[Paleopsych] JLE: Cariero, Heckman, and Masterov: Labor Market Discrimination and Racial Differences in Premarket Factors
checker at panix.com
Fri May 13 18:28:46 UTC 2005
Labor Market Discrimination and Racial Differences in Premarket Factors
The Journal of Law and Economics, vol. XLVIII (April 2005)
University College London
JAMES J. HECKMAN
University of Chicago
DIMITRIY V. MASTEROV
University of Chicago
This paper investigates the relative significance of differences in
cognitive skills and discrimination in explaining racial/ethnic wage gaps.
We show that cognitive test scores for exams taken prior to entering the
labor market are influenced by schooling. Adjusting the scores for
racial/ethnic differences in education at the time the test is taken
reduces their role in accounting for the wage gaps. We also consider
evidence on parental and child expectations about education and on
stereotype threat effects. We find both factors to be implausible
alternative explanations for the gaps we observe. We argue that policies
need to address the sources of early skill gaps and to seek to influence
the more malleable behavioral abilities in addition to their cognitive
counterparts. Such policies are far more likely to be effective in
promoting racial and ethnic equality for most groups than are additional
civil rights and affirmative action policies targeted at the workplace.
* This research was supported by a grant from the American Bar
Foundation and National Institutes of Health grant R01-HD043411. We thank
an anonymous referee for helpful comments. Carneiro was supported by
Fundação Ciência e Tecnologia and Fundação Calouste Gulbenkian. We thank
Derek Neal and Rodrigo Soares for helpful comments and Maria Isabel
Larenas, Maria Victoria Rodriguez, and Xing Zhong for excellent research
IT is well documented that civil rights policy directed toward the
South raised black economic status in the 1960s and 1970s.1 Yet
substantial gaps remain in the market wages of African-American males and
females compared with those of white males and females.2 There are sizable
wage gaps for Hispanics as well.
Columns 1 of Table 1 report, for various ages, the mean hourly log
wage gaps for a cohort of young black and Hispanic males and females.3 The
reported gaps are not adjusted for differences in schooling, ability, or
other market productivity traits. The table shows that, on average, black
males earned wages that were 25 percent lower than those of white males in
1990. Hispanic males earned wages that were 17.4 percent lower in the same
year. The gaps increase for males as the cohort ages. For women, there are
smaller gaps for blacks and virtually no gap at all for Hispanics, and the
gaps for women show no clear trend with age.4 Joseph Altonji and Rebecca
Blank5 report similar patterns using data from the March supplement to the
Current Population Survey for 196898.
TABLE 1 Change in the Black-White Log Wage Gap Induced by
Controlling for Age-Corrected AFQT Scores, 19902000
These gaps are consistent with claims of pervasive labor market
discrimination against minorities. Minority workers with the same ability
and training as white workers may be receiving lower wages. There is,
however, another equally plausible explanation consistent with the same
evidence. Minorities may bring less skill and ability to the market.
Although there may be discrimination or disparity in the development of
these valuable skills, the skills may be rewarded equally across all
demographic groups in the labor market. Clearly, a variety of intermediate
explanations that combine both hypotheses are consistent with the data
The two polar interpretations of market wage gaps have profoundly
different policy implications. If persons of identical skill are treated
differently in the labor market on the basis of race or ethnicity, a more
vigorous enforcement of civil rights and affirmative action in the
marketplace would appear to be warranted. On the other hand, if the gaps
are solely due to unmeasured abilities and skills that people bring to the
labor market, then a redirection of policy toward fostering skills should
be emphasized as opposed to a policy of ferreting out discrimination in
Derek Neal and William Johnson6 shed light on the relative empirical
importance of market discrimination and skill disparity in accounting for
wage gaps by race. Controlling for scholastic ability measured in the
mid-teenage years, they substantially reduce but do not fully eliminate
wage gaps for black males in 199091 data. They more than eliminate the
gaps for black females. Columns 2 in Table 1 show our version of the
estimates reported in the Neal-Johnson study, expanded to cover additional
years.7 For black males, controlling for an early measure of ability cuts
the black-white wage gap in 1990 by 76 percent. For Hispanic males,
controlling for ability essentially eliminates the wage gap with whites.
For women, the results are even more striking. Wage gaps are actually
reversed, and controlling for ability produces higher wages for minority
females. This evidence suggests that the endowments people bring to the
labor market play a substantial role in accounting for minority wage gaps.
This paper critically examines the Neal-Johnson argument and brings
fresh evidence to bear on it. With some important qualifications, our
analysis supports their conclusion that factors determined outside of the
market play the major role in accounting for minority-majority wage
differentials in modern labor markets.
In producing the wage gaps shown in Table 1, we follow a practice
suggested by Neal and Johnson and do not adjust for the effects of racial
and economic differences in schooling, occupational choice, or work
experience on wages. Racial and ethnic differences in these factors may
reflect responses to labor market discrimination and should not be
controlled for in regressions that estimate the "full effect" of race on
wages through all channels since doing so may spuriously reduce estimated
wage gaps by introducing a proxy for discrimination into the control
variables. While the motivation for their procedure is clear, their
qualitative claim is false. Including schooling in a wage regression
raises the estimated wage gaps and produces more evidence of racial
disparity. Gaps when schooling is fixed and not fixed are both of interest
and answer different questions.
Gaps in measured ability by ethnicity and race are substantial.
Figure 1 plots the ability distribution as measured by age-corrected Armed
Forces Qualification Test (AFQT) scores8 for males and females in the
National Longitudinal Survey of Youth of 1979 (NLSY79).9 As noted by
Richard Herrnstein and Charles Murray,10 ability gaps are a major factor
in accounting for a variety of racial and ethnic disparities in
socioeconomic outcomes. Stephen Cameron and James Heckman11 show that
controlling for ability, blacks and Hispanics are more likely to enter
college than are whites.12
(22 kB) FIGURE 1.Density of age-corrected AFQT scores for NLSY79
males born after 1961
Neal and Johnson13 argue that ability measured in the teenage years
is a "premarket" factor, meaning that it is not affected by expectations
or actual experiences of discrimination in the labor market. They offer no
explicit criterion for determining which factors are premarket and which
Schooling affects test scores,14 and levels of minority schooling are
lower than white schooling levels, both generally and in the samples used
by Neal and Johnson. Their test score is contaminated by schooling
attainment at the date of the test. When their test scores are adjusted
for this factor, adjusted wage gaps increase.
The gaps in ability evident in Figure 1 stem in part from lower
levels of schooling by minorities at the time of the test and may also
arise from lowered academic effort in anticipation of future
discrimination in the labor market. If skills are not rewarded fairly, the
incentive to acquire them is diminished for those subject to prejudicial
treatment. Discrimination in the labor market might sap the incentives of
children and young adults to acquire skills and abilities but may also
influence the efforts they exert in raising their own offspring. This
means that even after adjusting their test scores for schooling, measured
ability may not be a true premarket factor. Neal and Johnson15 mention
this qualification in their original paper, and their critics have
subsequently reiterated it.
The gaps in ability may also be a consequence of adverse
environments. Even if all wage gaps are due to ability, uncontaminated by
expectations of market discrimination, the appropriate policy for
eliminating ability gaps is not apparent from Table 1. Should policies
focus on early ages through enriched Head Start programs or on improving
schooling quality and reducing school dropout and repetition rates that
plague minority children at later ages?
This paper demonstrates that ability gaps open up very early.
Minorities enter school with substantially lower measured ability than
whites. The black-white ability gap widens as the children get older and
obtain more schooling, but the contribution of formal education to the
widening of the gap is small when compared to the size of the initial gap.
There is a much smaller widening of the Hispanic-white ability gap with
Our evidence and that of James Heckman, Maria Isabel Larenas, and
Sergio Urzua16 suggest that school-based policies are unlikely to have
substantial effects on eliminating minority ability gaps. Factors that
operate early in the life cycle of the child are likely to have the
greatest impact on ability.
The early emergence of ability gaps indicates that child expectations
play only a limited role in accounting for such gaps since very young
children are unlikely to have formed expectations about labor market
discrimination and to make decisions based on those expectations. However,
parental expectations of future discrimination may still play a role in
shaping children's outcomes.
The early emergence of measured ability differentials also casts
doubt on the empirical importance of the "stereotype threat"17 as a major
factor contributing to black-white test score differentials. The
literature on this topic claims that black college students at selective
colleges perform worse on tests when they are told that the tests may be
used to confirm stereotypes about black-white ability differentials. The
empirical importance of this effect is in dispute in the psychology
The children in our data are tested at very young ages and are
unlikely to be aware of stereotypes about minority inferiority or be
affected by the stereotype threat that has been empirically established
only for students at elite colleges. In addition, large gaps in test
scores are also evident for Hispanics, a group for whom the stereotype
threat has not been documented. The stereotype threat literature claims
that measured test scores for minorities understate their true ability.
Unless the effect is uniform across ability levels, incremental ability
should be rewarded differently between blacks and whites. We find no
evidence of such an effect.
Adjusting for the schooling attainment of minorities at the time that
they take tests provides an empirically important qualification to the
Neal-Johnson study.19 An extra year of schooling has a greater impact on
test scores for whites and Hispanics than for blacks. Adjusting the test
score for schooling disparity at the date of the test leaves more room for
interpreting wage gaps as arising from labor market discrimination.
This finding does not necessarily overturn the conclusions of the
Neal-Johnson analysis. At issue is the source of the gap in schooling
attainment at the date of the test. The Neal-Johnson premarket factors are
a composite of ability and schooling and are likely to reflect both the
life cycle experiences and the expectations of the child. To the extent
that they reflect expectations of discrimination as embodied in schooling
that affects test scores, the scores are contaminated by market
discrimination and are not truly premarket factors. An open question is
how much of the gap in schooling is due to expectations about future
The evidence from data on parents' and children's expectations tells
a mixed story. Minority child and parent expectations about the children's
schooling prospects are as optimistic at ages 1617 as those of their white
counterparts, although actual schooling outcomes of whites and minorities
are dramatically different. Differential expectations at these ages cannot
explain the gaps in ability evident in Figure 1.
For children 14 and younger, parent and child expectations about
schooling are much lower for blacks than for whites, although only
slightly lower for Hispanics than for whites. All groups are still rather
optimistic in light of subsequent schooling attendance and performance. At
these ages, differences in expectations across groups may lead to
differential investments in skill formation. While lower expectations may
be a consequence of perceived labor market discrimination, they may also
reflect child and parental perception of the lower endowments possessed by
minorities, so this evidence is not decisive.
A focus on cognitive skill gaps, while traditional,20 misses
important noncognitive components of social and economic success. We show
that noncognitive (behavioral) gaps also open up early. Previous work
shows that they play an important role in accounting for market wages.
Policies that focus solely on improving cognitive skills miss an important
and promising determinant of socioeconomic success and disparity that can
be affected by policy.21
The rest of the paper proceeds in the following way. Section II
presents evidence on the evolution of test score gaps over the life cycle
of the child. Section III discusses the evidence on stereotype threat.
Section IV presents our evidence on how adjusting for schooling at the
date of the test affects the conclusions of the Neal-Johnson analysis and
how schooling affects test scores differentially for minorities. Section V
discusses our evidence on child and parental expectations. Section VI
presents evidence on noncognitive skills that parallels the analysis of
Section II. Section VII concludes.
1 John J. Donohue & James J. Heckman, Continuous versus Episodic
Change: The Impact of Civil Rights Policy on the Economic Status of
Blacks, 29 J. Econ. Literature 1603 (1991).
2 The literature on African-American economic progress in the
twentieth century is surveyed in James J. Heckman & Petra Todd,
Understanding the Contribution of Legislation, Social Activism, Markets
and Choice to the Economic Progress of African Americans in the Twentieth
Century (unpublished manuscript, Am. Bar Found. 2001).
3 These gaps are for a cohort of young persons aged 2628 in 1990 from
the National Longitudinal Survey of Youth of 1979 (NLSY79). They are
followed for 10 years until they reach ages 3638 in 2000.
4 However, the magnitudes (but not the direction) of the female gaps
are less reliably determined, at least for black women. Derek Neal, The
Measured Black-White Wage Gap among Women Is Too Small, 112 J. Pol. Econ.
S1 (2004), shows that racial wage gaps for black women are underestimated
by these types of regressions since they do not control for selective
labor force participation. This same line of reasoning is likely to hold
for Hispanic women.
5 Joseph Altonji & Rebecca Blank, Gender and Race in the Labor
Markets, in 3C Handbook of Labor Economics 3143 (Orley Ashenfelter & David
Card eds. 1999).
6 Derek Neal & William Johnson, The Role of Premarket Factors in
Black-White Wage Differences, 104 J. Pol. Econ. 869 (1996).
7 We use a sample very similar to the one used in their study. It
includes individuals born only in 196264. This exclusion is designed to
alleviate the effects of differential schooling at the test date on test
performance and to ensure that the AFQT is taken before the individuals
enter the labor market, so that it is more likely to be a premarket
8 Age-corrected AFQT is the standardized residual from the regression
of the AFQT score on age at the time of the test dummy variables. AFQT is
a subset of four out of 10 Armed Services Vocational Aptitude Battery
(ASVAB) tests used by the military for enlistment screening and job
assignment. It is the summed score from the word knowledge, paragraph
comprehension, mathematics knowledge, and arithmetic reasoning ASVAB
9 In our Web appendix
show that the same patterns emerge when we divide the sample by gender.
10 Richard Herrnstein & Charles Murray, The Bell Curve (1994).
11 Stephen V. Cameron & James J. Heckman, The Dynamics of Educational
Attainment for Black, Hispanic, and White Males, 109 J. Pol. Econ. 455
12 Sergio Urzua, The Educational White-Black Gap: Evidence on Years
of Schooling (Working paper, Univ. Chicago, Dep't Econ. 2003), shows that
this effect arises from greater minority enrollment in 2-year colleges.
Controlling for ability, whites are more likely to attend and graduate
from 4-year colleges. Using the Current Population Survey, Sandra E. Black
& Amir Sufi, Who Goes to College? Differential Enrollment by Race and
Family Background (Working Paper No. w9310, Nat'l Bur. Econ. Res. 2002),
finds that equating the family background of blacks and whites eliminates
the black-white gap in schooling only at the bottom of the family
background distribution. Furthermore, the gaps are eliminated in the 1980s
but not in the 1990s.
13 Neal & Johnson, supra note 6.
14 See Karsten Hansen, James J. Heckman, & Kathleen Mullen, The
Effect of Schooling and Ability on Achievement Test Scores, 121 J.
Econometrics 39 (2004).
15 Neal & Johnson, supra note 6.
16 James Heckman, Maria Isabel Larenas, & Sergio Urzua, Accounting
for the Effect of Schooling and Abilities in the Analysis of Racial and
Ethnic Disparities in Achievement Test Scores (Working paper, Univ.
Chicago, Dep't Econ. 2004).
17 See Claude Steele & Joshua Aronson, Stereotype Threat and the Test
Performance of Academically Successful African Americans, in The
Black-White Test Score Gap 401 (Christopher Jencks & Meredith Phillips
18 See Paul Sackett, Chaitra Hardison, & Michael Cullen, On
Interpreting Stereotype Threat as Accounting for African AmericanWhite
Differences in Cognitive Tests, 59 Am. Psychologist 7 (2004).
19 Neal & Johnson, supra note 6.
20 See, for example, Christopher Jencks & Meredith Phillips, The
Black-White Test Score Gap (1998).
21 See Pedro Carneiro & James J. Heckman, Human Capital Policy, in
Inequality in America: What Role for Human Capital Policies? 77 (James
Heckman & Alan Krueger eds. 2003).
II. MINORITY-WHITE DIFFERENCES IN EARLY TEST SCORES AND EARLY ENVIRONMENTS
This section summarizes evidence from the literature and presents
original empirical work that demonstrates that minority-white cognitive
skill gaps emerge early and persist through childhood and the adolescent
years. Christopher Jencks and Meredith Phillips22 and Greg Duncan and
Jeanne Brooks-Gunn23 document that the black-white test score gap is large
for 3- and 4-year-old children. Using the Children of the NLSY79 (CNLSY)
survey, a sample of children of the mothers in the 1979 National
Longitudinal Survey of Youth data, a variety of studies show that even
after controlling for many variables such as individual, family, and
neighborhood characteristics, the black-white test score gap is still
sizable.24,25 These studies also document that there are large black-white
differences in family environments. Ronald Ferguson26 summarizes this
literature and presents evidence that black children come from much poorer
and less educated families than white children, and they are also more
likely to grow up in single-parent households. Studies summarized by
Ferguson27 find that the achievement gap is high even for blacks and
whites attending high-quality suburban schools.28 The common finding
across these studies is that the black-white gap in test scores is large
and that it persists even after one controls for family background
variables. Children of different racial and ethnic groups grow up in
strikingly different environments.29 Even after accounting for these
environmental factors in a correlational sense, substantial test score
gaps remain. Furthermore, these gaps tend to widen with age and schooling:
black children show slower measured ability growth with schooling or age
than do white children.
This paper presents additional evidence from the children of the
persons interviewed in the CNLSY. We have also examined the Early
Childhood Longitudinal Survey (ECLS) analyzed by Ferguson30 and Roland
Fryer and Steven Levitt31 as well as the Children of the Panel Study of
Income Dynamics (CPSID) and find similar patterns. We broaden previous
analyses to include Hispanic-white differentials. Figure 2 shows the
average percentile Peabody Individual Achievement Test (PIAT) Math32
scores for males in different age groups by race. (Results for females
show the same patterns and are available in our Web appendix.33 For
brevity, in this paper we focus only on the male results.) Racial and
ethnic test score gaps are found as early as ages 56 (the earliest ages at
which we can measure math scores in CNLSY data).34 On average, black 5-
and 6-year-old boys score almost 18 percentile points below white 5- and
6-year-old boys (that is, if the average white is at the 50th percentile
of the test score distribution, the average black is at the 32nd
percentile of this distribution). The gap is a bit smaller16 percentbut
still substantial for Hispanics. These findings are duplicated for many
other test scores and in other data sets and are not altered if we use
median test scores instead of means. Furthermore, as shown in Figure 3,
even when we use a test taken at younger ages, racial gaps in test scores
can be found at ages 12.35 In general, test score gaps emerge early and
persist through adulthood.
(21 kB) FIGURE 2.Percentile PIAT Math score by race and age group
for CNLSY79 males
(23 kB) FIGURE 3.Average percentile Parts of the Body Test score by
race and age for CNLSY79 males.
For brevity, we focus on means and medians in this paper. However,
Figures 1 and 4 illustrate that there is considerable overlap in the
distribution of test scores across groups in recent generations. Many
black and Hispanic children at ages 56 score higher on a math test than
the average white child. Statements that we make about medians or means do
not apply to all persons in these distributions.
(28 kB) FIGURE 4.Density of percentile PIAT Math scores at ages 56
for CNLSY79 males
Figure 2 shows that the black-white percentile PIAT Math score gap
widens with age. By ages 1314, the average black is ranked more than 22
percentiles below the average white. In fact, these gaps persist through
adulthood. At 1314, Hispanic boys are almost 16 points below the average
When blacks and Hispanics enter the labor market, on average they
have a much poorer set of cognitive skills than do whites. Thus, it is not
surprising that their average labor market outcomes are so much worse.
Furthermore, these skill gaps emerge very early in the life cycle,
persist, and, if anything, widen for some groups. Initial conditions
(early test scores) are very important since skill begets skill.36
The research surveyed by Pedro Carneiro and James Heckman37 suggests
that enhanced cognitive stimulation at early ages is likely to produce
lasting gains in achievement test scores in children from disadvantaged
environments. If the interventions are early enough, they also appear to
raise IQ scores, at least for girls.38 Home and family environments at
early ages, and even the mother's behavior during pregnancy, play crucial
roles in the child's development, and black children grow up in
environments that are significantly more disadvantaged than those of white
children. Figure 5 shows the distributions of long-term or "permanent"
family income for blacks, whites, and Hispanics.39 Minority children are
much more likely to grow up in low-income families than are white
children. In our Web appendix,40 we show that there are also large
differences in the level of education and cognitive ability (as measured
by the AFQT) of mothers in different ethnic and racial groups (see also
Figure 1). Maternal AFQT score is a major predictor of children's test
scores.41 Figure 6 documents that white mothers are much more likely to
read to their children at young ages than are minority mothers, and we
obtain similar results at other ages.42 Using this reading variable and
other variables in CNLSY such as the number of books, magazines, toys, and
musical recordings, family activities (eating together, outings), methods
of discipline and parenting, learning at home, television-watching habits,
parental expectations for the child (chores, time use), and home
cleanliness and safety, we can construct an index of cognitive and
emotional stimulationthe home score. This index is always higher for
whites than for minorities.43 The Web appendix also shows that blacks are
more likely than whites to grow up in single-parent homes. Hispanics are
less likely than blacks to grow up in a single-parent home, although they
are much more likely to do so than are whites.
(28 kB) FIGURE 5.Density of log permanent income for CNLSY79 males
(33 kB) FIGURE 6.How often mother reads to child at age 2 by race
and sex for CNLSY79 males and females. Each bar represents the number of
people who report falling in a particular reading frequency cell divided
by the total number of people in their race and sex group.
Even after controlling for numerous environmental and family
background factors, racial and ethnic test score gaps remain at ages 34
for most tests and for virtually all the tests at later ages. Figure 7
shows that, even after adjusting for measures of family background,44 the
black-white gap in percentile PIAT Math scores at ages 56 is almost 8
percentile points and at ages 1314 is close to 11 percentile points.
Hispanic-white differentials are reduced more by such adjustments, falling
to 7 points at ages 56 and to 4 points at ages 1314. For some tests,
differentials frequently are positive or statistically insignificant.45
Measured home and family environments play an important role in the
formation of these skills, although they are not the whole story.46
(28 kB) FIGURE 7.Adjusted percentile PIAT Math score by race and age
group for CNLSY79 males.
Early test scores for blacks and Hispanics are similar, although
Hispanics often perform slightly better. Figure 2 shows that for the PIAT
Math score, the Hispanic-black gap is about 2 percentile points.47 This is
much smaller than either the black-white or the Hispanic-white gap. For
the PIAT Math, the black-white gap widens dramatically, especially at
later ages, but the Hispanic-white gap does not change substantially with
age. For other tests, even when there is some widening of the
Hispanic-white gap with age, it tends to be smaller than the widening in
the black-white gap in test scores. In particular, when we look at the
AFQT scores displayed in Figure 1, which are measured using individuals at
ages 1623, Hispanics clearly have higher scores than do blacks. In
contrast, Figure 4 shows a strong similarity between the math scores of
blacks and Hispanics at ages 56, although there are other tests at which,
even at these early ages, Hispanics perform substantially better than
blacks. When we control for the effects of home and family environments on
test scores, the Hispanic-white test score gap either decreases or is
constant over time, while the black-white test score gap tends to widen
22 Jencks & Phillips, supra note 20.
23 Greg Duncan & Jeanne Brooks-Gunn, Consequences of Growing up Poor
24 In a similar study based on the Early Childhood Longitudinal
Survey (ECLS), Roland Fryer & Steven Levitt, Understanding the Black-White
Test Score Gap in the First Two Years of School, 86 Rev. Econ. Stat. 447
(2004), eliminates the black-white test score gap in math and reading for
children at the time they are entering kindergarten, although not in
subsequent years. However, the raw test score gaps at ages 34 are much
smaller in ECLS than in CNLSY and other data sets that have been used to
study this issue, so their results are anomalous in the context of a
25 For a description of CNLSY and NLSY79, see Bureau of Labor
Statistics, NLS Handbook, 2001 (2001).
26 Ronald Ferguson, Why America's Black-White School Achievement Gap
Persists (unpublished manuscript, Harvard Univ. 2002).
27 Ronald Ferguson, What Doesn't Meet the Eye: Understanding and
Addressing Racial Disparities in High Achieving Suburban Schools (Special
Ed., Policy Issues Rep. 2002).
28 This is commonly referred to as the "Shaker Heights study,"
although it analyzed many other similar neighborhoods.
29 See also the discussion in David J. Armor, Maximizing Intelligence
30 Ferguson, supra note 26.
31 Fryer & Levitt, supra note 24.
32 Peabody Individual Achievement Test in Mathematics (PIAT Math)
measures the child's attainment in mathematics as taught in mainstream
education. It consists of 84 multiple-choice questions of increasing
difficulty, beginning with recognizing numerals and progressing to
geometry and trigonometry. The percentile score was calculated separately
for each sex at each age.
33 Note 9 supra.
34 Instead of using raw scores or standardized scores, we choose to
use ranks, or percentiles, since test score scales have no intrinsic
meaning. Our results are not sensitive to this procedure.
35 This is not always the case for women, as shown in our Web
appendix (supra note 9). The Parts of the Body Test attempts to measure
the young child's receptive vocabulary knowledge of orally presented words
as a means of estimating intellectual development. The interviewer names
each of 10 body parts and asks the child to point to that part of the
body. The score is computed by summing the number of correct responses.
The percentile score was calculated separately for each sex at each age.
36 See James J. Heckman, Policies to Foster Human Capital, 54 Res.
Econ. 3 (2000).
37 Carneiro & Heckman, supra note 21.
38 See Frances Campbell et al., Early Childhood Education: Young
Adult Outcomes from the Abecedarian Project, 6 Applied Developmental Sci.
39 Values of permanent income are constructed by taking the average
of all nonmissing values of annual family income at ages 018 discounted to
child's age 0 using a 10 percent discount rate.
40 Note 9 supra.
41 For example, the correlation between percentile PIAT math score
and age-corrected maternal AFQT is .4.
42 See the results for all ages in our Web appendix, supra note 9.
43 As shown in our Web appendix (id.), where we document that both
cognitive and emotional stimulation indexes are always higher for whites
than for blacks at all ages.
44 Scores are adjusted by permanent family income, mother's
education, and age-corrected AFQT and home scores. "Adjusted" indicates
that we equalized the family background characteristics across all race
groups by setting them at the mean to purge the effect of disparities in
45 In our Web appendix (supra note 9), tables 1A and 1B report that
even after controlling for different measures of home environment and
child stimulation, the black-white test score gap persists, even though it
drops considerably. Results for other tests and other samples can be found
in our Web appendix. Even though for some test scores early black-white
test score gaps can be eliminated once we control for a large number of
characteristics, it is harder to eliminate them at later ages. In the
analysis presented here, the most important variable in reducing the test
score gap is the mother's cognitive ability, as measured by the AFQT.
46 However, the home score includes variables such as the number of
books, which are clearly choice variables and likely to cause problems in
this regression. The variables with the largest effect on the
minority-white test score gap are maternal AFQT and raw home score.
47 The test score is measured in percentile rank The black-white gap
is slightly below 18, while the Hispanic-white gap is slightly below 16.
This means that the black-Hispanic gap should be around 2.
III. THE STEREOTYPE THREAT
The fact that substantial racial and ethnic test score gaps open up
early in the life cycle of children casts doubt on the empirical
importance of the "stereotype threat." It is now fashionable in some
circles to attribute gaps in black test scores to racial consciousness on
the part of black test takers stemming from the way test scores are used
in public discourse to describe minorities.48 The claim is that blacks
perform below their true abilities on standardized tests when a stereotype
threat is present. The empirical importance of the stereotype threat in
accounting for test score differentials has been greatly overstated in the
popular literature.49 No serious empirical scholar assigns any
quantitative importance to stereotype threat effects as a major
determinant of test score gaps.
Stereotype threats could not have been important when blacks took the
first IQ tests at the beginning of the twentieth century, which documented
the racial differentials that gave rise to the stereotype. Yet racial IQ
gaps are comparable across time.50 Young children, like the ones studied
in this paper, are unlikely to have the heightened racial consciousness
about tests and their social significance of the sort claimed to be found
by Claude Steele and Joshua Aronson51 in college students at a few elite
universities. Moreover, sizable gaps are found for young Hispanic malesa
group for which the stereotype threat remains to be investigated.
Additional evidence on the unimportance of stereotype threat is
presented in Table 2.52 According to the stereotype threat literature,
minority test scores understate true ability. If stereotyping affects the
test score gap differently across ability levels, the effect of a unit of
ability on wages for a black should be different than it is for a white.
If the understatement is uniform across all ability levels, the
coefficient on a dummy variable for race is overstated in a log wage
regression (that is, measured discrimination is understated). If the
stereotype threat operates when minorities take the AFQT, their scores
should have a different incremental effect on wages than majority AFQT
scores.53 We test this hypothesis using the empirical model in Table 2. We
estimate the effect of black and Hispanic AFQT scores relative to the
effect of white AFQT scores on log wages as extracted from the NLSY79.
This amounts to testing for racial AFQT interactions in a log wage
equation. While there is some (weak) evidence that black scores have a
larger effect on log wages than white scores, the black-AFQT interaction
coefficients are small in magnitude and imprecisely determined. For
Hispanics, the estimated AFQT interaction coefficients are negative and,
again, not precisely determined. In our Web appendix, we also graph the
mean log wage by AFQT decile by race. There is no particular pattern of
convergence or divergence across ability levels when evaluated over common
TABLE 2 Pooled Log Wage Regressions for NLSY Males, 19902000
The stereotype literature substitutes wishful thinking for
substantial evidence. There is no evidence that it accounts for an
important fraction of minority-white test score gaps or that test scores
are not good measures of productivity.54
48 Steele & Aronson, supra note 17.
49 See the analysis in Sackett, Hardison, & Cullen, supra note 18.
50 Charles Murray, The Secular Increase in IQ and Longitudinal
Changes in the Magnitude of the Black-White Difference: Evidence from the
NLSY (paper presented at the Behavior Genetics Association Meeting,
Vancouver 1999), reviews the evidence on the evolution of the black-white
IQ gap. In the 1920sa time when such tests were much more unreliable and
black educational attainment much lowerthe mean black-white difference was
.86 standard deviations. The largest black-white difference appears in the
1960s, with a mean black-white difference of 1.28 standard deviations. The
difference ranges from a low of .82 standard deviations in the 1930s to
1.12 standard deviations in the 1970s. However, none of the samples prior
to 1960 are nationally representative, and the samples were often chosen
so as to effectively bias the black mean upward.
51 Steele & Aronson, supra note 17.
52 See our Web appendix (supra note 9) for evidence on females.
53 Let Y = 0 + 1T + , where E( T) = 0. The same equation governs
black and white outcomes. The term T is the true test score, and T* is the
test score under stereotype threat:
Suppose Cov(, U) = 0. Our Web appendix, supra note 9, shows that under
random sampling, the coefficient on the test score for whites is 1 and for
Intercepts are 0 and
where E(T) is the mean of T, is the variance of T, and is the variance
of U. Thus, the intercepts for blacks are upward biased. The slope for
blacks in general may be greater than or less than 1, depending on whether
the gap widens with T or shrinks . When = 0 (U = 0) and 1 = 1, the
slopes are the same for blacks and whites, but the intercepts for blacks
are upward biased. This method underestimates the amount of
54 A circular version of the stereotype threat argument would claim
that minorities also underperform at the workplace because of stereotype
threat there, so using measured wages to capture productivity understates
true black productivity. This form of the stereotype threat argument is
irrefutable. All measures are contaminated.
IV. THE DIFFERENTIAL EFFECT OF SCHOOLING ON TEST SCORES
We have established that cognitive test scores are correlated with
home and family environments and that test score gaps increase with age
and schooling. The research of Karsten Hansen, James Heckman, and Kathleen
Mullen55 and Heckman, Larenas, and Urzua56 shows that the AFQT scores used
by Neal and Johnson57 are affected by the schooling attainment of
individuals at the time they take the test. Therefore, one reason for the
divergence of black and white test scores over time may be differential
schooling attainments. Figure 8 shows the schooling completed at the test
date for the six demographic groups in the age ranges of the NLSY used by
Neal and Johnson. Blacks have completed (slightly) less schooling at the
test date than whites but substantially more than Hispanics.
(35 kB) FIGURE 8.Highest category of schooling completed at the test
date by race, sex, and age for NLSY79 males and females born after 1961.
Each bar represents the number of people who report falling in a
particular reading frequency cell divided by the total number of people in
their race and sex group.
Table 3 presents estimates of the effect of schooling at test date on
AFQT scores for individuals in different demographic groups in the NLSY,
using a version of the nonparametric method developed by Hansen, Heckman,
and Mullen.58 Their method isolates the causal effect of schooling
attained at the test date on test scores controlling for unobserved
factors that lead to selective differences in schooling attainment. This
table shows that the effect of schooling on test scores is much larger for
whites and Hispanics than it is for blacks over most ranges of schooling.
As a result, even though Hispanics have fewer years of completed schooling
than blacks at the time they take the AFQT, on average Hispanics score
better on the AFQT than do blacks.
TABLE 3 Effect of Years of Schooling on AFQT Scores for
Individuals in NLSY79
There are different explanations for these findings. Carneiro and
Heckman, Cunha and Heckman, and Cunha and colleagues59 suggest that one
important feature of the learning process is complementarity and
self-productivity between initial endowments of human capital and
subsequent learning.60 Higher levels of human capital raise the
productivity of learning.61 Since minorities and whites start school with
very different initial conditions, their learning paths can diverge
dramatically over time. A related explanation may be that blacks and
nonblacks learn at different rates because blacks attend lower quality
schools than whites.62
Janet Currie and Duncan Thomas63 show that test score gains of
participants in the Head Start program tend to fade completely for blacks
but not for whites. They suggest that one reason may be that blacks attend
worse schools than whites, and therefore blacks are not able to maintain
initial test score gains. Both early advantages and disadvantages as well
as school quality are likely to be important factors in the human capital
In light of the greater growth in test scores of Hispanics that is
parallel to that of whites, explanations based on schooling quality are
not entirely compelling. Hispanics start from similar initial
disadvantages in family environments and face school and neighborhood
environments similar to those faced by blacks.64 They also have early
levels of test scores similar to those found in the black population.65
To analyze the consequences of correcting for different levels of
schooling at the test date, we reanalyze the Neal-Johnson66 data using
AFQT scores corrected for the race- or ethnicity-specific effect of
schooling while equalizing the years of schooling attained at the date of
the test across all racial/ethnic groups. The results of this adjustment
are presented in Table 4. This adjustment is equivalent to replacing each
individual's AFQT score by the score we would measure if he or she would
have stopped his or her formal education after eighth grade.67 In other
words, we use eighth-grade-adjusted AFQT scores for everyone. Since the
effect of schooling on test scores is higher for whites than for blacks,
and whites have more schooling than blacks at the date of the test, this
adjustment reduces the test scores of whites much more than those for
blacks. The black-white male wage gap is cut only in half (as opposed to
76 percent) when we use this new measure of skill, and a substantial
unexplained residual remains. The adjustment has little effect on the
Hispanic-white wage gap, but a wage gap for black women emerges when using
the schooling-adjusted measure that did not appear in the original
TABLE 4 Change in the Black-White Log Wage Gap Induced by
Controlling for Schooling-Corrected AFQT Scores, 19902000
Adjusting for schooling at the date of the test reduces the test
score gap. This evidence raises the larger question of what a premarket
factor is. Neal and Johnson do not condition on schooling in explaining
black-white wage gaps, arguing that schooling is affected by expectations
of adverse market opportunities facing minorities and that conditioning on
such a contaminated variable would spuriously reduce the estimated wage
gap. We present direct evidence on this claim below.
Their reasoning is not entirely coherent. If expectations of
discrimination affect schooling, the very logic of their premarket
argument suggests that they should control for the impact of schooling on
test scores before using test scores to measure premarket factors. Neal
and Johnson68 assume that schooling at the time the test is taken is not
affected by expectations of discrimination in the market, while later
This distinction is arbitrary. A deeper investigation of the
expectation formation process and feedback is required. One practical
conclusion with important implications for interpretation of the evidence
is that the magnitude of the wage gap one can eliminate by performing a
Neal-Johnson analysis depends on the age at the time the test is taken. We
find that the earlier the test is taken, the smaller the unadjusted test
score gap, and the larger the fraction of the wage gap that is unexplained
by the residual. Figure 9 shows how adjusting measured ability for
schooling at the time of the test at different levels of attained
schooling affects the adjusted wage gap for black males. For example, the
log wage gap that we obtain when using eleventh-grade test scores
corresponds to that using an AFQT correction equal to 11. The later the
grade at which we adjust the test score, the lower the estimated gap. This
is because a test score gap opens up at later schooling levels, and hence
adjustment reduces the gap by a larger amount at later schooling levels.69
(30 kB) FIGURE 9.Residual black-white log wage gap in 1991 by grade
at which we evaluate schooling-corrected AFQT scores for NLSY79 males.
Finally we show that adjusting for "expectations-contaminated"
completed schooling by entering it as a direct regression in a log wage
equation does not operate in the fashion conjectured by Neal and Johnson.
Table 5 shows that when we adjust wage differences for completed schooling
as well as schooling-adjusted AFQT, wage gaps widen relative to the simple
adjustment. This runs contrary to the simple intuition that schooling
embodies expectations of market discrimination, so conditioning on it will
eliminate wage gaps.70 The deeper issue, not resolved in this paper or the
literature, is what productivity factors to condition on in measuring
discrimination. Schooling and measured ability are both valid candidate
productivity variables. Conditioning on them singly or jointly and
eliminating spurious endogeneity effects produces conceptually different
measures of the wage gap, all of which answer distinct but economically
interesting questions. Both variables may be affected by discrimination.
Looking only at outcome equations, one cannot settle what is a
productivity characteristic and what is contaminated and what is not.71,72
Deleting potential contaminated variables does not, in general, produce
the conceptually desired measure of discrimination.
TABLE 5 Change in the Black-White Log Wage Gap Induced by
Controlling for Schooling-Corrected AFQT Scores and Highest Grade
Ours is a worst-case analysis for the Neal-Johnson study.73 If we
assign all racial and ethnic schooling differences to expectations of
discrimination in the labor market, the results for blacks are less sharp
than Neal and Johnson claim. Yet even in the worst-case scenario,
adjusting for ability corrected for schooling and schooling as a direct
effect on wages substantially reduces minority-majority wage gaps over the
unadjusted case. The evidence presented in Section II about the early
emergence of ability differentials is reinforced by the early emergence of
differential grade repetition gaps for minorities documented by Cameron
and Heckman.74 Most of the schooling gap at the date of the test emerges
in the early years at ages when child expectations about future
discrimination are unlikely to be operative. One might argue that these
early schooling and ability gaps are due to parental expectations of poor
labor markets for minority children. We next examine data on child and
55 Hansen, Heckman, & Mullen, supra note 14.
56 Heckman, Larenas, & Urzua, supra note 16.
57 Neal & Johnson, supra note 6.
58 Hansen, Heckman, & Mullen, supra note 14. Heckman, Larenas, &
Urzua, supra note 16, presents a more refined analysis of the
racial/ethnic wage gap using the analysis of Hansen, Heckman, & Mullen
that supports all of our main conclusions. See also the note to Table 3.
59 Carneiro & Heckman, supra note 21; Flavio Cunha & James Heckman,
The Technology of Skill Formation (unpublished manuscript, Univ. Chicago
2004); Flavio Cunha et al., Interpreting the Evidence on Life Cycle Skill
Formation, in Handbook of Education Economics (Finis Welch & Eric Hanushek
eds., forthcoming 2005).
60 For example, see the model in Yoram Ben-Porath, The Production of
Human Capital and the Life Cycle of Earnings, 75 J. Pol. Econ. 352 (1967).
See also Cunha et al., supra note 59.
61 See the evidence in James Heckman, Lance Lochner, & Christopher
Taber, Explaining Rising Wage Inequality: Explorations with a Dynamic
General Equilibrium Model of Labor Earnings with Heterogeneous Agents, 1
Rev. Econ. Dynamics 1 (1998).
62 Cunha & Heckman, supra note 59, shows that complementarity implies
that early human capital increase the productivity of later investments in
human capital and that early investments that are not followed up by later
investments in human capital are not productive.
63 Janet Currie & Duncan Thomas, School Quality and the Longer-Term
Effects of Head Start, 35 J. Hum. Resources 755 (2000).
64 The evidence for CNLSY is presented in our Web appendix (supra
65 Heckman, Larenas, & Urzua, supra note 16, presents a more formal
analysis of the effect of schooling quality on test scores, showing that
schooling inputs explain little of the differential growth in test scores
among blacks, whites, and Hispanics.
66 Neal & Johnson, supra note 6.
67 However, the score is affected by attendance in kindergarten, 8
further years of schooling, and any school quality differentials in those
68 Neal & Johnson, supra note 6.
69 The figure omits the results for the 16-and-over category because
the low number of minorities makes correction of test scores to that level
much less reliable than correction to the other schooling levels. The
unadjusted line shows the black-white log wage gap we observe if we do not
depend on the grade to which we are correcting the test score. The
adjusted line shows the black-white log wage gap after we adjust for the
AFQT scores corrected to different grades. In our Web appendix (supra note
9), we present the same analysis for females and Hispanics.
70 The simple intuition, however, can easily be shown to be wrong, so
the evidence in these tables is not decisive on the presence of
discrimination in the labor market. The basic idea is that if both
schooling and the test score are correlated with an unmeasured
discrimination component in the error term, the bias for the race dummy
may be either positive or negative depending on the strength of the
correlation among the contaminated variables and their correlation with
the error term. See the discussion in our Web appendix (id.), where we
show that if both schooling and test score are correlated with factors
leading to discrimination in earnings, the estimated discrimination effect
may be upward or downward biased by adding schooling as a regressor.
71 See Robert Bornholz & James J. Heckman, Measuring Disparate
Impacts and Extending Disparate Impact Doctrine to Organ Transplantation,
48 Persp. Biology & Med. S95 (2005).
72 As pointed out to us by an anonymous referee, another reason for
excluding years of schooling from the log wage equation is that schooling
overstates the amount of human capital black children receive relative to
white children, say because of differential schooling quality. If this
effect is strong enough, including years of schooling will overstate the
racial wage differential. Table 3 shows that years of schooling for black
children have less effect on human capital (the test score) than years of
schooling for white children. However, Heckman, Larenas, & Urzua, supra
note 16, shows that measured schooling quality accounts for little of the
gap or the growth in the gap between blacks and whites.
73 Neal & Johnson, supra note 6.
74 Cameron & Heckman, supra note 11.
V. THE ROLE OF EXPECTATIONS
The argument that minority children perform worse on tests because
they expect to be less well rewarded in the labor market than whites for
the same test score or schooling level is implausible because expectations
of labor market rewards are unlikely to affect the behavior of children as
early as ages 3 or 4, when test score gaps are substantial across
different ethnic and racial groups. The argument that minorities invest
less in skills because both minority children and minority parents have
low expectations about their performance in school and in the labor market
has mixed empirical backing.
Data on expectations are hard to find, and when they are available
they are often difficult to interpret. For example, in the NLSY97, black
17- and 18-year-olds report that the probability of dying next year is 22
percent, while whites report a probability of dying of 16 percent.75 Both
numbers are absurdly high. Minorities usually report higher expectations
than whites of committing a crime, being incarcerated, and being dead next
year, and these adverse expectations may reduce their investment in human
capital. Expectations reported by parents and children for the adolescent
years for a variety of outcomes are given in our Web appendix.76
Schooling expectations measured in the late teenage years are very
similar for minorities and whites. They are slightly lower for Hispanics.
Table 6 reports the mean expected probability of being enrolled in school
next year for black, white, and Hispanic 17- and 18-year-old males. Among
those individuals enrolled in 1997, on average whites expect to be
enrolled next year with 95.7 percent probability. Blacks expect that they
will be enrolled next year with a 93.6 percent probability. Hispanics
expect to be enrolled with a 91.5 percent probability. If expectations
about the labor market are adverse for minorities, they should translate
into adverse expectations for the child's education. Yet these data do not
reveal this. Moreover, all groups substantially overestimate actual
enrollment probabilities. The difference in expectations between blacks
and whites is very small and is less than half the difference in actual
(realized) enrollment probabilities (81.9 percent for whites versus 76.4
percent for blacks). The gap is wider for Hispanics. Table 7 reports
parental schooling expectations for white, black, and Hispanic males for
the same individuals used to compute the numbers in Table 6. It shows
that, conditional on being enrolled in 1997 (the year the expectation
question is asked), black parents expect their sons to be enrolled next
year with a 90.9 percent probability, while for whites this expectation is
95.4 percent. For Hispanics, this number is lower (88.5 percent) but still
substantial. Parents overestimate enrollment probabilities for their sons,
but black parents have lower expectations than white parents. For females,
the racial and ethnic differences in parental expectations are smaller
than those for males.77
TABLE 6 Juvenile Expectations about School Enrollment in 1998:
TABLE 7 Parental Expectations about Youth School Enrollment in
1998: NLSY79 Males
For expectations measured at earlier ages the story is dramatically
different. Figures 10 and 11 show that, for the CNLSY group, both black
and Hispanic children and their parents have more pessimistic expectations
about schooling than do white children, and more pessimistic expectations
may lead to lower investments in skills, less effort in schooling, and
lower levels of ability. These patterns are also found in the CPSID and
(33 kB) FIGURE 10.Child's own expected educational level at age 10
by race and sex for CNLSY79 males and females. Each bar represents the
number of people who report falling in a particular educational level cell
divided by the total number of people in their race and sex group.
(34 kB) FIGURE 11.Mother's expected educational level for the child
at age 6 by race and sex for CNLSY79 males and females. Each bar
represents the number of people who report falling in a particular
educational level divided by the total number of people in their race and
If the more pessimistic expectations of minorities are a result of
perceived market discrimination, then lower levels of investment in
children that translate into lower levels of ability and skill at later
ages are attributable to market discrimination. Ability would not be a
premarket factor. However, lower expectations for minorities may not be a
result of discrimination but just a rational response to the fact that
minorities do not do as well in school as whites. This may be due to
environmental factors unrelated to expectations of discrimination in the
labor market. Whether this phenomenon itself is a result of discrimination
is an open question. Expectation formation models are very complex and
often lead to multiple equilibria and therefore are difficult to test
empirically. However, the evidence reported here does not provide much
support for the claim that the ability measure used by Neal and Johnson79
is substantially contaminated by expectational effects.
75 See our Web appendix, table 3, for evidence on expectations from
NLSY97 (supra note 9).
78 For CNLSY teenagers, expectations across racial groups seem to
converge at later ages. See our Web appendix (id.).
79 Neal & Johnson, supra note 6.
VI. THE EVIDENCE ON NONCOGNITIVE SKILLS
Controlling for scholastic ability in accounting for
minority-majority wage gaps captures only part of the endowment
differences between groups but receives most of the emphasis in the
literature on black-white gaps in wages. An emerging body of evidence,
summarized by Samuel Bowles, Herbert Gintis, and Melissa Osborne,80
Carneiro and Heckman,81 and Heckman, Stixrud, and Urzua,82 documents that
noncognitive skillsmotivation, self-control, time preference, and social
skillsare important in explaining socioeconomic success.83
The CNLSY has life cycle measures of noncognitive skills. Mothers are
asked age-specific questions about the antisocial behavior of their
children such as aggressiveness or violent behavior, cheating or lying,
disobedience, peer conflicts, and social withdrawal. The answers to these
questions are grouped in different indices.84 Figure 12 shows that there
are important racial and ethnic gaps in the antisocial behavior index that
emerge in early childhood. The higher the score, the worse the behavior.
By ages 56, the average black is roughly 10 percentile points above the
average white in the distribution of this score.85 The results shown in
Figure 13, where we adjust the gaps by permanent family income, mother's
education, and age-corrected AFQT and home scores, also show large
(26 kB) FIGURE 12.Average percentile antisocial behavior score by
race and age group for CNLSY79 males.
(30 kB) FIGURE 13.Adjusted percentile antisocial behavior score by
race and age group for CNLSY79 males.
Section II documents that minority and white children face
substantial differences in family and home environments while growing up.
The evidence presented in this section shows that these early
environmental differences account (in a correlational sense) for most of
the minority-white gap in noncognitive skills, as measured in the CNLSY.
Carneiro and Heckman87 document that noncognitive skills are more
malleable than cognitive skills and are more easily shaped by
interventions. More motivated children achieve more and have higher
measured achievement test scores than less motivated children of the same
ability. The largest effects of interventions in childhood and adolescence
are on noncognitive skills that promote learning and integration into the
larger society. Improvements in these skills produce better labor market
outcomes and less engagement in criminal activities and other risky
behavior. Promotion of noncognitive skill is an avenue for policy that
warrants much greater attention.
80 Samuel Bowles, Herbert Gintis, & Melissa Osborne, The Determinants
of Earnings: A Behavioral Approach, 39 J. Econ. Literature 1137 (2001).
81 Carneiro & Heckman, supra note 21.
82 James Heckman, Jora Stixrud, & Sergio Urzua, Evidence on the
Importance of Cognitive and Noncognitive Skills on Social and Economic
Outcomes (unpublished manuscript, Univ. Chicago 2004).
83 Some of the best evidence for the importance of noncognitive
skills in the labor market is from the General Education Development (GED)
program. This program examines high school dropouts to certify that they
are equivalent to high school graduates. In its own terms, the GED program
is successful. James J. Heckman & Yona Rubinstein, The Importance of
Noncognitive Skills: Lessons from the GED Testing Program, 91 Am. Econ.
Rev. 145 (2001), shows that GED recipients and ordinary high school
graduates who do not go on to college have the same distribution of AFQT
scores (the test graphed in Figure 1). Yet GED recipients earn the wages
of high school dropouts with the same number of years of completed
schooling. They are more likely to quit their jobs, engage in fighting or
petty crime, or be discharged from the military than are high school
graduates who do not go on to college or other high school dropouts.
Intelligence alone is not sufficient for socioeconomic success.
Minority-white gaps in noncognitive skills open up early and widen over
the life cycle.
84 The children's mothers were asked 28 age-specific questions about
frequency, range, and type of specific behavior problems that children
ages 4 and over may have exhibited in the previous 3 months. Factor
analysis was used to determine six clusters of questions. The responses
for each cluster were then dichotomized and summed to produce a raw score.
The percentile score was then calculated separately for each sex at each
age from the raw score. A higher percentile score indicated a higher
incidence of problems. The antisocial behavior index we use in this paper
consists of measures of cheating and telling lies, bullying and cruelty to
others, not feeling sorry for misbehaving, breaking things deliberately
(if age is less than 12), disobedience at school (if age is greater than
5), and trouble getting along with teachers (if age is greater than 5).
85 In our Web appendix (supra note 9), we show that these differences
are statistically strong. Once we control for family and home
environments, gaps in most behavioral indices disappear.
86 See our Web appendix, tables 2A and 2B, for the effect of
adjusting for other environmental characteristics on the antisocial
behavior score (id.).
87 Carneiro & Heckman, supra note 21.
VII. SUMMARY AND CONCLUSION
This paper discusses the sources of wage gaps between minorities and
whites. For all minorities but black males, adjusting for the ability that
minorities bring to the market eliminates wage gaps. The major source of
economic disparity by race and ethnicity in U.S. labor markets is in
endowments, not in payments to endowments.
This evidence suggests that strengthened civil rights and affirmative
action policies targeted at the labor market are unlikely to have much
effect on racial and ethnic wage gaps, except possibly for those
specifically targeted toward black males.88 Policies that foster
endowments have much greater promise. On the other hand, this paper does
not provide any empirical evidence on whether the existing edifice of
civil rights and affirmative action legislation should be abolished. All
of our evidence on wages is for an environment in which affirmative action
laws and regulations are in place.
Minority deficits in cognitive and noncognitive skills emerge early
and then widen. Unequal schooling, neighborhoods, and peers may account
for this differential growth in skills, but the main story in the data is
not about growth rates but rather about the size of early deficits.
Hispanic children start with cognitive and noncognitive deficits similar
to those of black children. They also grow up in similarly disadvantaged
environments and are likely to attend schools of similar quality.
Hispanics complete much less schooling than blacks. Nevertheless, the
ability growth by years of schooling is much higher for Hispanics than for
blacks. By the time they reach adulthood, Hispanics have significantly
higher test scores than do blacks. Conditional on test scores, there is no
evidence of an important Hispanic-white wage gap. Our analysis of the
Hispanic data illuminates the traditional study of black-white differences
and casts doubt on many conventional explanations of these differences
since they do not apply to Hispanics, who also suffer from many of the
same disadvantages. The failure of the Hispanic-white gap to widen with
schooling or age casts doubt on the claim that poor schools and bad
neighborhoods are the reasons for the slow growth rate of black test
scores. Deficits in noncognitive skills can be explained (in a statistical
sense) by adverse early environments; deficits in cognitive skills are
less easily eliminated by the same factors.
We have reexamined the Neal-Johnson89 analysis that endowments
acquired before people enter the labor market explain most of the
minority-majority wage gap. Neal and Johnson use an ability test taken in
the teenage years as a measure of endowment unaffected by discrimination.
They omit schooling in adjusting for racial and ethnic wage gaps, arguing
that schooling choices are potentially contaminated by expectations of
labor market discrimination. Yet they do not adjust their measure of
ability by the schooling attained at the date of the test, which would be
the appropriate correction if their argument were correct.
Adjusting wage gaps by both completed schooling and the
schooling-adjusted test widens the wage gaps for all groups. This
adjustment effect is especially strong for blacks. Nonetheless, half of
the black-white male wage gap is still explained by the adjusted score. At
issue is how much of the majority-minority difference in schooling at the
date of the test is due to expectations of labor market discrimination and
how much is due to adverse early environments. While this paper does not
settle this question definitively, test score gaps emerge early and are
more plausibly linked to adverse early environments. The lion's share of
the ability gaps at the date of the test emerge very early, before
children can have clear expectations about their labor market prospects.
The analysis of Sackett, Hardison, and Cullen90 and the emergence of
test score gaps in young children cast serious doubt on the importance of
stereotype threats in accounting for poorer black test scores. It is
implausible that young minority test takers have the social consciousness
assumed in the stereotype literature. If true, black skills are
understated by the tests, and the market return to ability should be
different for blacks than for whites. We find no evidence of such an
Gaps in test scores of the magnitude found in recent studies were
found in the earliest tests developed at the beginning of the twentieth
century, before the results of testing were disseminated and a stereotype
threat could have been "in the air." The recent emphasis on the stereotype
threat as a basis for black-white test score differences ignores the
evidence that tests are predictive of schooling attainment and market
wages. It diverts attention away from the emergence of important skill
gaps at early ages, which should be a target of public policy.
Effective social policy designed to eliminate racial and ethnic
inequality for most minorities should focus on eliminating skill gaps, not
on discrimination in the workplace of the early twenty-first century.
Interventions targeted at adults are much less effective and do not
compensate for early deficits. Early interventions aimed at young children
hold much greater promise than strengthened legal activism in the
88 However, even for black males, a substantial fraction of the
racial wage gap can be attributed to differences in skill.
89 Neal & Johnson, supra note 6.
90 Sackett, Hardison, & Cullen, supra note 18.
a.. Altonji, Joseph, and Blank, Rebecca. "Gender and Race in the Labor
Market." In Handbook of Labor Economics, edited by Orley Ashenfelter and
David E. Card, pp. 3C:31434259. New York: Elsevier Science Press, 1999.
b.. Armor, David J. Maximizing Intelligence. New Brunswick, N.J.:
Transaction Publishers, 2003.
c.. Ben-Porath, Yoram. "The Production of Human Capital and the Life
Cycle of Earnings." Journal of Political Economy 75 (1967): 35265.
d.. Black, Sandra E., and Sufi, Amir. "Who Goes to College? Differential
Enrollment by Race and Family Background." Working Paper No. w9310.
Cambridge, Mass.: National Bureau of Economic Research, 2002.
e.. Bornholz, Robert, and Heckman, James J. "Measuring Disparate Impacts
and Extending Disparate-Impact Doctrine to Organ Transplantation."
Perspectives in Biology and Medicine 48 (2005): S95S122.
f.. Bowles, Samuel; Gintis, Herbert; and Osborne, Melissa. "The
Determinants of Earnings: A Behavioral Approach." Journal of Economic
Literature 39 (2001): 113776.
g.. Bureau of Labor Statistics. NLS Handbook 2001. Washington, D.C.:
U.S. Department of Labor, 2001.
h.. Cameron, Stephen V., and Heckman, James J. "The Dynamics of
Educational Attainment for Black, Hispanic, and White Males." Journal of
Political Economy 109 (2001): 45599.
i.. Campbell, Frances; Ramey, Craig; Pungello, Elizabeth; Sparling,
Joseph; and Miller-Johnson, Shari. "Early Childhood Education: Young Adult
Outcomes from the Abecedarian Project." Applied Developmental Science 6
j.. Carneiro, Pedro, and Heckman, James J. "Human Capital Policy." In
Inequality in America: What Role for Human Capital Policies? edited by
James J. Heckman and Alan B. Krueger, pp. 77240. Cambridge, Mass.: MIT
k.. Cunha, Flavio, and Heckman, James J. "The Technology of Skill
Formation." Paper presented at the Society of Economic Dynamics and
Control, Florence, 2004, and at the American Economic Association annual
meeting, San Diego, 2004.
l.. Cunha, Flavio; Heckman, James J.; Lochner, Lance; and Masterov,
Dimitriy V. "Interpreting the Evidence on Life Cycle Skill Formation." In
Handbook of Education Economics, edited by Finis Welch and Eric Hanushek.
New York: North Holland, forthcoming 2005.
m.. Currie, Janet, and Thomas, Duncan. "School Quality and the
Longer-Term Effects of Head Start." Journal of Human Resources 35 (2000):
n.. Donohue, John J., and Heckman, James J. "Continuous versus Episodic
Change: The Impact of Civil Rights Policy on the Economic Status of
Blacks." Journal of Economic Literature 29 (1991): 160343.
o.. Duncan, Greg, and Brooks-Gunn, Jeanne. Consequences of Growing up
Poor. New York: Russell Sage, 1997.
p.. Ferguson, Ronald. "What Doesn't Meet the Eye: Understanding and
Addressing Racial Disparities in High Achieving Suburban Schools." Special
Edition, Policy Issues Report. Naperville, Ill.: North Central Regional
Educational Laboratory, 2002.
q.. Ferguson, Ronald. "Why America's Black-White School Achievement Gap
Persists." Unpublished manuscript. Cambridge, Mass.: Harvard University,
r.. Fryer, Roland, and Levitt, Steven. "Understanding the Black-White
Test Score Gap in the First Two Years of School." Review of Economics and
Statistics 86 (2004): 44764.
s.. Hansen, Karsten; Heckman, James J.; and Mullen, Kathleen. "The
Effect of Schooling and Ability on Achievement Test Scores." Journal of
Econometrics 121 (2004): 3998.
t.. Heckman, James J. "Policies to Foster Human Capital." Research in
Economics 54 (2000): 356.
u.. Heckman, James J.; Larenas, Maria Isabel; and Urzua, Sergio.
"Accounting for the Effect of Schooling and Abilities in the Analysis of
Racial and Ethnic Disparities in Achievement Test Scores." Working paper.
Chicago: University of Chicago, Department of Economics, 2004.
v.. Heckman, James J.; Lochner, Lance; and Taber, Christopher.
"Explaining Rising Wage Inequality: Explorations with a Dynamic General
Equilibrium Model of Labor Earnings with Heterogeneous Agents." Review of
Economic Dynamics 1 (1998): 158.
w.. Heckman, James J., and Rubinstein, Yona. "The Importance of
Noncognitive Skills: Lessons from the GED Testing Program." American
Economic Review 91 (2001): 14549.
x.. Heckman, James J.; Stixrud, Jora; and Urzua, Sergio. "Evidence on
the Importance of Cognitive and Noncognitive Skills on Social and Economic
Outcomes." Unpublished manuscript. Chicago: University of Chicago, 2004.
y.. Heckman, James J., and Todd, Petra. "Understanding the Contribution
of Legislation, Social Activism, Markets and Choice to the Economic
Progress of African Americans in the Twentieth Century." Unpublished
manuscript. Chicago: American Bar Foundation, 2001.
z.. Herrnstein, Richard, and Murray, Charles. The Bell Curve:
Intelligence and Class Structure in American Life. New York: Free Press,
aa.. Jencks, Christopher, and Phillips, Meredith. The Black-White Test
Score Gap. Washington, D.C.: Brookings Institution, 1998.
ab.. Murray, Charles. "The Secular Increase in IQ and Longitudinal
Changes in the Magnitude of the Black-White Difference: Evidence from the
NLSY." Paper presented at the Behavior Genetics Association meeting,
ac.. Neal, Derek. "The Measured Black-White Wage Gap among Women Is Too
Small." Journal of Political Economy 112 (2004): S1S28.
ad.. Neal, Derek, and Johnson, William. "The Role of Premarket Factors
in Black-White Wage Differences." Journal of Political Economy 104 (1996):
ae.. Sackett, Paul; Hardison, Chaitra; and Cullen, Michael. "On
Interpreting Stereotype Threat as Accounting for African AmericanWhite
Differences in Cognitive Tests." American Psychologist 59 (2004): 713.
af.. Steele, Claude, and Aronson, Joshua. "Stereotype Threat and the
Test Performance of Academically Successful African Americans." In The
Black-White Test Score Gap, edited by Christopher Jencks and Meredith
Phillips, pp. 40127. Washington, D.C.: Brookings Institution, 1998.
ag.. Urzua, Sergio. "The Educational White-Black Gap: Evidence on Years
of Schooling." Unpublished manuscript. Chicago: University of Chicago,
Department of Economics, 2003.
More information about the paleopsych