this procedure using a somewhat stylized example. Among 45-54 year old men in the
1990 Census, approximately 88% of Whites and 75% of Blacks are employed, while
approximately 10% of Whites and 17% of Blacks identify themselves as having a health problem
that limits the kind or amount of work they can do. For both Whites and Blacks, employment
rates are much higher for men without a health limitation than among men who are limited
(Figure 1). The overall rate for Whites represents the weighted average of the rates for the two
white populations:
EW
EWU SWU EWL SWL
(1)
where EW represents the overall non-employment rate for Whites; EWU and EWL represent the
employment rates for the unlimited and limited populations, respectively; and SWU and SWL
represent the share of the White population without and with limitations. The overall
employment rate for Blacks, E B , can be expressed in analogous fashion:
EB
E BU S BU E BL S BL
(2)
Here re-weighting the Black data involves reweighting the Black health-specific employment
rates ( E BU and E BL ) using the White population shares ( SWU and SWL ):
Eˆ B
E BU SWU E BL SWL
(3)
In our example, when Blacks are reweighted using the White prevalence of health limitations,
their employment rate rises from approximately 75% to 81% (Figure 2). This method of “direct
standardization” is a standard tool used by demographers (Kitagawa, 1964).
Researchers typically interpret Equation (3) in terms of a counterfactual, i.e., “what would
happen to Black employment rates if the distribution of health among Blacks improved to match
that of Whites.” However, this interpretation supposes that, when the health status of Blacks
improves, employment rates conditional on health status would remain the same. This is
unlikely to be the case, since policy interventions that improve the health of Blacks are likely to
affect the employment rates among Blacks with and without health limitations. Alternatively,
one can interpret Equation (3) as a descriptive statement. Subtracting (3) from (1) yields:
Page 7
EW Eˆ B
[ EWU E BU ]SWU [ EWL E BL ]SWL
(4)
The interpretation of (4) is easiest in the case where racial employment differentials for are
similar for those that are and are not limited (i.e., [ EWU EBU ] | [ EWL EBL ] ). In this case, the gap
in employment between Whites and re-weighted Blacks is just the employment gap conditional
on limitation status. More generally, the employment gap can be interpreted as the weighted
average of disability-specific gaps in employment, where the weights (S) are based on the White
health distribution. If reweighting makes little difference, and the expression in equation (4)
remains very close to the raw difference in employment rates, this suggests that health disparities
do not make an important contribution to the racial gap in employment rates. On the other hand,
if we find that reweighting makes a significant difference, as we do for some age groups, this
suggests that health disparities do contribute to the gap in employment rates .Thus, while
reweighted employment rates (as in 3) may not be an accurate estimate of the counterfactual,
comparisons between actual and reweighted employment rates do give us a sense regarding the
relative importance of health disparities in explaining employment rate disparities.
In this example, we standardized Blacks to have the health characteristics of Whites, an
approach that is conceptually consistent with the societal goal of improving outcomes among
disadvantaged groups. However, it is obviously also possible to standardize Whites to the Black
distribution of health. In the example at hand, these alternative simulations produce
substantively similar results, because the effect of health limitations on employment is similar
for Blacks and Whites. In general, however, the two alternative simulations need not account for
similar fractions of the overall Black/White gap in the outcome of interest (cf. Barsky et al.,
2002b).
We note an important issue that guides our empirical analyses. Our focus is on the
association between health status and labor-related outcomes. However, Blacks and Native
Americans are also disadvantaged relative to Whites with respect to other characteristics,
particularly educational attainment, that are themselves associated with both health status and
labor-related outcomes. Standardizing the health characteristics of Blacks to White levels will
also, implicitly, raise educational attainment in the standardized population. Since the
probability of employment increases sharply with education, our simple example confounds the
labor force effects of Black/White disparities in health with the effects of such disparities in
education. Standardizing on educational attainment alone would raise analogous issues.
In general, there is no way to identify the separate effects of health disparities and education
disparities on labor-related outcomes using our simulation methods. We therefore conduct
multiple simulations, controlling for age; age and health; age and education; and age, education
and health. Comparing simulations that standardize on age alone with ones that standardize on
age and health will tend to overestimate the effect of health per se on employment differentials,
since the effects of “health” includes the effects of the part of education that is correlated with
health. On the other hand, comparing simulations that standardize on age and education to ones
that standardize on age, education and health will tend to underestimate the impact of health
disparities, since it ignores the impact of health disparities that are correlated with educational
disparities. The true effect of disparities in health per se is likely to lie somewhere between
these bounds. For our example (Figure 2), Black/White disparities in health thus account for
somewhere between 21% and 39% of the Black/White gap in employment.
Page 8
The method just described is appropriate for standardizing on a limited number of categorical
variables (e.g. age, or the dichotomous measure of health limitation used in Figures 1 and 2). In
practice, we are interested in more complex reweighting. We therefore use multivariate
standardization techniques that are conceptually analogous to direct standardization but can
incorporate many, and continuous, stratifying variables (DiNardo et al., 1996; Heckman et al.,
1998; Barsky et al., 2002b).
To illustrate, we can extend our simple example to the case in which we want to simulate the
effect on employment of reweighting Blacks to reflect the distribution of health and other
characteristics observed among Whites. To do this, we first restrict the sample to Black and
White respondents. We then estimate a logistic regression in which the dependent variable is 1
if the respondent was Black and 0 otherwise. Explanatory variables include health and other
characteristics, such as age and education, depending on the simulation of interest. Also
included were interactions between health and education. The parameter estimates can be used to
calculate the predicted probability that a person with a particular set of explanatory variables
would be observed in the Black sample.
Based on these predicted probabilities, or propensities, Blacks who have characteristics that
are more typical of Whites are weighted up, while those whose characteristics are more
representative of the overall Black sample are weighted down. The reweighted (or
counterfactual) distribution from group a can then be used to calculate any characteristic of the
counterfactual outcome distribution (mean, median, percentiles) or to depict the entire
distribution of outcomes (kernel density estimation). Simulation methods are described in
further detail in appendix 2 and in Barsky et al. (2002b).
The results of these standardizations identify the extent to which differences in health
observed between Blacks and Whites can account for differences between these groups in laborrelated outcomes. We emphasize that we do not necessarily interpret this accounting as causal,
in the sense that ameliorating the health differences would necessarily yield proportional
reductions in the Black/White disparities in labor-related outcomes. Furthermore, to the extent
that observed health differences do not fully account for such disparities, we do not interpret the
residual differences in outcomes as causally related to being Black in any meaningful way. At
best, the remaining difference may be interpreted as an estimate of the importance of factors
other than health that effect race differences in earnings. At minimum, it as a measure of the
average gap in each outcome between Blacks and Whites, conditional on health.
FINDINGS
For reasons of space, we present results for two age groups: 25-34 and 55-64. Except as
noted, the patterns we report vary monotonically with age. Detail on other age groups is
provided in appendix 3.
Page 9
Descriptive Statistics
Figures 3 (for men) and 4 (for women) plot the prevalence of various types of health
limitation by age and race/ethnicity. For each of the four measures, prevalence rates are low for
men and women in their twenties and thirties but begin to rise after that. For each measure, the
prevalence is higher for Blacks and Native Americans than for Whites, and prevalence rises with
age more steeply than for Whites. As a result prevalence rates tend to diverge with age, so that
the absolute difference between Whites and Blacks/Native Americans increases.
Table 1 reports the prevalence of poor health by race/ethnicity, overall and by level of
educational attainment. The results show that the prevalence of each indicator of poor health is
higher for Blacks and Native Americans than for Whites, overall and within each education
category. Rates for Blacks and Native Americans are more similar, with prevalence being higher
for Blacks in some cells and for Native Americans in others. The overall prevalence of poor
health rises with age. In general, the size of the absolute disparity between Whites and the other
two groups rises somewhat with age, but the percentage disparity does not vary consistently by
age.
While the overall prevalence of poor health is inversely related to educational attainment,
there do not appear to be consistent gradients in health disparities across education categories.
Table 1 also includes a row in each panel presenting the prevalence of each health measure,
accounting for race/ethnicity disparities in the distribution of education. Across all ages,
differences in education account for between 52% and 84% of the race/ethnicity disparities in the
Census health measures. This finding is consistent with prior research (e.g., Schoenbaum and
Waidmann, 1997).
Poor Health and Labor-related Outcomes
Tables 2 and 3 present the effects of multivariate regression analysis that examine the
association between health status and employment, earnings, and participation in public
assistance programs and Social Security, respectively. Poor health is captured by a dummy
variable indicating whether respondents’ health limits the amount or type of work they can
perform. These models also include age and education as explanatory variables. Models are
stratified by race/ethnicity, age, and gender. The first row reports the result of models for the
overall sample in each race/ethnicity-, age-, and gender-specific cell, while the remainder report
results for models that are stratified for educational category. Cell values represent the
parameter estimate for poor health from the respective regressions. All regressions control for
age and education.
The first panel of Table 2 presents the association between poor health and employment
status. We used logistic regression to model this outcome, so the cell entries contain logit
parameters. These can be interpreted as the difference between people with and without health
limitation in the log odds of employment. Alternatively, the coefficients can be exponentiated to
give the odds of employment for people with health limitation relative to people without.
The parameter estimates are negative and very large in every gender, age, education, and
race/ethnicity cell, confirming that the association between self-reported work limitations and
employment is very strong across all these population groups. Thus, for instance, the coefficient
of -2.32 for Black 25-34 year old men implies that for this group, controlling for age and
education, the log odds of being employed is 2.32 lower for those reporting a limitation than for
Page 10
those not reporting a limitation – i.e. the odds of being employed for those reporting a limitation
is roughly 1/10 the odds of being employed for someone not reporting a limitation ( e-2.32|-1/10).
Parameter estimates are generally similar for younger White and Black men. The effect of
work limitations on employment is somewhat stronger for Black men at older working ages.
This pattern is consistent with expectations, since limitations among Black men are likely to be
more severe than those among Whites, and since Black men tend to be in more physically
demanding jobs. For Native American men, the effect of work limitations on employment is
generally weaker than for Whites. For women, parameter estimates for White and Native
Americans are generally similar, and those for Blacks are consistently larger (in absolute value).
The second panel of Table 2 presents the association between poor health and the natural log
of weekly earnings, among respondents who reported positive earnings in 1989. Regressions are
weighted by weeks worked, so that people working more weeks will receive more weight and
the samples should be representative of the working population in a typical week. The cell
entries contain parameter estimates for the effect of work limitations from linear regression
models; these coefficients can be interpreted loosely as the percent reduction in weekly earnings
associated with a work limitation. With one exception, the effect of poor health is negative and
large – reducing earnings by 20% to 40% – in every gender, age, education and race/ethnicity
cell. Effect sizes are generally similar across ethnicity groups, and there are no apparent patterns
in the differences that do exist.
Table 3 presents logit parameter estimates for the effect of work limitations on participation
in public assistance and Social Security, respectively. With the exception of the top age category
in the table, individuals in this age range participate in Social Security through the Disability
Insurance Program or if they have end stage renal disease. For both outcomes, parameter
estimates are positive and very large for every population group listed. Parameter estimates are
somewhat to substantially higher for college graduates than for those with less education, in
most age, gender and ethnicity categories. With a few exceptions, the effect of poor health on
participation in public assistance is larger for Whites than for Blacks, and for Blacks than for
Native Americans; most exceptions are among college graduates, perhaps because the overall
prevalence of public assistance is lowest for this group. In general, the same ethnicity gradients
hold for women with respect to receipt of Social Security; there are no consistent ethnicity
patterns for men.
Simulations
In the remainder of this section, we report the results of simulations in which the Black and
Native American samples are reweighted to reflect the distribution of age, education, and health
status observed among Whites in our Census sample. Table 4-6 present simulation results,
stratified by gender and age. In each table, the first row for each age category lists the actual
level of the outcome of interest among Whites. The remaining rows present the simulated level
of the outcome for Blacks and Native Americans, respectively, after accounting for
race/ethnicity differences in age; age and education; age and health; and age, education and
health. We note that, empirically, race/ethnicity differences in age alone account for little to
none of the disparity in any outcome; as a result, the level in the “age” row is very close to the
actual (unadjusted) level observed among Blacks and Native Americans, respectively.
The first panel of Table 4 shows simulations for employment. Comparing the first two rows
in each age category shows that employment rates are substantially lower for Black and Native
Page 11
American men and women than for Whites (except for Black women aged 55-64). Comparing
the third and fourth rows with the first indicates that education and health disparities can account
for a substantial fraction, but not all, of the race/ethnicity disparities in employment (except for
Black women). For example, 90% of White men aged 25-34 are employed, compared with an
(age-adjusted) rate of 73% of Black men in that age group. If Blacks in this group had the same
distribution of education and health as Whites, we estimate that their employment rate would
increase to 78%, thus narrowing the original differential by nearly one-third.
The fraction of race/ethnicity differences in employment that are accounted for by health
disparities increases substantially with age. For instance, following the logic described in the
Methods section, health disparities account for at least 3% (lower bound, comparing the age row
with the age and health row) and at most 9% (upper bound, comparing the age and education
row with the age, education and health row) of the employment gap between Black and White
men aged 25-34 – but between 17% and 41% of the gap between Black and White men aged 5564. This pattern is similar for Native American men and for Black and Native American women.
Interestingly, health disparities account for a larger fraction of the race/ethnicity gap in
employment for women than for men. We note that older black women have essentially
equivalent employment rates as White women; accounting for health disparities, the employment
rate of Black women would be substantially higher than for their White counterparts.
The second panel of Table 4 presents simulations for weekly earnings, among those with
positive earnings in 1989. Mean earnings for Black and Native American men and women in
each age group are substantially lower than for Whites. As in the first panel, adjusting for
race/ethnicity differences in education accounts for a substantial fraction of these disparities.
For instance, for men, education differences account for between one-third and half of the
earnings gap between employed Black and Native American men and Whites, respectively. We
estimate that if employed Black women in most age categories had the educational distribution
of Whites, they would have higher mean earnings.
Notably, although health per se is strongly associated with earnings (cf. Table 2),
race/ethnicity disparities in health account for little to none of the disparities in earnings, for any
of the age, gender and race/ethnicity categories. The large effect of health on employment
effectively eliminates measured health differences among the employed. Findings regarding
median earnings are substantively similar (see appendix 3).
Table 5 presents results for participation in Social Security. Participation is somewhat lower
among Whites than among Blacks or Native Americans, for each age and gender group.
However, differences nearly disappear in the older age category in the table. For all age groups,
health differences account for a very large fraction of race/ethnicity disparities in Social Security
participation, and often more than are accounted for by education disparities. This is presumably
because of the structural relationship between poor health and eligibility for Social Security
Disability Insurance. For men, health and education differences account for all, or nearly all, of
the race/ethnicity differences in Social Security participation (although the base rates are
relatively low).
Finally, Table 6 presents simulation results for mean household income (net of taxes and
public transfers). Not surprisingly, mean household income is substantially higher for Whites
than for Blacks and Native Americans. The results suggest that health disparities account for
relatively little of the substantial race/ethnicity gap in household income, for men or women.
However, through age 65 the explanatory power of health disparities increases consistently with
age, presumably reflecting the age gradient in the relationship between health disparities and
Page 12
employment, observed in Table 4 (the explanatory power of health disparities declines slightly
with age after age 65). Detailed results are included in appendix 3.. Health disparities account for
a maximum of 15% of the race/ethnicity gap in mean income (for Native American vs. White
men aged 55-64).
In contrast, disparities in education account for a much larger fraction of the race/ethnicity
gap in mean household income. Findings regarding median income are substantively similar
(these are available by request). There are number of reasons for this result. For one thing,
education disparities are strongly associated with disparities in earnings among the employed, in
addition to being associated with disparities in employment; in contrast, health disparities are
only associated with the latter. In addition, household earnings include spouse’s earnings
(among married couples), which help reduce the economic effects of the poor health of one
partner. In contrast, given very strong patterns of assortative mating, the education of spouses is
strongly correlated, reinforcing the association between respondents’ own education and
household income.
DISCUSSION
Using micro data from the 1990 Census, we have documented strong associations between
health and labor market outcomes for Blacks, Native Americans and Whites. For each of these
groups, people who reported having a health problem that limits work are dramatically less
likely to be working than similarly educated individuals without such limitations. In addition,
among those who were working, people with health limitations typically earned between 20 and
40 percent less than people without such limitations.
Overall, we found that Blacks and Native Americans, respectively, have worse labor market
outcomes, and worse health and lower educational attainment, than their White counterparts.
Race/ethnicity disparities in health could account for a significant part of the differences in
employment rates between Blacks and Native Americans on the one hand and Whites on the
other, particularly among men and women 45-64 years old. In contrast, observed race/ethnicity
disparities in health among employed people were relatively small, leading to our finding that
health disparities appeared to contribute very little to weekly earnings disparities between
race/ethnicity groups. To examine whether this finding was an artifact of using the Census
health measures, we conducted analogous simulations using the Health and Retirement Survey
and the more detailed health measures available there; results were substantively similar.
We also found that health disparities could account for a significant part of the higher
participation rates among Blacks and Native Americans in public assistance programs and
especially Social Security, relative to Whites. These results combined with those for
employment to produce some association between race/ethnicity disparities in health and
disparities in household income, particularly among people aged 45-64. This age gradient, and
that for employment, suggests that health differences – possibly due to accumulated economic
disadvantage over the life-course – reinforce the economic disadvantage of Blacks and Native
Americans relative to Whites that is evident at all ages. Overall, however, health disparities
accounted for relatively little of the race/ethnicity gaps in income, which appeared to be
associated much more strongly with disparities in education (this was true despite the strong
association between health and earnings).
One could think of our analyses as addressing the following counterfactual: “Suppose health
disparities between Blacks and Native Americans on the one hand and Whites on there other
Page 13
were eliminated. What effect would this have on the employment and earnings of Blacks and
Native Americans?” However, the reciprocal relationship between health status and economic
status calls into some question the meaningfulness of the counterfactual that motivates our
analysis. For instance, to the extent to which health disparities are a consequence of
race/ethnicity differences in access to health care, it is possible to imagine eliminating such
disparities without first eliminating economic disparities between the races. Alternatively, to the
extent that health disparities are initially due to economic disparities, it is impossible to imagine
eliminating health disparities without first eliminating economic disparities directly. It is for this
reason that we have described our work as primarily an accounting exercise.
As we have discussed, the health measures available in the Census are limited in a number of
ways. These limitations could lead us to over- or under-estimate the effect of health on labor
market outcomes (Bound, 1991). They could also induce a bias on estimates of the impact of
race/ethnicity on outcomes, regardless of whether we correctly measure the impact of health
itself. However, prior research suggests that the various potential sources of bias apparently
balance each other out (Bound, 1991; Bound et al, 1996, 1998,).
Our indicators of the labor market “costs” of health disparities are also limited. For instance,
omitting factors such as health insurance and private pensions is likely to lead us to understate
race/ethnicity disparities in labor-related outcomes. With the caveat that health care use is
endogenous with respect to health insurance coverage and economic status, we would also have
liked to account for out-of-pocket spending on medical care, since one reason that poor health
reduces economic well-being is because sick people purchase medical care.
Finally, we note that health disparities between Blacks and Native Americans and Whites,
respectively, are likely to have substantial impacts on well-being not captured in this analysis.
The measures of economic disparity examined here do not incorporate the costs of premature
mortality, but it is clear that any comprehensive conceptualization of economic disadvantage
could, and should, account for this (Murphy and Topel, 2001). Nor do they capture the value of
health for time spent outside of market work ( Hirth et al., 2000) or the impact of anticipated
poor health or mortality on human capital investments (Meltzer, 2001; Geronimus et al., 1999;
Geronimus, 1987). As these papers make clear, such costs are likely to be very substantial.
Page 14
BIBLIOGRAPHY
Anderson, K. H. and R. V. Burkhauser (1984), "The Importance of the Measure of Health in
Empirical Estimates of the Labor supply of Older Men," Economics Letters 16: 375-80.
Anderson, K. H. and R. V. Burkhauser (1985), “The Retirement-Health Nexus: A New
Measure of an Old Puzzle,” Journal of Human Resources 20: 315-30.
Barsky R, J Bound, K Charles, J Lupton (2002a) “A Precise Characterization of The BlackWhite Wealth Gap.” Manuscript, University of Michigan.
Barsky R, J Bound, K Charles, J Lupton (2002b). “Accounting for the Black-White Wealth Gap:
A Nonparametric Approach” Journal of American Statistical Association 97:663-673.
Blinder, A S (1973), “Wage Discrimination: Reduced Form and Structural Estimates,” Journal
of Human Resources, 8 (4), 436-455.
Bound J, M Schoenbaum, T Stinebrickner, T Waidmann (1998). “Modeling the Effects of
Health on Retirement Behavior,” Paper Presented at the National Bureau of Economic
Research Summer Institute..
Bound J, M Schoenbaum, T Waidmann (1995). “Race and Education Differences in Disability
Status and Labor Force Attachment,” Journal of Human Resources 30(Supplement):S227S267.
Bound J, M Schoenbaum, T Waidmann (1996). “Race Differences in Labor Force Attachment
and Disability Status,” The Gerontologist 36:311-321.
Bound, J (1991). “Self-reported vs. Objective Measures of Health in Retirement Models,”
Journal of Human Resources 26: 106-138.
Crimmins, E., Saito Y. and Reynolds, S. L. 1997. “Further Evidence on Recent Trends in the
Prevalence and Incidence of Disability Among Older Americans From Two Sources: the
LSOA and the NHIS.” Journals of Gerontology, Series B: Psychological Sciences and Social
Sciences 52B:S59-S71.
DiNardo, J, N M Fortin, N M. and T Lemieux 1996. “ Labor Market Institutions and the
Distribution of Wages, 1973-1992: A Semiparametric Approach.” Econometrica v64, n5:
1001-44.
Ferraro, K. F. 1980. "Self-ratings of Health among the Old and Old-old," Journal of Health and
Social Behavior 21: 377-83.
Geronimus, A.T. 1987. "On Teenage Childbearing and Neonatal Mortality in the United
States." Population and Development Review, 1987; 13,2:245-279.
Page 15
Geronimus, A.T., Bound J., Waidmann, T.A. .1999. "Health Inequality and Population Variation
in Fertility-Timing." Social Science and Medicine 49(12):1623-1636.
Geronimus, A. T., J Bound, T Waidmann, C Colen and D Steffick, 2001, “Inequality in Life
Expectancy, Functional Status, and Active Life Across Selected Black and White
Populations in the United States.” Demography, 38:227-251.
Harris, D. 1994, "The 1990 Census Count of American Indians." Social Science Quarterly, 75,
no. 2 : 580-593.Hayward, M.D. and M. Heron. 1999. “Racial Inequality in Active Life
among Adult Americans.” Demography 36: (1) 77-91.
Heckman, J. J., H Ichimura, and P. E. Todd. 1998. "Matching as an Econometric Evaluation
Estimator." The Review of Economic Studies, 65, 261-294.
Hirth RA, Chernew ME, Miller E, Fendrick AM, Weissert WG. (2000). “Willingness to pay for
a quality-adjusted life year: in search of a standard.” Medical Decision Making 20(3):332-42.
Idler, E. and Y. Benyamini. 1997. “Self-Rated Health and Mortality: A Review of Twenty-Seven
Community Studies.” Journal of Health and Social Behavior 39:21-37.
Kitagawa, E. (1964) “Standardized Comparisons in Population Research,” Demography 1:296315.
LaRue, A., L. Bank, L. Jarvic, and M. Hewtland (1979), "Health in Old Age: How Physicians'
Rating and Self-ratings Compare," Journal of Gerontology 34: 687-91.
Maddox, G., and E. Douglas (1973), “Self -Assessment of Health: A Longitudinal Study of
Elderly Subjects,” Journal of Health and Social Behavior 14: 87-93.
Manning, W., J. Newhouse, and J. Ware, Jr. 1982. “The Status of Health in Demand Estimation,
or Beyond Excellent, Good, Fair and Poor. Pp. 143-184 in Economic Aspects of Health,
edited by V. Fuchs. Chicago: University of Chicago Press.
Manning, W.G. Jr., J. P. Newhouse, and J. E. Ware Jr. (1982), "The Status of Health in Demand
Estimation, or Beyond Excellent, Good, Fair and Poor" in V. R. Fuchs (ed.), Economic
Aspects of Health (University of Chicago Press, Chicago).
Manton, K., C. Patrick, and K. Johnson. 1987. “Health Differentials between Blacks and Whites:
Recent Trends in Morbidity and Mortality.” Milbank Memorial Fund Quarterly 65 (Suppl.
1):129-199.
McGee DL, Liao Y, Cao G, Cooper RS. (1999) Self-reported health status and mortality in a
multiethnic US cohort. Am J Epidemiol 149(1):41-6.
Meltzer, D. O., 2001 “Effects of Health Disparities on Educational Investments,” paper
presented at the National Institute of Health Conference on Health Disparities, Washington,
D.C., 2001.
Page 16
Mossey, J. M. and E. Shapiro 1982. "Self-Rated Health: A Predictor of Mortality among the
Elderly," American Journal of Public Health 72: 800-808.
Murphy, K M. and R H. Topel, 2001. “Black-White Differences in the Economic Value of
Improving Health” paper presented at the National Institute of Health Conference on Health
Disparities, Washington, D.C., 2001.
Nagi, S. Z. 1969. "Congruency in Medical and Self-assessment of Disability," Industrial
Medicine 38: 74-83.
Oaxaca, R. 1973. "Male-Female Differentials in Urban Labor Markets." International Economic
Review, 14, 693-709.
Parsons, D. O. 1982., "The Male Labor Force Participation Decision: Health, Reported Health,
and Economic Incentives," Economica 49:81-91.
Rosenbaum, P. R., and Rubin, D. B. 1983, "The Central Role of the Propensity Score in
Observational Studies for Causal Effects." Biometrika, 70 (1), 41-55.
Rosenbaum, P. R., and Rubin, D. B. 1984, "Reducing Bias in Observational Studies Using
Subclassification on the Propensity Score." Journal of the American Statistical Association,
79 (387), 516-524.
Schoenbaum M, T Waidmann, 1997. “Race, Socioeconomic Status and Health: Accounting for
Race Differences in Health in the US Elderly,” Journals of Gerontology 52(Special):61-73.
Shalala DE, MH Trujillo, GJ Hartz, EL Paisano. 1999. Trends in Indian Health.
http://www.ihs.gov/PublicInfo/Publications/trends98/trends98.asp. Washington: Department
of Health and Human Services.
Snipp, C. M., 1989, American Indians : the first of this land. New York : Russell Sage
Foundation.
Social Security Administration. 2000. Social Security Bulletin: Annual Statistical Supplement,
2000. Washington, DC: U.S. Government Printing Office.
Waidmann, T, J Bound, and M Schoenbaum. 1995, "The Illusion of Failure: Trends in Selfreported Health of the U.S. Elderly." The Milbank Quarterly, 73, No. 2: 253-87.
Page 17
Figure 1: Proportion employed, by limitation status
(men aged 45-54)
Proportion employed
1.00
0.80
0.60
0.40
0.94
White
Black
0.86
0.42
0.20
0.22
0.00
Not limited
Health limits work
Limitation status
Figure 2: Standardized proportion employed
(men aged 45-54)
Proportion Employed
1.00
0.90
0.80
0.70
0.60
0.50
0.885
\
0.753
0.805
0.823
0.795
White (actual)
White (actual)
Black (actual)
Black, with health standardized to White levels
Black, with education standardized to White levels
Black, with educ & health standardized to White levels
Black
Native American
Fig. 3c: Proportion with mobility limit, male
White
Age
White
Age
Black
Native American
-
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
-
0.05
0.10
0.15
0.20
0.25
0.30
0.35
White
Age
Black
Native American
Fig. 3d: Proportion with personal care limit, male
Figures plotted are 5-year moving averages of weighted single-year prevalence estimates.
-
0.02
0.04
0.06
0.08
0.10
0.12
0.14
-
0.05
0.15
0.10
25
25
0.20
28
28
0.25
58
58
0.35
0.30
61
61
0.40
64
64
0.40
28
28
31
31
31
31
34
34
34
White
34
37
37
37
37
40
40
40
Age
Black
40
43
43
43
43
46
46
46
46
49
49
49
Native American
49
52
52
52
52
55
55
55
55
0.45
25
25
Fig. 3b: Proportion unable to work, male
58
58
Fig. 3a: Proportion with work limitation, male
61
61
Page 18
64
64
Black
Native American
Fig. 4c: Proportion with mobility limit, female
White
Age
White
Black
Age
Native American
-
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
-
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Fig. 4b: Proportion unable to work, female
White
Black
Age
Native American
Fig. 4d: Proportion with personal care limit, female
Figures plotted are 5-year moving averages of weighted single-year prevalence estimates.
-
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
-
0.05
0.10
0.15
0.20
58
58
0.25
61
61
0.30
64
64
0.35
25
25
28
28
28
28
31
31
31
31
34
34
34
White
34
37
37
37
37
40
40
40
Age
Black
40
43
43
43
43
0.40
25
25
46
46
46
46
49
Native American
49
49
49
52
52
52
52
55
55
55
55
58
58
Fig. 4a: Proportion with work limitation, female
61
61
Page 19
64
64
Page 20
Table 1: Proportion of people with health conditions that limit work, mobility, or personal care
Male
Female
Native
Native
AGE
EDUCATION GROUPa
White
Black
American
White
Black
American
HEALTH LIMITS KIND OR AMOUNT OF WORK
0.054
0.080
0.095
0.043
0.057
0.081
Whole sample
-0.062
0.063
-0.047
0.055
Ed standardized to Whites
25-34
0.136
0.151
0.140
0.113
0.109
0.117
HS dropouts
0.054
0.066
0.083
0.042
0.049
0.073
HS grads
0.016
0.020
0.038
0.018
0.022
0.042
Coll grads
0.200
0.285
0.350
0.165
0.295
0.317
Whole sample
-0.240
0.321
-0.254
0.288
55-64
0.318
0.347
0.414
0.275
0.376
0.384
HS dropouts
0.184
0.225
0.305
0.134
0.225
0.255
HS grads
0.095
0.133
0.210
0.085
0.116
0.206
Coll grads
HEALTH PREVENTS WORK
0.018
0.043
0.037
0.017
0.031
0.040
Whole sample
-0.029
0.020
-0.023
0.022
25-34
0.072
0.102
0.069
0.069
0.077
0.070
HS dropouts
0.015
0.029
0.027
0.015
0.023
0.033
HS grads
0.002
0.004
0.009
0.003
0.006
0.004
Coll grads
0.130
0.224
0.269
0.125
0.248
0.260
Whole sample
-0.175
0.225
-0.205
0.222
55-64
0.246
0.291
0.355
0.236
0.332
0.343
HS dropouts
0.108
0.155
0.204
0.093
0.173
0.181
HS grads
0.038
0.069
0.083
0.047
0.068
0.116
Coll grads
HEALTH LIMITS MOBILITY
0.012
0.025
0.021
0.011
0.019
0.023
Whole sample
-0.017
0.008
-0.014
0.010
25-34
0.047
0.061
0.039
0.045
0.047
0.035
HS dropouts
0.009
0.016
0.015
0.009
0.014
0.021
HS grads
0.002
0.005
0.011
0.003
0.004
0.005
Coll grads
0.044
0.090
0.099
0.054
0.115
0.116
Whole sample
-0.071
0.086
-0.098
0.100
55-64
0.085
0.118
0.129
0.101
0.149
0.151
HS dropouts
0.035
0.060
0.075
0.040
0.085
0.079
HS grads
0.016
0.030
0.044
0.023
0.040
0.076
Coll grads
HEALTH LIMITS PERSONAL CARE
0.019
0.067
0.050
0.016
0.066
0.048
Whole sample
-0.055
0.026
-0.057
0.035
25-34
0.053
0.108
0.086
0.050
0.111
0.072
HS dropouts
0.017
0.058
0.036
0.015
0.060
0.043
HS grads
0.007
0.031
0.038
0.007
0.030
0.017
Coll grads
0.048
0.123
0.106
0.049
0.127
0.100
Whole sample
-0.103
0.091
-0.112
0.090
55-64
0.083
0.150
0.138
0.087
0.157
0.124
HS dropouts
0.041
0.096
0.078
0.038
0.101
0.072
HS grads
0.019
0.057
0.053
0.021
0.062
0.087
Coll grads
a
NOTE: "Ed standardized" includes whole sample, with distribution of education standardized to White levels
SOURCE: Authors' calculations using 1990 US Census
EDUCATION GROUPd
NOTE: Linear regression, with dependent variable being the natural log of average weekly earnings, among those with non-zero earnings in 1989,
weighted by the number of weeks worked
NOTE: Represents largest standard error estimate for any cell within the column
NOTE: Pooled sample includes all levels of education; other rows are restricted to respondents in respective categories of educational attainment
d
e
age and educational attainment (within education category, as appropriate). Models are stratified by race/ethnicity and gender.
NOTE: Logistic regression, with dependent variable that is 1 if respondent was working for pay on April 1, 1990
b
c
Native
Native
Native
Native
White Black American White Black American White Black American White Black American
-2.32 -2.36
-1.54
-1.32 -1.95
-1.33
-0.33 -0.25
-0.15
-0.26 -0.21
-0.15
-2.38 -2.42
-1.54
-1.53 -2.05
-1.19
-0.37 -0.36
0.07
-0.30 -0.29
-0.13
-2.33 -2.34
-1.54
-1.33 -1.92
-1.41
-0.31 -0.20
-0.23
-0.27 -0.19
-0.15
-1.98 -2.08
-1.98
-0.96 -1.97
-0.75
-0.32 -0.22
-0.43
-0.22 -0.22
-0.24
-2.23 -2.84
-2.49
-1.82 -2.59
-2.06
-0.25 -0.24
-0.32
-0.22 -0.18
-0.44
-2.68 -3.13
-2.82
-2.17 -2.72
-2.10
-0.22 -0.26
-0.25
-0.18 -0.15
-0.20
-2.09 -2.53
-2.43
-1.70 -2.48
-2.17
-0.22 -0.20
-0.42
-0.22 -0.18
-0.40
-1.74 -2.20
-1.30
-1.51 -2.29
-1.99
-0.36 -0.30
-0.07
-0.34 -0.32
-1.28
0.05
0.18
0.58
0.04 0.14
0.50
0.02 0.07
0.29
0.03 0.07
0.26
Women
LOG WEEKLY EARNINGSc
Men
Women
NOTE: "Poor health" represented as a dummy variable that is 1 if respondent's health limits or prevents work; other explanatory variables are
a
Pooled sample
HS dropouts
25 to 34
HS grads
Coll grads
Pooled sample
HS dropouts
55 to 64
HS grads
Coll grads
Maximum Standard Errore
AGE
Men
EMPLOYEDb
Table 2: Parameter estimate for the effect of health limitation on labor market outcomesa
Page 21
age and education (within education category, as appropriate). Blank cells due to low participation rates plus limited sample size.
d
NOTE: Logistic regression, with dependent variable that is 1 if respondent received any income from Social Security (SSDI or ESRD) in 1989
b
c
Native American
1.70
1.59
1.76
-1.41
1.03
1.92
5.30
0.29
a
Table 3: Parameter estimate for the effect of health limitation on receipt of income from public transfer programsa
ANY INCOME FROM SOCIAL SECURITYb
Men
Women
AGE EDUCATION GROUPc
White
Black
Native American
White
Black
3.12
3.19
3.26
2.51
1.99
Pooled sample
25 to HS dropouts
3.22
2.99
3.37
2.23
1.67
34
3.13
3.32
3.34
2.60
2.20
HS grads
2.81
3.13
2.39
2.75
2.46
Coll grads
2.05
1.63
1.56
1.31
1.08
Pooled sample
55 to HS dropouts
2.01
1.56
1.28
1.08
0.93
64
2.10
1.76
2.06
1.45
1.35
HS grads
1.85
1.83
1.62
1.37
1.29
Coll grads
0.09
0.38
0.38
1.55
0.09
Maximum Standard Errord
Page 22
"Health" indicates that distribution of work limitation among Blacks/Native Americans is standardized to the White distribution
"Education" indicates that distribution of education among Blacks/Native Americans is standardized to the White distribution
"Age" indicates that distribution of age among Blacks/Native Americans is standardized to the White distribution
d
NOTE: Outcome is mean and median weekly earnings, among those with non-zero earnings in 1989, weighted by weeks worked
NOTE: Outcome is 1 if the respondent was working for pay on April 1, 1990 and 0 otherwise
b
c
Black
$595
$464
$498
$463
$497
$835
$612
$693
$614
$694
$ 0.07
MEAN WEEKLY EARNINGSc
Men
Women
Native
Native
American
Black
American
$595
$408
$408
$488
$386
$320
$551
$408
$356
$488
$386
$319
$554
$408
$356
$835
$397
$397
$670
$430
$388
$745
$462
$438
$683
$435
$379
$759
$464
$429
$ 0.83
$ 0.05
$ 0.70
NOTE: Table shows outcomes for Blacks/Native Americans, standardizing for race/ethnicity differences in various characteristics
"White" row represents actual level of each outcome observed among Whites in the age category.
a
Table 4: Simulated labor force outcomes for Blacks/Native Americansa
PROPORTION EMPLOYEDb
Men
Women
Native
Native
AGE REWEIGHTED BYa
Black
American
Black
American
0.900
0.900
0.719
0.719
White (actual)
0.729
0.703
0.655
0.567
age
25 to
0.778
0.758
0.709
0.646
age, education
34
0.744
0.717
0.664
0.580
age, health
0.783
0.765
0.712
0.653
age, education, health
0.654
0.654
0.440
0.440
White (actual)
0.545
0.486
0.441
0.342
age
55 to
0.597
0.554
0.491
0.407
age,
education
64
0.590
0.558
0.499
0.396
age, health
0.615
0.604
0.527
0.450
1.6E-05
1.8E-04
1.2E-05
1.5E-04
Page 23
Page 24
Table 5: Simulated participation in Social Security by Blacks/Native Americansa,b
Men
Women
AGE
REWEIGHTED BYa
Black
Native American
Black
Native American
0.012
0.012
0.011
0.011
White (actual)
0.015
0.015
0.018
0.020
age
25 to 34
0.012
0.012
0.015
0.016
age, education
0.012
0.011
0.017
0.018
age, health
0.011
0.011
0.015
0.015
0.189
0.189
0.210
0.210
White (actual)
0.193
0.215
0.210
0.218
age
55 to 64
0.166
0.194
0.194
0.211
age, education
0.173
0.182
0.189
0.186
age, health
0.158
0.166
0.180
0.182
Maximum Standard Errorc
1.3E-05
1.5E-04
1.0E-05
1.3E-04
a
NOTE: Simulations are analogous to those in Table 4
b
NOTE: Outcome is 1 if the respondent received any income from Social Security in 1989 and 0 otherwise
c
Page 25
Table 6: Simulated mean household income after taxes & transfersa,b
Men
AGE
REWEIGHTED BY
Black
Native American
$27,122
$27,122
White (actual)
$18,369
$16,862
age
25 to 34
$20,723
$19,530
age, education
$18,714
$17,212
age, health
$20,844
$19,660
$31,607
$31,607
White (actual)
$22,024
$20,193
age
55 to 64
$26,839
$24,110
age, education
$23,110
$21,911
age, health
$27,345
$25,257
$0.93
$17.21
a
Black
$28,583
$17,117
$19,195
$17,381
$19,304
$25,221
$15,438
$17,816
$16,634
$18,486
$0.36
Women
Native American
$28,583
$17,863
$20,329
$18,242
$20,581
$25,221
$15,392
$17,980
$16,750
$19,165
$5.55
NOTE: Simulations are analogous to those in Table 4
b
NOTE: Outcome is net household income in 1989 from non-asset sources, subtracting income, property,
FICA and Social Security taxes, and adding transfers from Social Security and public assistance
c
APPENDIX: Page 1
APPENDIX 1: Measuring Health
The Census includes only four health measures, each regarding physical disability:
whether health limits the kind or amount of work the respondent can do; whether health prevents
the respondent from working; whether health limits the respondent’s mobility; and whether
health limits the respondent’s ability to care for him/herself.1 Unless otherwise noted, we
account for three out of four of these measures in our analyses, excluding the indicator of
whether the respondent is unable to work out of concern that this latter variable may be
particularly endogenous to labor force outcomes (e.g., Waidmann et al., 1995).2
The Census health items are much more general than the Activities of Daily Living (ADLs)
or Instrumental Activities of Daily Living (IADLs) items that are ascertained on more detailed
surveys such as the Medical Expenditure Panel Survey or the National Health Interview Survey.
In addition, more detailed health surveys generally assess other dimensions of health, such as the
presence of specific chronic conditions, which the Census does not assess.3
Little empirical attention has been paid to the validity of the Census disability items.
Andresen et al. (2000) found that responses to the more global Census questions do not
correspond precisely to responses to narrow questions about specific, individual ADLs or
IADLs. However, this finding is neither surprising nor particularly relevant here. The issues of
more immediate concern for this analysis are first, whether differences across race/ethnicity
groups in responses to the respective Census questions accurately reflect differences across these
groups in the age-specific functioning of the respective populations; and second, whether
observed differences in functioning as measured by the Census items across labor market groups
(e.g. the employed versus the non-employed) reflect true differences in functioning between
these groups.
Global self-reported measures of health, although subjective, are highly correlated with
clinical measures of morbidity and predictive of subsequent death, health care utilization, and
labor market behavior (Idler and Benyamini, 1997; Nagi, 1969; Maddox and Douglas, 1973;
LaRue et al., 1979; Ferraro, 1980; Mossey and Shapiro, 1982; Manning et al., 1982). What is
more, this appears to be true across racial and ethnic groups (McGee et al, 1998). Our own
results (reported below) mirror these findings, in that we observe a very strong association
between the functional limitation measure in the Census and labor market outcomes, overall and
for different race/ethnicity groups.
However, even if the Census health measures are reliable indicators of actual health, this
does not imply that using such measures as proxies for health when examining racial differences
in labor market outcomes will yield valid results. At issue is whether the Census measures are
systematically biased.
There are a number of reasons to be suspicious of the use of the kind of survey measures
1
The Census questionnaire asks, “Does this person have a physical, mental or other health condition that has lasted
for 6 or more months and which (a) limits the kind or amount of work this person can do at a job? (b) prevents this
person from working at a job? Because of a health condition that has lasted 6 or more months, does this person
have any difficulty (a) going outside the home, for example to shop or visit a doctor’s office? (b) taking care of his
or her own personal needs, such as bathing, dressing or getting around inside the home?”
2
Empirically, simulations that controlled for all four Census health measures yielded substantively similar results.
3
In addition, Census data are based on a mail survey, and are sometimes provided by proxy respondents (although
both are also true of data from some more detailed health surveys).
APPENDIX: Page 2
available on the Census for studying labor market outcomes (Parsons, 1982; Anderson and
Burkhauser, 1984, 1985; Bound, 1991; Waidmann et al., 1995). First, respondents are being
asked for subjective judgments, and there is no reason to expect that these judgments will be
entirely comparable across respondents. Second, responses may not be independent of the labor
market outcomes of interest. Third, since health may represent one of the few “legitimate”
reasons for a working aged man to be out of work, men out of the labor force may mention
health limitations to rationalize their behavior. Fourth, since disability benefits are available
only for those deemed incapable of work, men and women will have a financial incentive to
identify themselves as disabled, an incentive that will be particularly high for those for whom the
relative rewards from continuing to work are low. Lastly, since there is ample evidence that
responses to questions about health status depend on social context (e.g., Waidmann et al, 1995),
it seems possible that Blacks or Native Americans would answer the Census questions using a
different implicit scale than do whites.
Each of these problems will lead to a different kind of bias. The lack of comparability across
individuals represents measurement error that is likely to lead us to underestimate the impact of
health on labor market outcomes, while the endogeneity of self-reported health is likely to lead
us to our exaggerate its impact. Biases in our estimation of health's impact on outcomes will also
induce biases on coefficients of any variables correlated with health. In addition, the dependence
of self-reported health on the economic environment will induce a bias on estimates of the
impact of race/ethnicity on outcomes, regardless of whether we correctly measure the impact of
health itself. Finally, if race/ethnicity has a direct impact on reporting behavior, this will limit
the extent to which health disparities can account for differences in labor market outcomes,
independent of whether we correctly measure the impact of health itself.
To get a better understanding of these issues, it will be helpful to consider a simple statistical
model of the association between health and demographic factors on labor market outcomes. To
be concrete, we consider earnings (again, as an example of the type of outcome examined here),
denoted by Y. Y is a function of health status, denoted Ș; race, R, and other unmeasured
components, H . We are interesting estimating the extent to which race differences in health can
account for race differences in the labor market outcome we are considering. For this purpose,
we would ideally like to estimate two regressions:
Y
E 1 R O1K H
Y
E 1cR H c
(1)
By construction, H is uncorrelated with both R and Ș, and H c is uncorrelated with R ( H is the part
of Y not accounted for by R and Ș. Here, E 1c measures the race gap in earnings, while E 1
measures the racial gap adjusting for health differences between the races. The comparison
between these parameters gives a measure of the extent to which racial health disparities in
health account for racial disparities in labor market outcomes. Valid estimates of this
comparison clearly require valid estimates of E 1 .
In particular, we are interested in how the fact that we have only a proxy for Ș affects our
estimate of E 1 . We do not directly observe Ș but rather an indicator, self-reported disability,
APPENDIX: Page 3
denoted by H. H depends on health status Ș , but also potentially on R, and, again on other
random components P 1 ,
H = E 2 R + O2 K + P1
(2)
Since Ș is unobserved, the sign of O 1 in (1) is arbitrary, but if larger values of Ș are
associated with better health then we would expect that O 1 should be positive as well.
Assuming the same for H, O 2 will also be positive. For convenience of exposition, we define R
in a way that E 1 is also positive (so in regressions that include Blacks and Whites, or Native
Americans and whites, R might be an indicator for being white). If, conditional on actual health
Ș, Blacks (or Native Americans) are no more nor less likely to report themselves in poor health,
then E 2 = 0 . On the other hand, if Blacks are less likely to report themselves in poor health,
conditional on actual health, then E 2 < 0 , while if blacks are more likely to report to report
themselves in poor health, E 2 > 0 .
One possibility for estimating equation (1) is to use H as a proxy for Ș. There are a variety
of econometric problems with doing so. The correlation between H and P 1 introduces a
simultaneity bias on Ô 1 (the estimate of O 1 ), while variance in P 1 introduces errors-in-variables
bias. Errors in estimates of O 1 , and the dependence of H on R, translate into errors in estimates
of E 1 . In particular, letting rK ,R represent the correlation between Ș and R, and ȡ the
correlation between H and P 1 ; and normalizing O 2 to equal 1, it is easy to show that:
O 1 V K2 (1 - rK2 ,R ) + V H V P U
Oˆ1 =
V K2 (1 - rK2 ,R ) + V 2P
1
(3)
1
Eˆ 1 = E 1 + ( O 1 - Oˆ1 )
V K ,R
- Oˆ1 E 2
V 2R
(4)
As long as U > 0 , this correlation will impart an upward bias on Ô 1 , while V 2P will impart
1
the standard errors-in-variables downward bias on Ô 1 . Which of these is dominant depends on
the relative strength of these two forces. The bias on Eˆ 1 will depend on the biases on Ô 1 and
E 2 , respectively. Thus, even if the errors-in-variables and the simultaneity biases on Ô 1 were
to cancel, we will still tend to underestimate E 1 if E 2 > 0 and overestimate ȕ1 if E 2 < 0 .
The above expressions make clear that the biases on Ô 1 and Eˆ 1 may be quite substantial
even when H is a reliable measure of Ș (i.e., even when V 2P is quite small). They also make
1
clear that the magnitude and even the direction of the bias depend on the magnitude of several
different correlations. Even if the Census health measures are highly correlated with actual
health, estimates using these as a proxy for health may not give reliable results. Likewise, even
if the Census measures often represent rationalization, their use may not necessarily exaggerate
the role of health on labor-related outcomes.
APPENDIX: Page 4
Beliefs about the kinds of bias involved using the Census health measures as a proxy for
actual health implicitly reflect judgments about all quantities involved in the above expressions.
In particular, (4) illustrates that the validity of our analysis depends on whether differences in
the Census health measures across labor market groups (e.g. the employed versus the nonemployed) accurately reflect true health differences between these groups (i.e. on whether
Oˆ1 | O 1 ), and on whether differences across racial groups in responses to Census questions
accurately reflect differences across these groups in the age-specific functional capacities of the
respective populations (i.e. on whether Eˆ 2 | 0 .)
In previous work, we have taken advantage of more detailed information on health status
available in other datasets (e.g., the Health and Retirement Survey) to address the potential
biases that may affect estimates of E 1 (and O1 ), using instrumental variables methods (Bound,
1991; Bound et al, 1996, 1999). We have also used this work to compare results based on
instrumental variables methods with those based directly on the type of health measures
available in the Census (Bound et al., 1998). In general, our findings regarding the effects of
health on labor force outcomes have been consistent across different datasets and across
estimation methods. Indeed, if anything, the results from this previous work suggest that the
“naïve” use of Census-type health measures (or self-rated general health) tends to understate the
effects of health on labor market outcomes. This suggests that, empirically, the endogeneity and
errors-in-variables biases described above approximately offset each other.
One addition issue mentioned above is that the Census measures of physical limitation may
not accurately reflect race/ethnicity differences in health. In general terms, our results (reported
below) are consistent with race and age gradients in disability observed elsewhere, e.g., Blacks
have higher disability rates than Whites, with disparities most pronounced in young through
middle-adulthood and narrowing in old age. We have also compared the Census health measures
with comparable information from the National Health Interview Survey (NHIS) and found
similar patterns across these datasets (e.g., Waidmann et al., 1995; additional results are
available by request). Multivariate logistic regressions of disability status by age, race, and
education using the1994 disability supplement to the NHIS (NCHS 1996) and the 1990 Census,
respectively, show patterns that are quite similar across the two surveys.
APPENDIX: Page 5
APPENDIX 2: Simulation Methods
As we have described, the focus of this paper is to assess the possible effects of racial/ethnic
disparities in health (which we can denote by H) on employment rates, earnings, and income of
different racial/ethnic groups (we will refer to earnings, denoted by Y, as an example of the type
of outcome examined here). To examine these issues, we use data on Whites and on other
race/ethnicity groups in the population with systematically lower health status that Whites; in
essence, we reweight the other race/ethnicity populations in such a way that their distribution of
health is approximately coincident to the distribution of health actually observed in the White
population. In the context of discrete data, our method would be identical to standardization
methods familiar to demographers.
For expositional purposes, we focus in this section on disparities in H and Y between Whites
and Blacks (denoted by w and b, respectively). Throughout this section, we will adopt the
convention that capital letters indicate population variables, and lower-case letters indicate a
particular value of a variable. A standard method of assessing the extent to which race
differences in Y can be explained by race differences in H is given by Blinder (1973) and Oaxaca
(1973). The Blinder-Oaxaca decomposition is most often discussed in terms of regressions fitted
to sample data. For our purposes, however, it is convenient to begin with a slightly more
abstract discussion of the Blinder-Oaxaca logic, described in terms of conditional expectation
functions. This approach does not begin with the any particular presumption about the shape of
that the conditional expectation function for earnings given health. Further, by focusing
explicitly on the distribution of health for Blacks and for Whites, it sets the stage for all of the
later analysis. The exposition in this section is closely adapted from Barsky et al. (2001).
We denote race as R {w, b} . We let E[Y | R, H ] be the conditional expected function given
race and health, and g ( H | R) the distribution of health conditional on race. The expected value
of earnings for a given race is then given by
E[Y | R]
³ E[Y | R, h]g (h | R)dh
(5)
The Blinder-Oaxaca decomposition, which focuses on mean differences between groups, is thus
implicitly based on (5). The method performs the following thought experiment: imagine
integrating the conditional expectation of earnings for Whites over the Black health distribution,
Eb [Y | w] { ³ E[Y | w, h]g (h | b)dh
(6)
With this construction in mind, the Blinder-Oaxaca method decomposes the group difference in
mean earnings into the portion attributable to differences in the health distribution and the
portion attributable to differences in the conditional expectations function. These two
components are represented by the first and second terms in the following equality,
APPENDIX: Page 6
E[Y | w] Eb [Y | w] Eb [Y | w] E[Y | b]
³ E[Y | w, h]g (h | w) g (h | b)dh ³ E[Y | w, h] E[Y | b, h g (h | b)dh
E[Y | w] E[Y | b]
(7)
We note that differences in the health distribution are evaluated here using the White
conditional expectations function. Alternatively, differences in the earnings distributions may be
evaluated using the expectations function for Blacks, thereby focusing on the quantity
E w [Y | b] { ³ E[Y | b, h]g (h | w)dh
(8)
and carrying out the analogous decomposition. The two decompositions will produce similar
results as long as the slopes of the conditional wealth functions are similar.
Researchers typically interpret (6) and (8) in terms of counterfactuals. They regard (4) as
addressing the question: “what would happen to the mean earnings of Blacks if Black health
status were to rise to White levels?” Likewise, they see (6) as assessing the impact on the mean
earnings of Whites of a hypothetical downward shift of White health to levels characteristic of
the Black population. In the analyses presented here, we report outcomes among Blacks (and
Native Americans) when health (and other) characteristics of Blacks (and Native Americans)
are standardized to levels observed empirically among Whites. This approach has the
expositional advantage that Blacks and Native Americans are standardized to the same reference
group (i.e., Whites). In addition, it seems conceptually consistent with the original motivation
for examining health disparities, namely to address the inequity of worse outcomes among
certain race/ethnicity groups in the population.
As discussed above, (8) is frequently evaluated using regressions in which the relevant
outcome (i.e., earnings) is the dependent variable, and the explanatory variables include an
indicator for race and measures of health status. For reasons discussed in detail in Barsky et al.
(2001), we take a different approach here. In particular, we avoid making any assumption about
the functional form of the conditional expected earnings function in (4) by instead noting the
relationship between the Black and White health distribution. By Bayes’ rule,
g ( H | w) Y ( H ) g ( H | b) where the ‘weight’, Y (H ) , is given by
Y (H )
Pr( w | H ) Pr( w)
Pr(b | H ) Pr(b)
(9)
Thus, (8) can be evaluated by
E w [Y | b]
³Y (h) E[Y | b, h]g (h | b)dh
(10)
which is the weighted expected value of Black earnings where the weights reflect differences in
the health distributions between Black and White households.
To evaluate (4), we can use the empirical Black earnings distribution, thereby avoiding a
APPENDIX: Page 7
functional specification for E >Y | b, h @ . This requires an estimate of the weighting
function, Y (H ) . However, Y (H ) is a simple transformation of the propensity-to-score function
that has received considerable attention in both the statistics and econometrics literature
(Rosenbaum and Rubin, 1983, 1984; Heckman, Ichimura and Todd, 1998).
As highlighted by DiNardo, Fortin and Lemieux, in addition to yielding information on
mean effects of health disparities on earnings differences, the re-weighting method allows us to
examine the effect of standardization on the entire distribution of outcomes. Using the
weighting function (9) and focusing on the full Black earnings distribution
k (Y | b) ³ f (Y | b, h) g (h | b)dh , where f (Y | b, H ) is the conditional earnings distribution given
race and health, we can consider the following counterfactual earnings distribution,
k w Y | b { ³ f Y | b, h g h | wdh
³Y (h) f Y | b, h g h | b dh
(11)
where the last equality is simply the Black earnings distribution re-weighted in such a way that
the health distributions of Blacks and Whites are coincident. While the Blinder-Oaxaca
decomposition focuses on the mean counterfactual, (11) expresses the distributional
counterfactual.
The results of these standardizations identify the extent to which differences in health
observed between Blacks and Whites can account for race differences in earnings (or other labor
market outcomes). As we discuss further below, however, we do not necessarily interpret this
accounting as causal, in the sense that ameliorating the health differences would necessarily
yield that amount of narrowing in earnings differences. Furthermore, to the extent that observed
health differences do not fully account for differences in earnings between Blacks and Whites,
we do not interpret the residual earnings differences as causally related to being Black in any
meaningful way. At best, the remaining difference may be interpreted as an estimate of the
importance of factors other than health that effect race differences in earnings. At minimum it as
a measure of the average conditional (on health) gap in earnings between Blacks and Whites.
APPENDIX: Page 8
APPENDIX 3: Detailed Results
In the text we presented results for selected age groups. Here we present results for a more
complete set of age groups and labor market outcomes. For most outcomes, we examined
respondents aged 25-64, because the outcomes were less salient or difficult to interpret for both
younger and older groups. (For instance, employment rates are low for all race/ethnicity groups
after age 65, while Social Security participation is close to universal.) For total household
income post-tax/transfers, we also examined results for respondents aged 65 and older. Each
table in the appendix corresponds to the main table with the same number and title.
We note that Appendix Tables 4, 5, 6a and 6b contain results for outcomes that were not
reported in the text. Specifically, Appendix Table 4 includes results for median weekly earnings;
Table 5 contains results for participation in public assistance programs; Table 6a includes results
for mean and median household income, excluding taxes and public transfers; and Table 6b
includes results for median household income, net of taxes and public transfers. All Appendix
Tables include information on age groups not reported in the text.
APPENDIX: Page 9
Appendix Table 1: Proportion with health limitations
Male
AGE
25-34
34-44
45-54
55-64
25-34
34-44
45-54
55-64
EDUCATION GROUPa
White
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
0.054
-0.136
0.054
0.016
0.078
-0.198
0.084
0.033
0.109
-0.228
0.106
0.046
0.200
-0.318
0.184
0.095
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
0.018
-0.072
0.015
0.002
0.029
-0.117
0.028
0.006
0.052
-0.149
0.043
0.012
0.130
-0.246
0.108
0.038
Female
Native
Native
Black
American
White
Black
American
HEALTH LIMITS KIND OR AMOUNT OF WORK
0.080
0.095
0.043
0.057
0.081
0.062
0.063
-0.047
0.055
0.151
0.140
0.113
0.109
0.117
0.066
0.083
0.042
0.049
0.073
0.020
0.038
0.018
0.022
0.042
0.125
0.172
0.063
0.094
0.131
0.100
0.154
-0.079
0.116
0.209
0.235
0.164
0.171
0.192
0.110
0.162
0.061
0.082
0.121
0.047
0.115
0.031
0.037
0.073
0.180
0.235
0.096
0.174
0.217
0.146
0.203
-0.144
0.190
0.254
0.313
0.204
0.269
0.300
0.149
0.209
0.084
0.135
0.173
0.075
0.124
0.044
0.061
0.143
0.285
0.350
0.165
0.295
0.317
0.240
0.321
-0.254
0.288
0.347
0.414
0.275
0.376
0.384
0.225
0.305
0.134
0.225
0.255
0.133
0.210
0.085
0.116
0.206
HEALTH PREVENTS WORK
0.043
0.037
0.017
0.031
0.040
0.029
0.020
-0.023
0.022
0.102
0.069
0.069
0.077
0.070
0.029
0.027
0.015
0.023
0.033
0.004
0.009
0.003
0.006
0.004
0.071
0.073
0.028
0.056
0.069
0.051
0.058
-0.043
0.054
0.146
0.131
0.113
0.126
0.134
0.054
0.063
0.025
0.043
0.055
0.016
0.029
0.007
0.013
0.022
0.118
0.139
0.056
0.126
0.144
0.086
0.110
-0.098
0.116
0.191
0.213
0.153
0.217
0.224
0.083
0.113
0.043
0.087
0.106
0.030
0.038
0.015
0.028
0.047
0.224
0.269
0.125
0.248
0.260
0.175
0.225
-0.205
0.222
0.291
0.355
0.236
0.332
0.343
0.155
0.204
0.093
0.173
0.181
0.069
0.083
0.047
0.068
0.116
a
APPENDIX: Page 10
Appendix Table 1: Proportion with health limitations (continued)
Male
AGE
25-34
34-44
45-54
55-64
25-34
34-44
45-54
55-64
EDUCATION GROUPa
White
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
0.012
-0.047
0.009
0.002
0.016
-0.067
0.013
0.004
0.023
-0.066
0.019
0.007
0.044
-0.085
0.035
0.016
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
Whole sample
HS dropouts
HS grads
Coll grads
0.019
-0.053
0.017
0.007
0.021
-0.068
0.020
0.009
0.030
-0.069
0.027
0.012
0.048
-0.083
0.041
0.019
Female
Native
Black
American
White
Black
HEALTH LIMITS MOBILITY
0.025
0.021
0.011
0.019
0.017
0.008
-0.014
0.061
0.039
0.045
0.047
0.016
0.015
0.009
0.014
0.005
0.011
0.003
0.004
0.035
0.036
0.017
0.031
0.025
0.029
-0.025
0.075
0.066
0.065
0.064
0.024
0.028
0.014
0.025
0.010
0.019
0.007
0.009
0.053
0.062
0.029
0.064
0.039
0.052
-0.051
0.087
0.091
0.074
0.106
0.036
0.051
0.022
0.046
0.015
0.026
0.010
0.018
0.090
0.099
0.054
0.115
0.071
0.086
-0.098
0.118
0.129
0.101
0.149
0.060
0.075
0.040
0.085
0.030
0.044
0.023
0.040
HEALTH LIMITS PERSONAL CARE
0.067
0.050
0.016
0.066
0.055
0.026
-0.057
0.108
0.086
0.050
0.111
0.058
0.036
0.015
0.060
0.031
0.038
0.007
0.030
0.075
0.056
0.019
0.074
0.061
0.046
-0.064
0.120
0.088
0.064
0.120
0.066
0.052
0.017
0.068
0.035
0.024
0.008
0.034
0.094
0.073
0.030
0.099
0.078
0.060
-0.087
0.128
0.111
0.068
0.138
0.079
0.058
0.025
0.085
0.044
0.033
0.012
0.046
0.123
0.106
0.049
0.127
0.103
0.091
-0.112
0.150
0.138
0.087
0.157
0.096
0.078
0.038
0.101
0.057
0.053
0.021
0.062
a
Native
American
0.023
0.010
0.035
0.021
0.005
0.037
0.031
0.070
0.029
0.021
0.070
0.058
0.103
0.056
0.023
0.116
0.100
0.151
0.079
0.076
0.048
0.035
0.072
0.043
0.017
0.052
0.045
0.091
0.042
0.034
0.072
0.061
0.105
0.056
0.034
0.100
0.090
0.124
0.072
0.087
EDUCATION GROUPd
NOTE: Linear regression, with dependent variable being the natural log of average weekly earnings, among those with non-zero earnings in 1989,
weighted by the number of weeks worked
e
age and educational attainment (within education category, as appropriate). Models are stratified by race/ethnicity and gender.
NOTE: Logistic regression, with dependent variable that is 1 if respondent was working for pay on April 1, 1990
d
c
LOG WEEKLY EARNINGSc
Men
Women
Native
Native
Black American White Black American
-0.25
-0.15
-0.26 -0.21
-0.15
-0.36
0.07
-0.30 -0.29
-0.13
-0.20
-0.23
-0.27 -0.19
-0.15
-0.22
-0.43
-0.22 -0.22
-0.24
-0.25
-0.23
-0.25 -0.27
-0.28
-0.33
-0.31
-0.28 -0.28
-0.35
-0.21
-0.20
-0.24 -0.27
-0.22
-0.30
-0.25
-0.25 -0.22
-0.52
-0.25
-0.18
-0.26 -0.26
-0.18
-0.27
-0.30
-0.27 -0.25
-0.25
-0.23
-0.11
-0.24 -0.24
-0.16
-0.33
-0.14
-0.30 -0.35
-0.08
-0.24
-0.32
-0.22 -0.18
-0.44
-0.26
-0.25
-0.18 -0.15
-0.20
-0.20
-0.42
-0.22 -0.18
-0.40
-0.30
-0.07
-0.34 -0.32
-1.28
0.07
0.29
0.03 0.07
0.26
b
a
Pooled sample
HS dropouts
25 to 34
HS grads
Coll grads
Pooled sample
HS dropouts
35 to 44
HS grads
Coll grads
Pooled sample
HS dropouts
45 to 54
HS grads
Coll grads
Pooled sample
HS dropouts
55 to 64
HS grads
Coll grads
Maximum Standard Errore
AGE
Women
Native
Native
White Black American White Black American White
-2.32 -2.36
-1.54
-1.32 -1.95
-1.33
-0.33
-2.38 -2.42
-1.54
-1.53 -2.05
-1.19
-0.37
-2.33 -2.34
-1.54
-1.33 -1.92
-1.41
-0.31
-1.98 -2.08
-1.98
-0.96 -1.97
-0.75
-0.32
-2.65 -2.62
-1.89
-1.62 -2.47
-1.77
-0.32
-2.81 -2.83
-2.05
-1.95 -2.43
-1.94
-0.36
-2.65 -2.54
-1.84
-1.65 -2.47
-1.75
-0.29
-2.44 -2.51
-2.10
-1.16 -2.50
-1.73
-0.38
-2.86 -3.02
-2.43
-1.88 -2.81
-2.32
-0.31
-3.13 -3.27
-2.26
-2.16 -2.77
-2.28
-0.30
-2.78 -2.85
-2.58
-1.83 -2.80
-2.37
-0.29
-2.63 -2.77
-2.37
-1.62 -3.07
-2.45
-0.38
-2.23 -2.84
-2.49
-1.82 -2.59
-2.06
-0.25
-2.68 -3.13
-2.82
-2.17 -2.72
-2.10
-0.22
-2.09 -2.53
-2.43
-1.70 -2.48
-2.17
-0.22
-1.74 -2.20
-1.30
-1.51 -2.29
-1.99
-0.36
0.05
0.18
0.58
0.04 0.14
0.50
0.02
Men
EMPLOYEDb
Appendix Table 2: Parameter estimate for the effect of health limitation on labor market outcomesa
APPENDIX
Page 11
NOTE: Logistic regression, with dependent variable that is 1 if respondent received any income from Social Security (SSDI or ESRD) in 1989
e
age and education (within education category, as appropriate). Blank cells due to low participation rates plus limited sample size.
NOTE: Logistic regression, with dependent variable that is 1 if respondent received any public assistance (AFDC, SSI, or other) in 1989
d
c
b
a
Appendix Table 3: Parameter estimate for the effect of health limitation on receipt of income from public transfer programsa
ANY INCOME FROM SOCIAL SECURITYc
ANY INCOME FROM PUBLIC ASSISTANCEb
Men
Women
Men
Women
EDUCATION
Native
Native
Native
Native
AGE GROUPd
White Black American White Black American White Black American White Black American
2.95
2.69
2.07
1.93
1.25
1.11
3.12
3.19
3.26
2.51
1.99
1.70
Pooled sample
2.83
2.62
2.24
1.66
1.02
0.87
3.22
2.99
3.37
2.23
1.67
1.59
25 to HS dropouts
34
2.98
2.75
1.93
2.06
1.36
1.27
3.13
3.32
3.34
2.60
2.20
1.76
HS grads
3.38
2.69
-2.66
2.64
0.58
2.81
3.13
2.39
2.75
2.46
-Coll grads
2.83
2.51
1.77
2.40
1.80
1.47
3.34
3.22
3.08
2.45
2.01
1.79
Pooled sample
2.76
2.52
1.44
2.21
1.59
1.34
3.63
3.36
3.28
2.15
1.61
1.37
35 to HS dropouts
44
2.81
2.46
1.93
2.47
1.91
1.45
3.32
3.15
2.93
2.52
2.22
1.83
HS grads
3.14
3.09
3.52
2.74
2.57
2.65
3.08
3.15
3.55
2.73
2.66
3.39
Coll grads
2.86
2.61
2.12
2.87
2.12
1.85
3.63
3.14
3.55
2.70
2.09
1.69
Pooled sample
2.88
2.62
2.20
2.62
1.94
1.50
3.73
2.99
3.38
2.42
1.66
1.24
45 to HS dropouts
54
2.85
2.57
1.96
3.06
2.38
2.35
3.62
3.30
3.93
2.83
2.46
2.32
HS grads
2.85
2.77
2.55
3.00
2.41
-3.47
3.31
2.47
2.76
3.07
2.64
Coll grads
2.01
1.96
1.46
2.41
1.97
1.63
2.05
1.63
1.56
1.31
1.08
1.41
Pooled sample
2.12
1.98
1.51
2.35
1.86
1.46
2.01
1.56
1.28
1.08
0.93
1.03
55 to HS dropouts
64
1.88
1.90
1.37
2.48
2.26
2.02
2.10
1.76
2.06
1.45
1.35
1.92
HS grads
1.88
2.37
-2.26
1.93
3.18
1.85
1.83
1.62
1.37
1.29
5.30
Coll grads
e
0.10
0.47
0.82
0.10
0.29
1.37
0.09
0.38
1.55
0.09
0.29
1.30
Maximum Standard Error
APPENDIX
Page 12
0.900
0.703
0.758
0.717
0.765
0.912
0.735
0.786
0.772
0.812
0.885
0.696
0.756
0.760
0.798
0.654
0.486
0.554
0.558
0.604
0.900
0.729
0.778
0.744
0.783
0.912
0.764
0.813
0.789
0.823
0.885
0.755
0.806
0.796
0.825
0.654
0.545
0.597
0.590
0.615
White (actual)
age
Native
American
Black
REWEIGHTED BYa
0.441
0.491
0.499
0.527
0.678
0.733
0.721
0.756
0.440
0.731
0.778
0.746
0.785
0.697
0.655
0.709
0.664
0.712
0.745
0.719
Black
0.342
0.407
0.396
0.450
0.566
0.644
0.621
0.684
0.440
0.637
0.693
0.664
0.711
0.697
0.567
0.646
0.580
0.653
0.745
0.719
Native
American
Women
$612
$693
$614
$694
$636
$716
$638
$717
$835
$599
$664
$601
$664
$907
$464
$498
$463
$497
$819
Mean
$595
Men
$476
$556
$477
$557
$538
$620
$538
$620
$662
$496
$558
$502
$558
$744
$372
$403
$372
$403
$682
Median
$517
Black
$670
$745
$683
$759
$675
$738
$674
$733
$835
$606
$666
$614
$670
$907
$488
$551
$488
$554
$819
Mean
$595
$496
$538
$496
$538
$538
$603
$541
$620
$662
$496
$538
$500
$558
$744
$372
$414
$372
$414
$682
Median
$517
Native American
$430
$462
$435
$464
$466
$499
$469
$499
$397
$471
$504
$472
$504
$440
$386
$408
$386
$408
$453
Mean
$408
a
$310
$342
$310
$344
$375
$414
$382
$414
$331
$393
$417
$393
$419
$372
$310
$333
$313
$333
$382
$0.70
$388
$438
$379
$429
$397
$428
$400
$430
$397
$404
$441
$408
$445
$440
$320
$356
$319
$356
$453
Mean
$408
$278
$323
$284
$324
$314
$346
$318
$349
$331
$310
$333
$310
$338
$372
$269
$299
$269
$301
$382
Median
$364
Native American
Women
Median
$364
Black
WEEKLY EARNINGSc
1.6E-05 1.8E-04
1.2E-05
1.5E-04
$0.07
$0.83
$0.05
NOTE: Table shows outcomes for Blacks/Native Americans, standardizing for race/ethnicity differences in various characteristics
"White" row represents actual level of each outcome observed among Whites in the age category.
"Age" indicates that distribution of age among Blacks/Native Americans is standardized to the White distribution
"Education" indicates that distribution of education among Blacks/Native Americans is standardized to the White distribution
"Health" indicates that distribution of work limitation among Blacks/Native Americans is standardized to the White distribution
b
NOTE: Outcome is 1 if the respondent was working for pay on April 1, 1990 and 0 otherwise
c
NOTE: Outcome is mean and median weekly earnings, among those with non-zero earnings in 1989, weighted by weeks worked
d
age, health
age, health, ed
55 to 64 age, education
age
age, health
age, health, ed
White (actual)
age
age, health
age, health, ed
White (actual)
age
age, health
age, health, ed
White (actual)
AGE
Men
PROPORTION EMPLOYEDb
Appendix Table 4: Simulated labor force outcomes for Blacks/Native Americansa
APPENDIX
Page 13
REWEIGHTED BYa
Native American
0.017
0.052
0.037
0.045
0.035
0.015
0.058
0.042
0.045
0.034
0.015
0.074
0.051
0.050
0.040
0.022
0.094
0.055
0.069
0.047
9.7E-05
Black
0.040
0.168
0.131
0.162
0.129
0.027
0.116
0.089
0.106
0.085
0.022
0.105
0.078
0.083
0.068
0.030
0.138
0.103
0.101
0.083
7.1E-06
Native American
0.040
0.149
0.111
0.140
0.106
0.027
0.115
0.090
0.100
0.082
0.022
0.112
0.074
0.081
0.058
0.030
0.146
0.102
0.107
0.079
9.1E-05
Black
0.012
0.015
0.012
0.012
0.011
0.018
0.028
0.022
0.021
0.019
0.029
0.047
0.036
0.033
0.030
0.189
0.193
0.166
0.173
0.158
1.3E-05
Native American
0.012
0.015
0.012
0.011
0.011
0.018
0.031
0.025
0.018
0.017
0.029
0.047
0.040
0.025
0.025
0.189
0.215
0.194
0.182
0.166
1.5E-04
NOTE: Outcome is 1 if the respondent received any income from Social Security in 1989 and 0 otherwise
Black
0.011
0.018
0.015
0.017
0.015
0.018
0.030
0.025
0.026
0.024
0.024
0.043
0.037
0.033
0.030
0.210
0.210
0.194
0.189
0.180
1.0E-05
Native American
0.011
0.020
0.016
0.018
0.015
0.018
0.029
0.027
0.024
0.022
0.024
0.042
0.034
0.030
0.026
0.210
0.218
0.211
0.186
0.182
1.3E-04
ANY INCOME FROM SOCIAL SECURITYc
Men
Women
NOTE: Outcome is 1 if the respondent received any income from public assistance in 1989 and 0 otherwise
d
c
Black
0.017
0.040
0.030
0.033
0.027
0.015
0.046
0.032
0.036
0.029
0.015
0.052
0.036
0.038
0.032
0.022
0.071
0.048
0.056
0.044
7.0E-06
NOTE: Simulations are analogous to those in Appendix Table 4
b
a
White (actual)
age
25 to
age, education
34
age, health
age, health, ed
White (actual)
age
35 to
age, education
44
age, health
age, health, ed
White (actual)
age
45 to
age, education
54
age, health
age, health, ed
White (actual)
age
55 to
age, education
64
age, health
age, health, ed
AGE
Appendix Table 5: Simulated participation in transfer programs for Blacks/Native Americansa
ANY INCOME FROM PUBLIC ASSISTANCEb
Men
Women
APPENDIX
Page 14
Median
$32,253
$15,717
$18,707
$16,127
$19,014
$40,854
$22,577
$26,878
$23,652
$26,878
$38,000
$20,923
$25,211
$22,577
$25,937
$20,427
$9,931
$13,976
$12,041
$15,094
$7.82
Native American
Mean
Median
$37,005
$32,253
$20,864
$16,127
$24,619
$20,427
$21,404
$17,146
$24,965
$20,691
$47,929
$40,854
$28,355
$22,577
$32,344
$26,878
$29,497
$23,749
$33,092
$27,308
$46,443
$38,000
$26,667
$19,100
$31,496
$24,526
$28,784
$21,502
$33,064
$25,802
$29,740
$20,427
$15,964
$8,856
$19,610
$12,984
$18,071
$11,476
$21,453
$16,000
Women
income, & income from estates/trusts)
NOTE: Outcome is gross household income in 1989 from non-asset sources (earned income & benefits; excludes interest, dividends, royalties, rental
c
$0.65
Mean
$37,005
$20,140
$23,228
$20,511
$23,378
$47,929
$28,226
$32,339
$29,000
$32,665
$46,443
$26,813
$30,971
$28,478
$31,804
$29,740
$16,730
$20,079
$18,512
$21,089
Black
NOTE: Simulations are analogous to those in Appendix Table 4, but all models also account for race/ethnicity differences in marital status
b
a
Appendix Table 6a: Household Income before taxes & transfersa,b
Men
Black
Native American
AGE REWEIGHTED BYa
Mean
Median
Mean
Median
$35,238
$30,860
$35,238
$30,860
White (actual)
$22,725
$18,910
$20,365
$16,219
age
25 to
$26,097
$21,502
$24,274
$20,427
age, education
34
$23,225
$19,352
$20,882
$17,202
age, health
$26,279
$21,723
$24,475
$20,427
age, health, ed
$48,761
$41,929
$48,761
$41,929
White (actual)
$32,637
$28,813
$28,887
$24,727
age
35 to
$38,402
$34,403
$34,158
$30,000
age,
education
44
$33,718
$30,103
$30,414
$26,301
age, health
$38,862
$34,403
$35,415
$31,178
age, health, ed
$52,719
$44,079
$52,719
$44,079
White (actual)
$34,979
$31,178
$30,895
$25,802
age
45 to
$42,059
$37,629
$36,767
$32,253
age, education
54
$36,656
$32,253
$33,332
$28,938
age, health
$42,873
$38,328
$38,287
$33,515
age, health, ed
$40,650
$32,253
$40,650
$32,253
White (actual)
$26,352
$20,280
$23,368
$17,202
age
55 to
$33,555
$27,202
$29,223
$24,000
age,
education
64
$28,066
$22,102
$26,157
$20,427
age, health
$34,373
$28,277
$31,171
$26,340
age, health, ed
c
Maximum Standard Error
$1.05
$10.81
APPENDIX
Page 15
$24,820
$15,976
$18,210
$16,285
$18,380
$33,156
$23,475
$27,391
$24,183
$27,654
$34,632
$24,748
$29,544
$25,880
$30,008
$27,199
$18,349
$23,365
$19,644
$23,970
$18,007
$11,703
$15,401
$12,257
$15,686
$12,965
$8,400
$10,198
$8,575
$10,380
$27,122
$18,369
$20,723
$18,714
$20,844
$36,281
$25,580
$29,439
$26,291
$29,737
$38,545
$27,083
$31,754
$28,154
$32,261
$31,607
$22,024
$26,839
$23,110
$27,345
$21,881
$15,103
$19,361
$15,678
$19,656
$15,947
$11,056
$13,619
$11,354
$13,835
White (actual)
age
age, health
age, health, ed
White (actual)
age
age, health
age, health, ed
White (actual)
age
age, health
age, health, ed
White (actual)
age
age, health
age, health, ed
White (actual)
age
age, health
age, health, ed
White (actual)
Men
$11,668
$13,909
$11,933
$14,154
$15,947
$14,710
$17,378
$15,424
$18,154
$21,881
$20,193
$24,110
$21,911
$25,257
$31,607
$20,193
$24,110
$21,911
$25,257
$38,545
$23,150
$26,680
$24,115
$27,451
$36,281
$16,862
$19,530
$17,212
$19,660
$27,122
$8,610
$9,600
$8,803
$9,684
$12,965
$11,460
$14,359
$11,986
$14,863
$18,007
$16,844
$21,125
$18,703
$22,338
$27,199
$16,844
$21,125
$18,703
$22,338
$34,632
$20,780
$24,080
$21,866
$25,009
$33,156
$14,596
$17,279
$15,000
$17,454
$24,820
Native American
Mean
Median
$6,938
$7,950
$7,112
$8,063
$9,955
$10,236
$12,109
$10,806
$12,455
$16,627
$15,438
$17,816
$16,634
$18,486
$25,221
$21,396
$24,192
$22,485
$24,729
$34,407
$22,708
$25,483
$23,233
$25,705
$35,837
$17,117
$19,195
$17,381
$19,304
$28,583
Mean
a
$0.93
$17.21
$0.36
NOTE: Simulations are analogous to those in Appendix Table 4
b
NOTE: Outcome is net household income in 1989 from non-asset sources, subtracting income, property, FICA and Social
Security taxes, and adding transfers from Social Security and public assistance
c
age
75 and
age, education
over
age, health
age, health, ed
Median
Mean
REWEIGHTED BY
AGE
Black
Appendix Table 6b: Household Income after taxes & transfersa,b
Black
$5,173
$5,800
$5,268
$5,899
$7,250
$7,024
$8,471
$7,435
$8,832
$13,186
$11,239
$14,006
$12,709
$14,893
$20,615
$17,530
$20,735
$18,738
$21,207
$30,326
$19,190
$22,050
$19,793
$22,285
$32,477
$14,204
$16,141
$14,473
$16,245
$26,142
Median
Women
$5.55
$7,668
$8,572
$7,886
$8,816
$9,955
$12,175
$13,810
$12,815
$14,168
$16,627
$15,392
$17,980
$16,750
$19,165
$25,221
$21,546
$24,692
$22,890
$25,677
$34,407
$23,006
$25,663
$23,774
$26,169
$35,837
$17,863
$20,329
$18,242
$20,581
$28,583
$5,646
$6,271
$5,764
$6,406
$7,250
$8,647
$10,687
$9,313
$11,000
$13,186
$11,300
$14,568
$13,411
$16,065
$20,615
$16,842
$20,305
$18,447
$21,125
$30,326
$19,393
$22,276
$20,265
$22,790
$32,477
$14,980
$17,618
$15,393
$17,855
$26,142
Native American
Mean
Median
APPENDIX
Page 16

labor market consequences of race differences in health

Transcription

Similar documents

2015

Unlimited 1% earnings on pUrchases. impressive

February - Contra Costa Health Services

"If these findings are correct, they suggest that it is the white students

$X,XXX.XX*

Harvest Newsletter Fall 2008

UNTIL MONDAY Haka`s throat gesture `has no place in sport`

Families and Social Class - Sociology at the University of

TOPICS No 27 - Trevallyn Primary School

Census areas