You are on page 1of 26

Board of Regents of the University of Wisconsin System

Experience, Tenure, and Wage Growth of Young Black and White Men
Author(s): Bernt Bratsberg and Dek Terrell
Source: The Journal of Human Resources, Vol. 33, No. 3 (Summer, 1998), pp. 658-682
Published by: University of Wisconsin Press
Stable URL: http://www.jstor.org/stable/146337
Accessed: 07-01-2016 07:44 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Board of Regents of the University of Wisconsin System and University of Wisconsin Press are collaborating with
JSTOR to digitize, preserve and extend access to The Journal of Human Resources.

http://www.jstor.org

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Experience, Tenure, and Wage
Growth of Young Black
and White Men

Bernt Bratsberg
Dek Terrell

ABSTRACT
This paper studies the source of differences in wage growth between
young black and white workers. Focusing on "terminal" high school
graduates from the National Longitudinal Survey of Youth, we estimate
the returns to on-the-job tenure and general labor market experience us-
ing ordinary least squares, Altonji and Shakotko, and Topel estimators.
Results from all three estimators indicate that returns to general experi-
ence for black workers trail those for white workers, but that black work-
ers earn equal if not higher returns to tenure than do white workers.

I. Introduction

The difference in pay across races has long been a problem of sig-
nificant interest to labor economists. Recently, several authors have documented a
narrowing of the pay gap between white and black workers (Freeman 1981; Smith
and Welch 1989; Donahue and Heckman 1991).' Smith and Welch (1989) show that
the improvement in black relative wages has been particularly marked among young
workers. The improvement in the relative wages of young black workers has been
attributed both to increases in the quality of the schooling of black youth (Smith
and Welch 1989; Card and Krueger 1992) and to the enactment of affirmative action
policies (Freeman 1981; Donahue and Heckman 1991). Whatever the cause of the
Bernt Bratsberg is an associate professor of economics at Kansas State University. Dek Terrell is an
associate professor of economics at Louisiana State University. The authors thank Joe Altonji, Jim
Ragan, Steve Trejo, and two anonymous referees for helpful comments. The data used in this article
can be obtained beginning November 1998 through October 2001 from Bernt Bratsberg, Department of
Economics, Kansas State University, Manhattan, KS 66506, or from Dek Terrell, Department of Eco-
nomics, Louisiana State University, Baton Rouge, LA 70803.
[Submitted October 1993; accepted July 1997]
1. While this literaturedocumentsa substantialnarrowingof the black-whitewage gap between 1965
and 1980, the trendslowed and may have reversedduringthe 1980s (O'Neill 1990; Juhn,Murphy,and
Pierce 1991; Grogger1996).
THE JOURNAL OF HUMAN RESOURCES * XXXIII * 3

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 659

narrowing of the gap in wages between young black and white workers, the lasting
effect of this trend rests on the equality of wage growth across these groups.
Lazear (1979) warns that trends showing a narrowing of racial gaps in pecuniary
wages may be illusory, and stresses the importance of differences in wage growth
in creating discrepancies in lifetime compensation across races. Indeed, Lazear notes
that differences in the growth of wages over a career pose a far greater problem for
young black workers than any gap in starting salaries. This paper presents new empir-
ical evidence on differences in wage growth for young black and white workers
focusing on two critical components of wage growth, the returns to on-the-job tenure
and the returns to general work experience.
Despite the importance of differences in returns to tenure and experience across
races, the existing empirical literature offers no conclusive evidence on the size or
even the direction of such differentials. In particular, early studies yield conflicting
evidence on the difference in these returns across races.2 Moreover, most prior stud-
ies fail to take into account the recent debate on wage growth and job tenure, which
stresses that traditional methods of estimating returns to tenure and experience suffer
from potential biases caused by correlations of tenure and experience with unobserv-
able determinants of wages (Abraham and Farber 1987; Altonji and Shakotko 1987;
Marshall and Zarkin 1987; Topel 1991).
Two recent studies do address differences in the returns to general experience and
job tenure across races accounting for possible correlation between tenure and the
errorin the wage equation. Wolpin (1992) estimates parameters of a structuralmodel
using data from the National Longitudinal Survey of Youth (NLSY) and finds that
black workers receive a substantially smaller return to general labor market experi-
ence than white workers. His results also indicate that wages of black workers rise
faster with on-the-job tenure than do wages of white workers. Altonji and Shakotko
(1985) find similar evidence that slower wage growth of black workers stems from
lower returns to general experience in a sample of prime aged men drawn from the
Panel Study of Income Dynamics (PSID). Although these studies indicate the poten-
tial for a consensus on this question, more evidence is needed to settle the issue.
The literature contains several alternative methods of estimating returns to tenure
and the debate has not produced conclusive evidence on the superiority of one
method over others. This paper applies three estimation techniques to establish con-
clusive evidence on racial differences in the returns to general and specific labor
market experience. We estimate these effects using panels of young males drawn
from the NLSY, limiting the investigation to terminal high school graduates.3
This paper estimates the returns to tenure and experience using ordinary least
squares, Altonji and Shakotko (1987), and Topel (1991) estimators. All three estima-

2. For example,Lazear(1979) and O'Neill (1990) reportlower wage growthwith generalexperiencefor


blackworkers,whileLouryandGarman(1993) findhigherreturnsto generalexperienceforblackworkers.
Flanagan(1974) finds similarreturnsto experienceacross races, but higherreturnsto tenurefor older
black workersand lower returnsto tenurefor youngerblack workersthan comparablewhite workers.
Also, Shapiro(1984) finds lower returnsto tenurefor blackworkersthanwhite workers.
3. By focusing on panel data from a narrowlydefinedbirthcohort such as the NLSY, we avoid bias
causedby vintageeffects-which may be particularlyimportantfor black workersdue to differencesin
the qualityof educationacross birth cohorts. See Reich (1988) and Smith and Welch (1988) for two
alternativeviewpointson the importanceof vintageeffects.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
660 The Journal of Human Resources

tors find that black workers receive far lower returnsto general experience than white
workers, but results also suggest that black workers earn equal or even higher returns
to tenure than those of their white counterparts. The results appear consistent with
employer learning as in the Farmer and Terrell (1996) model, although furtherinves-
tigation is needed to explain the result. The results also offer a comparison of the
ordinary least squares, Altonji and Shakotko, and Topel estimators.
The paper is organized as follows. Section II presents a description of the data and
the construction of key variables. Section III outlines the methodology and discusses
possible reasons for discrepancies in returns to tenure and experience across races.
Section IV presents estimates of the returns to tenure and experience for black and
white workers. Section V contains auxiliary regressions used to check the sensitivity
of our results to alternative specifications of the model and to alternative measures
of tenure and experience. Section V also addresses the bias of each estimator to
further assess the reliability of comparisons across races. We conclude in section
VI by discussing the implications of our results in the context of several theoretical
models.

II. Data4

A large literature (Brown and Light 1992; Light and Ureta 1995;
Topel 1991; Williams 1994) warns that estimates of returns to on-the-job tenure and
general labor market experience may be quite sensitive to measurement errors or
assumptions used to construct tenure and experience variables. This study uses an-
nual data from 1979 through 1991 for young black and white males drawn from the
nonmilitary subsamples of the NLSY and the accompanying workhistory data set.
The detailed workhistory provides weekly data on the work status of subjects,
allowing construction of precise measures of on-the-job tenure and labor market
experience.
One problem in measuring work experience is determining the start of the career.
Light (forthcoming) investigates this important issue and finds that estimates of re-
turns to tenure and experience are sensitive to assumptions about the starting point
of the career, particularly in samples of young workers. Moreover, in a full sample
of workers complementarity between education and post-education training may lead
to differences in wage growth across races because of differences in educational
attainment between black and white workers. Following Wolpin (1992), we address
these issues by focusing on terminal high school graduates5 and assume that the
career begins at high school graduation.
Given the starting point for the career, our workhistory based measures of tenure
and experience count only weeks the subject reports working. The data appendix
supplies a detailed description of the construction of the tenure and experience vari-
ables as well as sample restrictions imposed to ensure data quality. Table 1 contains
summary statistics for these variables and others used in the analyses, listed sepa-

4. The appendixcontainsa detaileddescriptionof the constructionof samplesand key variables.


5. Terminalhigh school graduatesrefersto individualswho graduatehigh school but obtainno further
education.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 661

Table 1
Summary Statistics

t-Ratio for
White Black Ho: Equality
Variable Workers Workers of Means

ln(wage) 1.9999 1.8085 18.86


(0.4240) (0.4038)
Aln(wage) 0.0389 0.0329 0.70
(0.2056) (0.2511)
Tenure 2.7507 2.1139 8.04
(2.8701) (3.3250)
Tenuresquared 15.8022 10.5471 8.29
(30.0140) (23.5272)
Potential tenure 2.8487 2.1655 10.41
(2.9506) (2.5164)
Potential tenure squared 16.8193 11.0187 8.75
(31.6296) (24.4762)
Old job 0.6415 0.5329 8.89
(0.4796) (0.4990)
Experience 5.4825 4.9044 7.07
(3.4943) (3.2212)
Experience squared 42.2663 34.4243 7.70
(45.8902) (38.9515)
Potential experience 6.7799 6.7538 0.29
(3.8132) (3.6392)
Potential experience squared 60.5050 58.8518 1.21
(58.0015) (54.0931)
Initial experience 2.7318 2.7905 0.84
(2.8544) (2.8048)
Initial potential experience 3.9312 4.5883 7.54
(3.3975) (3.5701)
Part-time 0.0691 0.1144 6.08
(0.2537) (0.3183)
Married, spouse present 0.4431 0.2214 20.32
(0.4968) (0.4153)
SMSA 0.6532 0.7318 7.01
(0.4760) (0.4431)
Health 0.0172 0.0147 0.82
(0.1302) (0.1205)
Union 0.2471 0.3007 4.80
(0.4314) (0.4587)
Northeast 0.2155 0.1404 8.26
(0.4112) (0.3475)
NorthCentral 0.3922 0.1382 26.13
(0.4883) (0.3452)
South 0.2751 0.6698 34.39
(0.4466) (0.4704)
Quit 0.6042 0.5255 3.74
(0.4892) (0.4996)

Note: Samplesizes are5,568 for whiteworkersand2,308 for blackworkers,exceptfor Aln(wage)(3,113


whiteworkersand 1,036blackworkers)andQuit(1,564 white workersand862 blackworkers).Standard
deviationsare reportedin parentheses.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
662 The Journalof HumanResources

ratelyfor white and black workers.The table also suppliest-ratiosfor the null hy-
pothesisthatsamplemeansareequalacrossraces.Perhapsthe most strikingcontrast
between black and white workersemerges when comparingactual experienceto
potentialexperience(experiencemeasuredin calendartime since the inceptionof
the career).
Whilethereis no differencein samplemeansof potentialexperience,whitework-
ers on averagehave significantlymore actualwork experiencethanblack workers.
Light andUreta(1995) note thatpotentialand actualmeasuresyield quitedifferent
estimatesof returnsto experience,particularlyfor women,in samplesof whitework-
ers drawnfrom the NLS Cohortsof Men 14-24 in 1966 and Women 14-24 in
1968. Forcomparisonsacrossraces,the use of actualexperienceappearseven more
problematicdueto largedifferencesin the accumulationof workexperienceby black
andwhite workers.Forthis reason,we rely on the precisemeasuresof actualtenure
andexperiencethroughoutmostof the paper.We returnto the issue of how estimated
returnsare affectedby the use of potentialexperiencein section V.

III. Methodology

The primaryobjectiveof this paperconsistsof estimatingthe returns


to tenureandgeneralexperiencefor youngblackandwhiteworkers.Severaltheoreti-
cal models predicta relationshipbetweenwages andjob tenureas well as general
labormarketexperience.Theoryalso suppliesa numberof reasonsfor differences
in this relationshipacrossraces. In the humancapitalframeworkof Becker (1975)
and Mincer (1974), employees accumulategeneralhumancapital with additional
years of labormarketexperienceand receive higherwages in all jobs as a result.
In Becker'sformulation,employeesalso accumulatejob-specifichumancapitalwith
tenureon the job which resultsin higherwages with the currentemployer.
Humancapitaltheorylinks the returnsto experienceandtenureto returnsto gen-
eral and specifichumancapital,but the returnsto experienceandtenuremay differ
by race. For example,discriminationagainsta groupcould lower their returnsto
generalhumancapitalandthuslead to lowerreturnsto generalexperience.An alter-
nativeexplanationis thatthe groupsimply accumulatesless humancapitalin each
year of work experience,eitherby choice or, as Lazear(1979) argues,because of
lower levels of trainingprovidedby employers.
Anotherexplanationfor differencesin returnsto tenureand experienceacross
groups stems from priorassessmentsof ability conditionalon group.Farmerand
Terrell(1996) proposea model in which initial wages dependon the employer's
prioropinionof groupability,which may differ from actualability.Over time the
employerupdatesher assessmentof workerabilityandwages basedon observations
of individualoutput.Individualsfrom groupswhose averageproductivityis under-
ratedby employersearnhigherreturnsto tenurebecausetheirwages adjustas em-
ployerslearntheirtrue abilitythroughobservingoutput.6
Otherexplanationsof the relationshipbetween tenureand experiencefocus on

6. No suchupdatingoccurswith generalexperience.Workerswho changejobs areagainevaluatedbased


on priorsof groupabilitybecauseno individualoutputhas been observed.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 663

incomplete information about worker ability or the quality of the job match between
worker and firm. For example, in the Jovanovic (1979; 1984) search models workers
and employers learn about the quality of the job match over time and wages adjust
to reflect true productivity. Ultimately, the discovery of bad matches leads to quits,
while good matches survive. Wages rise with both experience and tenure in this
model, however the source of wage growth lies in an improvement of the quality
of the job match over time and not in productivity gains from the accumulation of
tenure and experience as in human capital models.
Given the importance of job match quality, specify the wage equation:

(1) wijt = Zin + Tijt + Xijt + C.i + ij + vijt

where i is an individual index; j indexes jobs; and t denotes time period. The equation
uses wijtto denote log wages; Zi is a vector of observable individual characteristics;
Tijtdenotes tenure on the job; and Xijtrepresents total accumulated labor market
experience.7 Our analysis allows for the possibility that the parameters 71,(a, and P
differ across races by specifying separate wage equations for each group.
The specification in equation (1) includes Eito capture unobserved heterogeneity
across individuals and rlj to reflect the quality of job match between worker and
firm. If tenure and experience are correlated with rij, as the basic job match model
predicts, ordinary least squares (OLS) yields inconsistent estimates of the parameters
of the wage equation. Thus, in addition to OLS, this paper uses two estimators, those
of Altonji and Shakotko (1987) and Topel (1991), to estimate the impact of tenure
and experience on wages.
Altonji and Shakotko (1987) propose an instrumental variables estimator to purge
the model of correlation between tenure and the quality of job-match using the devia-
tion of current tenure from mean tenure on the current job as an instrument for
tenure. By construction, this instrument is uncorrelated with ?i and eij,and is highly
correlated with current tenure.
Topel's estimation procedure is equivalent to instrumenting tenure using this devi-
ation variable, initial experience and other independent variables. In practice, Topel
proposes a two stage procedure to identify the returns to tenure and experience. The
first step consists of estimating the wage equation in first differences for all continu-
ing jobs:
(2) Awijt = ca + p + Avit.
Topel's methodology relies on the fact that the job match and individual errors are
constant over the duration of the job and thus disappear when the equation is esti-
mated in first differences. Using the estimate of oc + p from the first stage, Topel runs
a second stage regression to divide the estimate into returns to tenure and experience:
(3) wt - Tijt(a + P) = Zi, + XoijtP + i + ij+ Vijt
where Xoijtdenotes initial experience on job j. If initial experience is uncorrelated
with job match or individual heterogeneity, the Topel estimator yields consistent
estimates of the returns to tenure and experience.

7. The empirical specifications expand to include quadratic and quartic polynomials in both tenure and
experience, omitted here for ease of exposition.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
664 The Journal of Human Resources

Unfortunately, all three estimators fail to supply consistent estimates if initial and
current experience are correlated with unobserved individual heterogeneity and the
quality of the job match. Section IV contains ordinary least squares, Altonji and
Shakotko, and Topel estimates of the returns to tenure and experience for black
workers and white workers. Unless the bias of estimators differs substantially by
race, the comparison of returns to tenure and experience across races will not be
compromised by bias. We returnto the issue of bias in section V, focusing on differ-
ences in bias across races.

IV. Empirical Results

This section presents estimates of returns to tenure and experience


for black and white workers. Table 2, Table 3, and Figure 1 summarize results for
both races. In order, each table and figure presents results from ordinaryleast squares,
the Altonji and Shakotko estimator, and the Topel estimator. Comparisons across
races generally appear robust to alternative definitions and model specifications; sec-
tion V contains results exploring sensitivity to assumptions in greater detail.
Table 2 lists parameter estimates from the wage model used to investigate the
returns to tenure and experience. Excluding the coefficients on tenure and experience
variables, the parameter estimates appear similar across estimators, and signs and
magnitudes appear consistent with prior studies.8 The coefficients on the tenure and
experience terms drive any differences in wage growth across races and are the
primary focus of this study. To allow easy comparisons of wage growth across races,
Table 3 supplies predicted cumulative returns to tenure and experience computed
from the parameter estimates in Table 2.
Figure 1 contains plots of the cumulative wage growth due to tenure and experi-
ence for each race. Throughout, the solid line graphs the wage path for white workers
and the dashed line represents the wage path for black workers. The results for all
three estimators presented in Figure 1 and Table 3 corroborate the earlier work of
Altonji and Shakotko (1985) and Wolpin (1992). For all three estimators, black
workers receive much lower returns to general labor market experience, but earn
similar, if not higher returns to tenure than white workers.
Consider first returns to tenure. Using the OLS estimates, a black worker may
expect cumulative wage growth of 12.0 percent (e01132- 1) as the return to three
years of tenure compared 13.0 percent for a white worker. The Altonji and Shakotko
estimator attributes essentially zero wage growth to three years of tenure for both
races, but point estimates predict slightly higher returns for black workers. Likewise,
the Topel estimator indicates higher wage growth after three years of tenure for black
workers, 11.4 percent, compared to 9.5 percent for white workers. Although the
predicted returns to tenure vary substantially across estimators within race, all three
estimators find similar returns to tenure for black and white workers. In fact, for all
three estimators the standard errors in Table 3 suggest that differences in returns to
tenure across races at each time horizon might be explained by sampling errors.
A similar comparison of returns to general labor market experience across races

8. See Cain (1986) for a summaryof studiescomparingwage regressionsacrossraces.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 665

Table 2
Log Wage Regressions by Race

II. Altonjiand
I. OLS Shakotko III. Topel

White Black White Black White Black


Variable Workers Workers Workers Workers Workers Workers

Tenure 0.0511 0.0419 0.0045 0.0049 0.0384 0.0441


(0.0056) (0.0094) (0.0089) (0.0162) (0.0053) (0.0104)
Tenuresquared -0.0034 -0.0014 -0.0021 -0.0007 -0.0027 -0.0027
(0.0005) (0.0010) (0.0007) (0.0014) (0.0008) (0.0018)
Experience 0.0452 0.0279 0.0604 0.0383 0.0460 0.0239
(0.0057) (0.0089) (0.0063) (0.0103) (0.0031) (0.0058)
Experiencesquared -0.0014 -0.0013 -0.0014 -0.0012 -0.0017 -0.0013
(0.0004) (0.0007) (0.0005) (0.0008) (0.0007) (0.0017)
Part-time -0.1752 -0.0469 -0.1826 -0.0640 -0.1873 -0.0596
(0.0191) (0.0241) (0.0193) (0.0248) (0.0190) (0.0244)
Married,spouse present 0.0889 0.1038 0.1027 0.1126 0.1076 0.1282
(0.0105) (0.0194) (0.0107) (0.0198) (0.0116) (0.0223)
SMSA 0.1228 0.0677 0.1232 0.0539 0.1263 0.0635
(0.0102) (0.0178) (0.0102) (0.0183) (0.0102) (0.0180)
Health -0.0084 -0.0452 -0.0049 -0.0465 -0.0093 -0.0419
(0.0359) (0.0613) (0.0362) (0.0623) (0.0360) (0.0614)
Union 0.2891 0.2408 0.3093 0.2543 0.2931 0.2442
(0.0110) (0.0166) (0.0114) (0.0171) (0.0111) (0.0167)
Northeast -0.1344 -0.0633 -0.1090 -0.0532 -0.1300 -0.0610
(0.0171) (0.0381) (0.0175) (0.0387) (0.0171) (0.0381)
NorthCentral -0.1467 -0.2338 -0.1257 -0.2337 -0.1442 -0.2362
(0.0157) (0.0382) (0.0160) (0.0388) (0.0157) (0.0383)
South -0.1276 -0.2118 -0.1168 -0.2152 -0.1258 -0.2127
(0.0166) (0.0345) (0.0168) (0.0350) (0.0166) (0.0345)
Constant 1.6676 1.6877 1.6616 1.7097 1.6872 1.7149
(0.0206) (0.0416) (0.0210) (0.0425) (0.0182) (0.0402)
R2 0.3263 0.2349 0.3153 0.2116 0.2905 0.1906
Standarderror 0.3484 0.3541 0.3512 0.3594 0.3491 0.3554

Note: Samplesizes are 5,568 for white workersand 2,308 for blackworkers,except for first-stageTopel
resultswhicharebasedon subsamplesof 3,113 forwhiteworkersand1,036forblackworkers.TheAltonji-
Shakotkoestimatorinstrumentsfor Tenureand TenureSquaredusing predictedvalues from auxilaryre-
gressionsof TenureandTenureSquaredon all othervariablesin the wage regressionandwithinjob-spell
mean deviatesof Tenureand TenureSquared.The Topel estimatoris a two-stageprocedurewherepre-
dicted on-the-jobwage growth-estimated in a first-stagedifferenceequation-is nettedout of the ob-
servedwage to formthe dependentvariablein the second-stageregressionof predictedstartingwages on
initialexperience.Standarderrorsarereportedin parentheses.Forthe Topelestimator,standarderrorsare
correctedto reflect samplingerrorin the first-stageestimatesusing the methodof Murphyand Topel
(1985).

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
666 The Journal of Human Resources

Table 3
Cumulative Returns to Tenure and Experience

1 Year 2 Years 3 Years 4 Years 5 Years 6 Years

I. OLS
Returns to tenure
White workers 0.0477 0.0886 0.1226 0.1499 0.1703 0.1840
(0.0051) (0.0093) (0.0125) (0.0149) (0.0164) (0.0174)
Black workers 0.0404 0.0782 0.1132 0.1455 0.1750 0.2017
(0.0085) (0.0151) (0.0202) (0.0237) (0.0259) (0.0271)
Returns to experience
White workers 0.0438 0.0848 0.1229 0.1583 0.1908 0.2205
(0.0053) (0.0098) (0.0135) (0.0164) (0.0186) (0.0201)
Black workers 0.0266 0.0505 0.0718 0.0904 0.1063 0.1196
(0.0083) (0.0151) (0.0206) (0.0248) (0.0277) (0.0295)
II. Altonji and Shakotko
Returns to tenure
White workers 0.0025 0.0009 -0.0048 -0.0146 -0.0285 -0.0465
(0.0083) (0.0155) (0.0215) (0.0264) (0.0303) (0.0335)
Black workers 0.0041 0.0067 0.0080 0.0077 0.0059 0.0026
(0.0150) (0.0276) (0.0379) (0.0461) (0.0524) (0.0572)
Returns to experience
White workers 0.0590 0.1153 0.1689 0.2197 0.2679 0.3127
(0.0060) (0.0112) (0.0155) (0.0190) (0.0218) (0.0239)
Black workers 0.0372 0.0720 0.1044 0.1345 0.1622 0.1876
(0.0095) (0.0176) (0.0242) (0.0295) (0.0334) (0.0362)
III. Topel
Returns to tenure
White workers 0.0357 0.0660 0.0908 0.1102 0.1242 0.1328
(0.0048) (0.0088) (0.0120) (0.0145) (0.0165) (0.0181)
Black workers 0.0414 0.0773 0.1078 0.1328 0.1523 0.1664
(0.0094) (0.0168) (0.0226) (0.0269) (0.0303) (0.0333)
Returns to experience
White workers 0.0443 0.0851 0.1226 0.1566 0.1871 0.2143
(0.0029) (0.0054) (0.0076) (0.0094) (0.0111) (0.0126)
Black workers 0.0226 0.0425 0.0599 0.0746 0.0867 0.0961
(0.0053) (0.0098) (0.0136) (0.0166) (0.0191) (0.0211)

Note: Standard errors are given in parentheses. All predicted returns are created using the coefficients in
Table 2.

reveals a very different pattern. All three estimators predict substantially higher re-
turns to experience for white workers than black workers and the gap grows with
the level of experience. Consider the cumulative wage growth attributable to five
years of general experience, roughly the mean experience level in our sample. Ordi-
nary least squares estimates predict that a white worker will receive 21.0 percent
cumulative wage growth from five years of general experience, while a comparable
black worker receives only 11.2 percent cumulative wage growth. The Altonji and
Shakotko estimator indicates white workers achieve 30.7 percent cumulative wage
growth with five years of experience compared to 17.6 percent wage growth for a
black worker with five years of experience. The Topel estimator attributes 20.6 per-

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
.2' .2- .2
t*
0
.44

1 .1 White Workers .1 .1

0
0- O- 0-

I.5
4
6 4 0 2 4 6 6
Tenure Tenure

4)

to .4- .4- .4-

White Workers.
.3 - .3- .3 -
bo

.2- .2- .2-


0

-S
.1 - .1-

orkr
.1- A.'^
0- Black Workers 0- 0-
6 3E 4 6 3 6 9 6
Experience Experience

I. OLS II. Altonji-Shakotko

Figure 1
Tenure and Experience Profiles by Race

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
668 The Journal of Human Resources

cent cumulative wage growth to five years of experience for white workers and
finds that black workers receive only 9.1 percent wage growth at the same level of
experience. Again the predicted wage growth attributableto experience varies across
estimator, but the relative wage growth is similar for all three estimators-white
workers earn much higher returns to experience than black workers and the standard
errors in Table 3 indicate that this difference cannot be attributedto sampling error.
Summing tenure and experience effects gives an estimate of total wage growth
for workers who stay on a single job. Not surprisingly, at all levels of experience,
all three estimators predict slower wage growth for black workers continuing on a
single job than for white workers. For example, after five years on a single job, OLS
predicts 43.5 percent (e-1908+ 0.1703 - 1) wage growth for a white worker and only
32.5 percent wage growth for a black worker. Parameter estimates from the Altonji
and Shakotko procedure imply 27.0 percent wage growth for the white worker com-
pared to 18.3 percent for the black worker, and results from the Topel procedure
suggest 36.5 percent wage growth for white workers and 27.0 percent wage growth
for black workers with five years of tenure and experience.
Prior studies using the NLSY provide estimates of the returnsto tenure and experi-
ence for white males and allow some verification of the above results. Light and
McGarry (forthcoming) estimate returns to tenure and experience for white males
of varying education levels using several alternative model specifications. Their
specification most comparable to the Altonji and Shakotko estimator predicts cumu-
lative wage growth over five years of 46.1 percent for a white worker holding only
one job, with only 1.9 percent cumulative wage growth due to tenure. The total wage
growth exceeds our estimates from Table 3, but the small returns to tenure mirror
our results based on the Altonji and Shakotko estimator. The most elaborate error
specification used by Light and McGarry which "confirms Topel's prediction (p.
19)" estimates tenure returns similar to our results in Table 3 but estimates larger
returns to general experience. A comparison of our OLS results to those of Light
and McGarry reveals similar findings.9
The comparison across estimators appears consistent with prior studies using PSID
data (Altonji and Shakotko 1987; Topel 1991). For both races, the OLS and Topel
estimators attribute a large portion of cumulative wage growth to returns to tenure
while the Altonji and Shakotko instrumental variables approach finds very small
returns to tenure, attributing almost all wage growth to experience. These results
also suggest that Williams' (1991) finding of very similar returns to tenure for the
Topel and the Altonji and Shakotko estimators in data from Income Maintenance
Experiments may be an anomaly not often replicated in other data sets.

V. Sensitivity Analyses

Potentially biased estimates of returns to tenure from ordinary least


squares motivated the derivation of both the Altonji and Shakotko and the Topel

9. Williams(1994) also findshighertotalwage growthin a sampleof college graduatesdrawnfromthe


NLSY. For example,the fixed effects model in Williams'studypredicts51.3 percentwage growthafter
five years of post-collegeexperience.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 669

estimators. Unfortunately, if job-match and individual heterogeneity are correlated


with initial and current experience, these estimators also may yield biased and incon-
sistent estimates of returns to tenure and experience. Results in section IV indicate
substantial differences in predicted returns across estimators and, not surprisingly,
work by Altonji and Williams (1992) suggests that returns often lie somewhere in
between the estimates from the Altonji and Shakotko procedure and those obtained
using the Topel methodology.
Because this study focuses on a comparison across races, biased estimates of re-
turns do not necessarily compromise the key conclusion from our empirical model.
Indeed all three estimators support the same conclusion-black workers receive a
comparable or even higher return to tenure than white workers, but earn a far lower
return to general labor market experience. Only a difference in bias across races
challenges this conclusion. In this section, we evaluate the bias in the Altonji and
Shakotko and the Topel estimators. First, however, we examine the sensitivity of
the key results in section IV to the chosen metric of tenure and experience and to
the choice of empirical specification.

A. Model Specification
The workhistory-based measures of tenure and experience accumulate by one week
only if the worker actually works that week. In the literature, it is far more common
to use potential tenure and experience (time since the inception of the job-spell or
the career). In panel A of Table 4, we report results after reestimating the models
using the conventional measures of tenure and experience.
The use of potential tenure and experience accentuates the key findings from sec-
tion IV, as the gap between tenure profiles now increases slightly in favor of black
workers, and the gap between experience profiles is more pronounced in favor of
white workers. For example, using the Altonji and Shakotko estimator, tenure spe-
cific wage growth after three years on a job is 7.0 percent for black workers and
4.1 percent for white workers. In contrast, after five years the cumulative wage
growth with experience is 18.6 percent for white workers and only 5.4 percent for
black workers. Indeed, for black workers results from all three estimators appear
sensitive to the choice of metric for experience as the estimators show extremely
small returns to potential experience. These results echo findings for female workers
elsewhere (Malkiel and Malkiel 1973; Light and Ureta 1995) warning that potential
experience is a poor proxy for actual labor market experience.
In panel B of Table 4, we summarize results after expanding the three estimators
so that the regressors also include a quadratic polynomial of a time trend.'? Again,
the key findings from Section IV remain intact. While tenure profiles are very similar

10. In the Topel estimator,the time polynomialis addedto the second stage of the estimationprocedure
and the dependentvariableof the second stage is adjustedto accountfor a time trendcomponentin the
estimatedlineartenureand experienceeffect from the first stage. In the spiritof Altonji and Williams
(1992), we obtainestimatesof the time trendcomponentfrom OLS wage regressionson the variables
listed in Table 2 and a time trend.The estimatesare -0.027185 for white workersand -0.015974 for
blackworkers.Ignoringthis adjustmentreducescoefficientestimateson tenureby one-fifthandexperience
by one-fourth(comparedto thosereportedin Table4), buthas little effect on the comparisonacrossraces.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
670 The Journal of Human Resources

Table 4
Sensitivity Analyses

II. Altonjiand
I. OLS Shakotko III. Topel

White Black White Black White Black


Assumption/Variable Workers Workers Workers Workers Workers Workers

A. Potentialtenureand experience
Tenure 0.0607 0.0534 0.0206 0.0264 0.0446 0.0592
(0.0051) (0.0087) (0.0080) (0.0144) (0.0049) (0.0096)
Tenuresquared -0.0036 -0.0019 -0.0024 -0.0013 -0.0027 -0.0036
(0.0005) (0.0009) (0.0006) (0.0012) (0.0007) (0.0016)
Experience 0.0270 0.0062 0.0369 0.0129 0.0403 -0.0062
(0.0051) (0.0079) (0.0055) (0.0086) (0.0028) (0.0050)
Experiencesquared -0.0005 -0.0003 -0.0006 -0.0005 -0.0016 0.0004
(0.0003) (0.0005) (0.0004) (0.0006) (0.0006) (0.0012)
B. Regressionincludestime trend
Tenure 0.0418 0.0337 0.0125 0.0086 0.0407 0.0482
(0.0055) (0.0094) (0.0087) (0.0161) (0.0053) (0.0105)
Tenuresquared -0.0027 -0.0007 -0.0013 -0.0001 -0.0027 -0.0027
(0.0005) (0.0010) (0.0007) (0.0014) (0.0008) (0.0019)
Experience 0.0899 0.0539 0.1023 0.0652 0.0709 0.0358
(0.0064) (0.0100) (0.0069) (0.0110) (0.0032) (0.0059)
Experiencesquared -0.0032 -0.0025 -0.0034 -0.0025 -0.0017 -0.0013
(0.0005) (0.0008) (0.0005) (0.0009) (0.0007) (0.0017)
Time -0.0580 -0.0369 -0.0558 -0.0345 -0.0473 -0.0320
(0.0067) (0.0108) (0.0067) (0.0109) (0.0061) (0.0103)
Time squared 0.0020 0.0013 0.0017 0.0009 0.0014 0.0013
(0.0007) (0.0007) (0.0005) (0.0007) (0.0004) (0.0007)
C. Actualexperienceendogeneous
Tenure 0.0539 0.0507 0.0108 0.0215 0.0413 0.0539
(0.0058) (0.0100) (0.0089) (0.0162) (0.0053) (0.0102)
Tenuresquared -0.0035 -0.0017 -0.0020 -0.0009 -0.0027 -0.0027
(0.0006) (0.0011) (0.0007) (0.0015) (0.0008) (0.0019)
Experience 0.0395 0.0094 0.0544 0.0208 0.0432 0.0141
(0.0067) (0.0117) (0.0072) (0.0127) (0.0031) (0.0057)
Experiencesquared -0.0012 -0.0006 -0.0014 -0.0011 -0.0017 -0.0013
(0.0005) (0.0010) (0.0005) (0.0011) (0.0007) (0.0017)

Note:Regressionsalso includecontrolvariableslistedin Table2. In panelA, potentialtenureandexperi-


ence are measuredas calendartime since inceptionof job or graduation.In panel B, the quadratictime
trendis addedto the regressionin Table2 for OLS and Altonjiand Shakotko;and the Topel procedure
adds the quadratictime trendto the second step regressionafteradding0.027815 to the combinedfirst
ordertenureand experienceeffect for white workersand 0.015974 for black workers.In panelC, OLS
uses potentialexperienceas IV for actualexperience;the Altonjiand Shakotkoestimatoruses potential
tenureandexperiencein formingIVs for Tenure,TenureSquared,Experience,andExperienceSquared;
andthe Topelprocedureuses potentialinitialexperienceas IV for actualinitialexperience.See also note
to Table2.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsbergand Terrell 671

for the two races,black workersexperiencefar lower wage growthwith experience


thando white workers.Interestingly,the inclusionof the time-trendraisesthe esti-
matedreturnsto generallabormarketexperiencefor all threeestimators.To illus-
trate,evaluatedat five yearsof experience,cumulativewage growthis 44.8 percent
(OLS),53.2 percent(AltonjiandShakotko),and36.6 percent(Topel)for whitework-
ers and 23.0 percent,30.2 percent,and 15.7 percentfor black workers.1
Althoughnot reportedin Table4, we also experimentedwith alternativespecifica-
tions of the tenureand experienceprofiles.l2Altonjiand Shakotko(1987) find that
most tenure-specificwage growth occurs duringthe first year of a job-spell, and
arguefor the inclusionof an indicatorvariableset to unity if the job-spell is older
thanone yearin the regressionmodel.Moreover,MurphyandWelch (1990) demon-
stratethatthe commonlyused quadraticrepresentation of experienceseverelyunder-
statesearlycareerwage growth,andfavorthe quarticspecification.Consistentwith
the findingof Altonji and Shakotko,when we add the OLDJOBindicatorvariable
to the specification,results show positive first-yeareffects in the tenureprofiles.l3
However,cumulativereturnsto both tenureand experienceare virtuallyunaffected
by the inclusionof the indicatorvariable.Resultsfromquartictenureandexperience
specificationsindicate that, for white workers, the quarticrepresentationyields
slightly more concave profiles (indicatingfasterwage growthin early years) than
the quadraticrepresentation,but that for black workers,the quarticrepresentation
yield nonconcavetenureand experienceprofiles.More importantly,however, the
principalfindingsof SectionIV arenot underminedby the expandedrepresentations
of tenureand experienceprofiles.
In sum, resultsfrom the specificationanalysesindicatethat estimationof tenure
andexperienceprofilesis highlysensitiveto the use of actualversuspotentialtenure
and experienceand to the inclusionof a time trendin the empiricalmodel, but is
largelyunaffectedby more flexible representations of tenureand experience.How-
ever, our principalfindingthatyoung black workersreceive at least as high returns
to job-specifictenurebut much smallerreturnsto generallabormarketexperience
thando theirwhite counterparts,remainsrobustto a varietyof empiricalspecifica-
tions.14

B. IndividualHeterogeneityand Actual Experience


In a young samplewith a narrowage distributionsuch as ours,a substantialamount
of the variationin actualworkexperiencelikely stemsfromvariationin employment.

11. Interestingly,the veryhighreturnsto experiencethatwe reportin panelB arein line withwhatwe find
whenwe use a wage deflatorobtainedfromMurphyandWelch(1992) insteadof the personalconsumption
expenditureindex. We speculatethat use of the consumptionexpenditureindex understatesreal wage
growthduringthe 1980s, and thatthe time trendpicks up this effect.
12. Completeresultsfor all specificationsdiscussedin this section are availablefrom the authorsupon
request.
13. The OLS, AltonjiandShakotko,andTopelcoefficientestimateson OLDJOBare0.0357, 0.0270, and
0.0245 for white workersand 0.0134, 0.0215, and 0.0147 for black workers.
14. Table 1 indicatesthata largerproportionof blackworkersare unionizedin our samples.As an addi-
tionalsensitivitycheck,we reestimatedthe modelsdiscussedin sectionIIIfor samplesexcludingjob spells
wherewages were set by collective bargaining.The resultsare very similarto the basic resultsreported
above, and are availableuponrequest.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
672 The Journal of Human Resources

If spells of unemployment and nonparticipation during early stages of the career are
related to individual heterogeneity in wages, this could induce a positive correlation
between the workhistory based measure of experience and the individual specific
error component in the wage equation (Ei). In turn, this leads to an upward bias in
the estimated return to general labor market experience as the coefficient estimate
also captures the positive association between actual experience and individual heter-
ogeneity.
A plausible instrument that is correlated with actual experience but not with indi-
vidual heterogeneity is potential experience.15 In panel C of Table 4, we present
results from two-stage least squares procedures using potential experience as an in-
strumental variable for actual experience. Under the OLS heading, we construct the
IV by regressing (actual) experience and its square on potential experience and its
square and the other variables in the wage model. In the Altonji and Shakotko proce-
dure, we construct the IV using potential experience and job-spell deviations in po-
tential tenure.'6'17In the Topel procedure, we use potential initial experience as IV
for actual in the second step.
Results show that instrumenting for actual experience reduces the estimated wage
growth with general labor market experience for all three estimators-which is con-
sistent with the above argument that actual experience is positively correlated with
unobserved individual heterogeneity in wages. Moreover, while instrumenting yields
relatively small changes to the estimated experience profiles for white workers, the
impact on estimates for black workers is substantial. Compared to the predicted wage
growth with five years of experience reported in Table 3, after instrumenting pre-
dicted wage growth is reduced by about 10 percent for all three estimators for white
workers, and by 70 percent (OLS), 50 percent (Altonji and Shakotko), and 56 percent
(Topel) for black workers."8
Although our results indicate that individual heterogeneity bias may be more se-
vere in the sample of black workers, accounting for such bias leads to a larger deficit
in returns to experience for black workers and higher relative returnsto tenure. Thus,
ratherthan compromising the central results of this paper, accounting for heterogene-
ity bias reinforces the patterns observed in Section IV.19

15. Altonjiand Shakotko(1987) recommendusing this instrumentwhen actualexperienceis correlated


with ei but not with yij.
16. Using deviationsin actual(ratherthanpotential)tenurewhen formingthe instrumenthad virtually
no effect on the results.
17. We also followed Parent(1996) and constructedinstrumentsin the Altonjiand Shakotkoprocedure
usingindividualspecificmeandeviationsin actualexperience.Comparedto Table2, resultsindicatehigher
wagegrowthwithtenureandlowerwagegrowthwithexperienceforbothraces,butcomparabledifferential
growthby race.
18. Instrumenting for actualexperiencemay be even moreimportantwhenthe regressionincludesa time
trend(such as those reportedin Table 4, panel B). When we reestimatethe models in panel B treating
actualexperienceas endogeneousin the wage equation,resultsroughlyparallelthose discussedin the
presentsection.Specifically,predictedwage growthwith generallabormarketexperienceremainspracti-
cally unchangedby the IV procedurefor whiteworkers,butfalls substantially(althoughby lessermagni-
tudes thanthose above) for blackworkers.
19. Althoughthereare parallelsbetweenresultsin this section and those in the previoussection (where
we used potentialexperienceas a proxyfor actualexperience),we warnthatourresultsdo notjustify the
commonapproachof using potentialexperiencein the wage regression.For both races, using potential

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 673

C. Job-MatchHeterogeneityand Experience
Job shopping and job matching models (Johnson 1978; Jovanovic 1979) predict that
the quality of the job match improves with labor market experience. Both Altonji
and Shakotko (1987) and Topel (1991) address the correlation between the quality
of the job match (rlij) and experience, and supply methods of assessing the bias in
their estimators stemming from such correlation.
Altonji and Shakotko suggest that the asymptotic bias of their IV estimator may
be approximated using standard analysis of omitted variables. The bias depends pri-
marily on the expected job match quality (rlij)conditional on the level of experience,
which Altonji and Shakotko approximate through an analysis of quit behavior. As-
suming that improvements in job quality occur after quits and not layoffs, a logit
regression of quits on a polynomial in experience supplies the predicted probability
of a quit with each level of experience and cumulative totals of this predicted proba-
bility give the expected number of improvements in job quality conditional on expe-
rience. The product of the expected number of improvements and an assumed in-
crease in rlijwith each quit supplies the needed measure of expected job match quality
conditional on experience. The bias is then estimated by plugging the measure of
expected match quality into a standard bias formula.
In our samples, results from logit regressions of quits on a quadratic polynomial
in experience show minor differences in the probability of quits across races. For
example, after five years of experience the logit regressions predict 1.18 quits for
white workers compared to 1.22 quits for black workers. Table 5 contains estimates
of bias combining logit results with assumed values of 0.025 and 0.05 log points
for the improvement in job match quality per quit. The results indicate larger biases
in our sample than in that of Altonji and Shakotko (1987), attributableto the NLSY
samples having less experience and tenure and a higher incidence of quits than the
PSID sample used in the earlier study. Assuming equal improvements in job match
quality per quit implies almost identical estimates of the bias in cumulative returns
to both tenure and experience for each race. For example, assuming wage growth
of 0.05 log points per quit, the results suggest that the Altonji and Shakotko estimator
understates cumulative returns to three years of tenure by 0.0487 log points for white
workers and 0.0488 log points for black workers. Again assuming a 0.05 log point
increase in rij per quit, the Altonji and Shakotko estimator overstates returns to five
years of experience by 0.0821 for white workers and 0.0820 for black workers. These
calculations indicate that biases stemming from failure to account for job shopping
do not affect the estimated gaps between black and white workers in returns to tenure
and experience.
While improvements in match quality following quits induce a negative bias in
the tenure slope, any loss of match quality following layoffs could lead the Altonji
and Shakotko estimator to overstate the tenure slope. To complete the analysis of
bias in the Altonji and Shakotko estimator, Table 5 therefore also lists estimates of

experienceas a proxyfor actualexperienceleadsto estimatesthatseverelyunderstatethe experienceslope


and overstatethe tenureslope.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Table 5
Estimates of Bias from Failure to Control for Correlation Between Experience and Quality of Job M

Bias in Bias in
Bias in Coefficientof: Returnto Returnt
AssumedChangein rij per Quit or ThreeYears Five Yea
Layoff (A & S)/Assumedb (Topel) Tenure Tenure2 Experience Experience2 of Tenure of Experi

A. Whiteworkers ~~~~~~~~~~~~~~A. .v~o workers~~


WV~hite
Altonji& Shakotko
Al IQ = 0.025, Ar IL 0 -0.0084 0.0001 0.0087 -0.0001 -0.0244 0.041
ATIrQ= 0.05, AltIL = 0 -0.0167 0.0002 0.0174 -0.0002 -0.0487 0.082
ArlIQ=0.025, AilL = -0.025 -0.0003 -0.0000 0.0002 0.0000 -0.0009 0.030
ArilQ = 0.05, Ar|L = -0.05 -0.0006 -0.0000 0.0005 0.0000 -0.0018 0.060
Topel
b= 0 0.0030 0 -0.0030 0 0.0090 -0.0150
b 0.01 -0.0070 0 0.0070 0 -0.0390 0.065
B. Black workers
Altonji& Shakotko
ArllQ= 0.025, AnlL = 0 -0.0084 0.0001 0.0087 -0.0001 -0.0244 0.041
ATrQ = 0.05, AT1IL= 0 -0.0168 0.0002 0.0175 -0.0002 -0.0488 0.082
Al IQ = 0.025, Ar IL = -0.025 -0.0000 -0.0000 -0.0000 0.0000 -0.0002 0.000
ATIIQ = 0.05, ArilL = -0.05 -0.0001 -0.0000 -0.0000 0.0000 -0.0004 0.000
Topel
b= 0 0.0004 0 -0.0004 0 0.0012 -0.0020
b = 0.01 -0.0096 0 0.0096 0 -0.0312 0.052

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 675

bias allowing for both gains in match quality after quits and losses of match quality
after layoffs.20
Borjas (1984) and Farber (1996) report higher incidences of layoffs and displace-
ment among black workers and our results appear consistent with the previous stud-
ies. Logit regressions of layoffs on a quadraticpolynomial in experience predict 0.74
layoffs for white workers and 1.07 layoffs for black workers with five years experi-
ence. Given an assumed value for the decrease in rij per layoff, the larger incidence
of layoffs translates into a larger estimated layoff bias in the tenure slope for black
workers. As indicated in Table 5, however, when combined with the bias resulting
from quits, the total bias in the Altonji and Shakotko estimator appears quite minor-
particularly in the sample of black workers. For example, assuming a 0.025 log point
decrease in rij per layoff and a similar increase in rij per quit, our calculations suggest
that the Altonji and Shakotko estimator understates returns to three years of tenure
by 0.0009 log points for white workers and 0.0002 log points for black workers.
These calculations also predict that the estimator overstates the returns to five years
of experience by 0.0300 log points for white workers and 0.0002 log points for black
workers. In sum, the bias analyses suggest that biases in the Altonji and Shakotko
estimator are either very similar across races or negligible, so that biases cannot
explain the patterns in wage growth by race reported in Section IV.
Similar calculations for the Topel estimator are based on an auxiliary regression
of the job match component lij on experience and tenure:

(4) lij = Tijta + Xijtb + uijt

If Equation 4 describes the correlations between the quality of the job match and
tenure and experience, OLS applied to the wage model supplies biased estimates of
aoand P, as E[ c] = ax + a and E[P] = + b. Using Equation (3), the bias of the
Topel estimator can be expressed as:

(5) E[&] a - b - yx0(a + b) and

(6) E[P] = + b + yT(a + b)

where yxoTis the OLS coefficient from a regression of tenure on initial experience.
Although a and b are not observed, the sum a + b is estimated consistently by
subtracting the OLS estimate of ax + P from the intercept in the first stage of the

20. The literatureoffers little guidanceas to the size of the expectedloss of match qualityfollowing
layoffs.ExceptionsareAltonjiandWilliams(1992) andKeithandMcWilliams(1995). Basedon samples
fromthe PSID, AltonjiandWilliamscalculatechangesin matchqualityfollowing a layoff rangingfrom
-0.20 to -0.01 log points dependingon initial experienceand tenure.Using samplesfrom the NLSY,
KeithandMcWilliamsreportthateach pastinvoluntaryseparationreducesthe currentwage by 0.016 log
points.In our samples,wage changesfor workerswho are laid off are on averagepositive for both races
(butshowa sharpnegativerelationwith initialtenure).Becausethe meanwage growthbetweeninterviews
for workerswho stay on the samejob is less than0.04 log pointsfor bothraces(see Table 1), we conclude
that the loss of job matchqualityis very unlikelyto be as large as 0.05 log points (the extremevalue
assumedin Table 5) in our samples.Farber(1996) and Neal (1995) reportinsignificantdifferencesby
race in wage changesfollowing displacement.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
676 The Journalof HumanResources

Topelmodel.Topel assertsthatb > 0 basedon the fact thatthe qualityof job match
will improvewith experience,and the literaturegenerallyassumesthat a > 0 as
high qualityfirm-workercombinationsare unlikelyto end in quits or layoffs.21
Table5 also containsthe bias calculationssuggestedby Topel with two assump-
tions aboutthe size of b. We find a small bias for bothraces, similarto thatfound
by Topel (1991) in samplesdrawnfrom the PSID. Assumingequalcorrelationbe-
tweenjob matchqualityand experienceacrossraces generatessimilarbiases in the
Topel estimatesfor blackand white workers.For example,assumingb = 0.01, the
Topel bias calculationspredictthat our estimatesoverstatereturnsto five years of
experienceby 0.0650 log points for white workersand 0.0520 log pointsfor black
workers.22 Again assumingb = 0.01, bias calculationssuggestthatour Topel wage
path understates the cumulativewage growthdue to threeyearsof tenureby 0.0390
for whiteworkersand0.0312 for blackworkers.Overallbiasesappearsimilaracross
races and the calculationsindicatethatbias in the Topel estimatordoes not greatly
influencecomparisonsbetweenblack workersand white workers.
The approximationsof bias in the two estimatorsdependon assumedvalues of
an unobservedvariablein the wage model-the qualityof the job match.A simple
model of discriminationsuggests a greatercorrelationbetweenexperienceandjob
matchqualityfor black workers.Consideran economy with a mix of firms,some
discriminatingagainstblackworkersandothersnot discriminating.Overtime black
workersleave discriminatingemployersto look for nondiscriminating firms.Assum-
ing thatsome succeedin movingto nondiscriminating firms,the discriminationsup-
plies an additionalsourceof improvementin qualityof job matchthatis not present
in the sampleof white workers.This argumentsuggests that b is largerfor black
workersthanfor white workersandthatblackworkersmay expect largerimprove-
mentsin job matchwith experience-which leads to a largerdownwardbias in the
returnto tenurefor black workersin both the Altonji and Shakotkoand the Topel
estimators.
As a finalnote, we wish to interjecta wordof cautionregardingthe bias calcula-
tions.For bothestimators,the bias calculationsrequireassumptionsaboutthe unob-
servedqualityof job matchcomponentof wages.In addition,the Topelbias calcula-
tions fail to accountfor bias due to individualheterogeneity.For the Altonji and
Shokotkocalculations,a very large improvementin qualityof job matchfor white
workersand much smallerimprovementfor black workerscould lead to a larger
upwardbias in returnsto generalexperienceanddownwardbias in returnsto tenure
for white workers.While the discussionof discriminationabove arguesagainstthis
possibility,the fact thatbias exists andcannotbe exactlymeasuredclearlyindicates
the need for moreresearchin this area.We close by notingthatno bias calculation
suggestedby eitherAltonjiand Shokotko(1987) or Topel (1992) indicatesthatthe
largeestimatedgap in returnsto generalexperienceacrossraces and similarreturns
to returnsto tenurefor black and white workersis the resultof bias.

21. Topel (1991) arguesthata is unsignedand may well be negativein surveydata.


22. In our data,the estimatesof YXTare -0.2562 for white workersand -0.2270 for black workers.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsbergand Terrell 677

VI. Conclusions and Implications

This paperprovidesevidenceof negligibledifferencesin the returns


to tenureacrossracialgroups,but significantlylower returnsto experiencefor black
workersthan for white workers.The interpretationof these results differs across
theoreticalmodels. Perhapsthe simplestinterpretationis that, comparedto white
workers,blackworkersaccumulateless generalhumancapitalovertime,butroughly
equal amountsof specific humancapital.
Lazear(1979) postulatesthatdifferencesin on-the-jobtraininglead to disparities
in wage growthforblackandwhiteworkers.DuncanandHoffman(1979) andLynch
(1988) corroboratethis hypothesisin partby providingevidence of less on-the-job
trainingfor black workers.In the frameworkof Lazear,our resultsimply that the
on-the-jobtraininggiven to white workers,but not to black workers,leads to accu-
mulationof generalratherthan specific humancapital. Our results are therefore
consistentwith Lazear'spropositionthat employersrespondto affirmativeaction
legislation by providing on-the-jobtrainingin place of higher wages for white
workers.
Becker (1971) models discriminationas arising from the disutilitythat agents
receive from associatingwith membersof the minoritygroup.In this context,the
key implicationof our results lies in the fact thatblack workersstayingwith one
employerfare betterthan those changingjobs. One possible interpretationof this
resultis thatthe disutilityof the discriminatingworkersfalls afterassociatingwith
membersof the minoritygroup for a period of time. The decline in disutilityof
fellow workersincreasesthe value of minorityworkersto the firmand contributes
to the observedreturnto tenurefor those workers.
Farmerand Terrell(1996) suggest a model in which employersincorrectlyper-
ceive differencesin workerability acrossraces.In this model, the parityin returns
to tenureacrossracesoccursdue to an updatingof employerassessmentsof worker
ability that occurs only for workersaccruingtenurewith one employer.Dynamic
extensionsof statisticaldiscriminationmodels, such as Aignerand Cain (1977) and
Lundbergand Startz(1983), may predictsimilarresults.Minoritywages may grow
fasterwith tenurethanmajoritywages to makeup for biases in initialwages caused
by a noisy signal of initial ability, or, as in Oettinger(1996), ex ante uncertainty
aboutmatchproductivitymay be greaterfor minorityworkers,hinderingefficient
job mobility and slowing their wage growthwith experience.
This paperprovidesevidence on an importantsourceof racialinequalityin life-
time earnings.Whateverthe reasonfor the racialgap in valuationof generallabor
market experience, this paper provides a clear indication of the place where
improvementis neededto equalizewages. The lower returnsto experienceamong
black workersplace an ever wideninggap between the wages of black and white
workers that must be addressedif racial equalizationof lifetime earnings is to
occur.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
678 The Journal of Human Resources

Appendix 1
Data Appendix
This appendix details how we define the samples underlying our analyses and de-
scribes the construction of key variables. Variables listed in capital letters refer to
variables names as defined in the NLSY workhistory data set.

A. Samples
We examine wage growth for young black and white workers using panels of male
youths drawn from the 1979 though 1991 waves of the National Longitudinal Survey
of Youth (NLSY). We focus on "terminal high school graduates," and include in
our sample only individuals who had completed high school (but exclude GEDs)
and who did not enroll in college by the end of the sample period. The NLSY data
provide information on the last month and last year of enrollment in school as well
as the month and year the respondent graduatedfrom high school. We define a termi-
nal high school graduate as someone whose last month enrolled in school coincides
with the month he graduated from high school. Because graduation and last enroll-
ment dates are given only in month and year, we assume graduation and enrollment
dates occur on the fifteenth of the month.
Of the 12,686 individuals interviewed in 1979, 1,613 were black males and 3,790
were white males. From the full NLSY samples, we exclude the military subsamples,
leaving 1,451 black males and 3,181 white males in the 1979 interview. Of these, we
identified 386 black males and 778 white males as terminal high school graduates.23
The longitudinal data provide up to 13 observations for each individual. Individu-
als enter our sample in the year following graduation. For each year, we consider
only the job defined as the CPS job (or currentjob) in the NLSY data. The sample
excludes observations with missing values for variables used in the analyses as well
as observations where the individual is self-employed, in the armed forces, or em-
ployed in agriculture or by the government. To eliminate reporting errors, we also
drop observations where the respondent reports earning less than $1.50 or more
than $100 per hour (nominal wages are deflated using the personal consumption
expenditures index, 1987 = 100) or reports wage loss exceeding 67 percent or wage
growth exceeding 600 percent from the previous interview. These restrictions leave
sample sizes of 5,568 observations for white workers and 2,308 observations for
black workers. The final samples include observations for 1,078 individuals (728
white and 350 black) and 3,483 jobs (2,287 for white workers and 1,196 for black
workers).

B. Tenure and Experience Variables


The construction of tenure and experience series is based on the detailed information
provided in the NLSY workhistory data. We view the career as beginning immedi-
23. Thirty-twoadditionalindividualswho otherwisesatisfy our sample requirementsreporteda high
school graduationdatewithinthreemonthsof the last datethey wereenrolledin school.Inclusionof these
individualsin the samplehad only small effects on our results.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsbergand Terrell 679

ately following schoolingand defineboth experienceand tenurevariablesbased on


weeks of workexperiencefollowingthe last dateof enrollmentin school.The NLSY
workhistoryweekly variableSTATUSreportsthe work statusin each week of the
survey.For individualsgraduatingfromhigh school afterJanuary1, 1978, we con-
structthe experiencevariableby giving one week of experiencefor every week in
which the respondentworked(definedby STATUS)betweenthe date of graduation
and the interviewdate.24The samplealso includes299 individuals(222 white and
77 black) who graduatedfrom high school prior to 1978, and unfortunatelydata
priorto 1978 is limitedto weeks workedin 1975, 1976, and 1977. For these years,
we allow workersto accumulateexperienceequalto the minimumof reportedweeks
worked in a year and the numberof weeks remainingin the year after gradua-
tion.
We definethe tenurevariableusing a combinationof NLSY workhistoryreports
of the firstandlast week thatthe individualworkedon eachjob duringthe interview
year (START,STOP), STATUS,and the graduationdate. For each interview,the
NLSY workhistorydataincludeinformationon five jobs, labeledjobs one through
five. For continuingjobs, the variablePREVsuppliesinformationneededto match
thejob numberin the currentyearto thejob numberduringthe last interviewperiod.
Duringthe year of graduation,we give the respondentone week of tenurefor each
week workedbetweenthe morerecentdate of graduationor job startand the inter-
view date.In otheryears,continuingjobs arematchedto ajob numberin theprevious
year and tenureis initializedas the weeks of tenurerecordedfor thatjob at the
previousinterviewdate.New jobs areinitializedas zero duringthe startweek. Indi-
viduals then receive an additionalweek of tenurefor each week workedbetween
the initializationweek and the interviewweek (or the date the job ended, STOP).
Althoughin the final analysiswe consideronly datafor the CPSjob, the algorithm
generatestenureseriesfor everyjob heldby the respondentbetween1979 and 1991.
Ourtenureseries differsfrom the tenurevariableprovidedin the NLSY datain
two importantways. First, we do not allow accumulationof tenurewhile in high
school, consistentwith our assumptionthatthe careerbegins aftergraduation.Sec-
ond, the NLSY tenurevariableallows individualsto accumulatetenureduringtem-
poraryleaves and layoffs. Using the STATUSvariablein the workhistorydata,our
algorithmonly accumulatestenurein weeks thatthe employee actuallyworks.
When matchingcurrentjobs to pastjobs, we discovereda minorproblemin the
NLSY definitionof STOPand START.In general,the STARTweek in the current
year is set equalto the previousinterviewweek plus one andSTOPin the previous
year is set to the previousinterviewweek. Howeverin some cases STARTin the
currentyear is equalto STOPin the previousyear and in othersSTARTand STOP
fail to coincidewith interviewweeks. The firstproblemleads to doublecountingof
tenureaccruedduringthe interviewweek, while the secondcausestenureandexperi-
ence to accumulateat differentrates.OuralgorithmadjustsSTARTandSTOPduring
each year to match interviewweek and eliminatethese problems.Both problems
arepresentin thetenurevariableprovidedin the NLSYdata,however,butthe correc-

24. In extractingthese variablesfromthe NLSY workhistoryCD, we encountereda formattingproblem


in the extractionsoftwarethataccompaniedthe dataset. A programwrittenin C thatcorrectsthis problem
is availablefrom the authorson request.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
680 The Journal of Human Resources

tion is so minor that empirical results using the NLSY series are not likely to be
compromised.

C. Job Spells and Job Separations


For each interview, we focus on the CPS job (or currentjob) and use the constructed
tenure variable and the provided job characteristics for the job identified as the CPS
job in the NLSY data. We define two wage observations as belonging to the same
job spell if the CPS job in the current year was also the CPS job the last time the
respondent was interviewed. However, we allow temporary breaks in the job spell;
if a particularjob reappears as the CPS job within two years, our algorithm initializes
current tenure as the amount of tenure last observed on that job.
When a respondent leaves a job, the NLSY provides information on the reason
for the job separation. Based on the response, we code an indicator variable (quit)
as unity if the reason for the job separation appears voluntary (and not a lay-off,
plant closing, discharge, and so on). Thus, we have 862 observations on the quit
variable in the panel of black workers and 1,564 observations in the panel of white
workers.

References

Abraham,KatharineG., and HenryS. Farber.1987. "Job Duration,Seniority,and Earn-


ings." AmericanEconomicReview77(3):278-97.
Aigner,Dennis, and GlennCain. 1977. "StatisticalTheoriesof Discriminationin Labor
Markets."Industrialand LaborRelationsReview30(2):175-87.
Altonji,JosephG., and RobertA. Shakotko.1985. "Do Wages Rise with Job Seniority?"
WorkingPaperNo. 1616. Cambridge,Mass.:NationalBureauof EconomicResearch,
August.
- . 1987. "Do Wages Rise with Job Seniority?"Reviewof EconomicStudies54(3):
437-59.
Altonji,JosephG., and Nicolas Williams. 1992. "The Effects of LaborMarketExperience,
Job Seniority,and Job Mobilityon Wage Growth."WorkingPaperNo. 4133. Cam-
bridge,MA: NationalBureauof EconomicResearch,August.
Becker,GaryS. 1971. TheEconomicsof Discrimination.Chicago,IL: Universityof Chi-
cago Press.
. 1975. HumanCapital.Chicago,IL: Universityof ChicagoPress.
Borjas,GeorgeJ. 1984. "Race, Turnover,and Male Earnings."IndustrialRelations23(1):
73-89.
Brown,JamesN., andAudreyLight. 1992. "Interpreting PanelData on Job Tenure."Jour-
nal of LaborEconomics10(3):219-57.
Cain, Glen C. 1986. "The EconomicAnalysisof LaborMarketDiscrimination:A Sur-
vey." In Handbookof LaborEconomics,vol. 1, ed. OrleyAshenfelterand RichardLa-
yard,693-785. Amsterdam:ElsevierPublishingCompany.
Card,David, and Alan B. Krueger.1992. "School Qualityand Black-WhiteRelativeEarn-
ings: A DirectAssessment."QuarterlyJournalof Economics107(1):151-200.
Donahue,JohnH., and JamesHeckman.1991. "ContinuousVersusEpisodicChange:The
Impactof Civil RightsPolicy on the EconomicStatusof Blacks." Journalof Economic
Literature29(4):1603-43.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
Bratsberg and Terrell 681

Duncan,Greg J., and Saul Hoffman.1979. "On-the-JobTrainingand EarningsDifferences


by Race and Sex." Review of Economics and Statistics 61(4):594-603.
Farber,HenryS. 1994. "The Analysis of InterfirmWorkerMobility." Journalof Labor
Economics 12(4):554-93.
. 1996. "The ChangingFace of Job Loss in the United States, 1981-1993." Work-
ing Paper360, IndustrialRelationsSection. Princeton,NJ: PrincetonUniversity,March.
Farmer,Amy F., and Dek Terrell.1996. "Discrimination,BayesianUpdatingof Employer
Beliefs, and HumanCapitalAccumulation,"EconomicInquiry34(2):205-20.
Flanagan,RobertJ. 1974. "LaborForce Experience,Job Turnover,and Racial Wage Dif-
ferentials." Review of Economics and Statistics 56(4):521-29.
Freeman,RichardB. 1981. "Black Progressafter 1964: Who has Gainedand Why?" In
Studiesin LaborMarkets,ed. S. Rosen, 247-94. Chicago:Universityof ChicagoPress.
Grogger,Jeff. 1996. "Does School QualityExplainthe Recent Black/WhiteWage
Trend?" Journal of Labor Economics 14(2):231-53.
Jovanovic,Boyan. 1979. "Job Matchingand the Theoryof Turnover."Journalof Politi-
cal Economy 87(5):972-90.
. 1984. "Matching,Turnover,and Unemployment."Journalof Political Economy
92(1):108-22.
Juhn,Chinhui,Kevin M. Murphy,and BrooksPierce. 1991. "Accountingfor the Slow-
down in Black-White Wage Convergence." In Workers and Their Wages: Changing Pat-
terns in the UnitedStates, ed. MarvinH. Kosters.Washington,DC: AmericanEnterprise
Inst. Press.
Keith, Kristen,and AbagailMcWilliams.1995. "The Wage Effects of CumulativeJob Mo-
bility." Industrial and Labor Relations Review 49(1):121-37.
Lazear,Edward.1979. "The Narrowingof Black-WhiteWage DifferentialsIs Illusory."
American Economic Review 69(4):553-64.
Light,Audrey.Forthcoming."EstimatingReturnsto Schooling:When Does the CareerBe-
gin?" Economics of Education Review.
Light, Audrey,and KathleenMcGarry.Forthcoming."Job ChangePatternsand the Wages
of Young Men." Review of Economics and Statistics.
Light, Audrey,and ManuelitaUreta. 1995. "EarlyCareerWorkExperienceand Gender
Wage Differentials." Journal of Labor Economics 13(1):121-54.
Loury,LindaD., and David Garman.1993. "AffirmativeAction in HigherEducation."
American Economic Review 83(2):99-103.
Lundberg,Shelly, and RichardStartz1983. "PrivateDiscriminationand Social Interven-
tion in CompetitiveLaborMarkets."AmericanEconomicReview73(3):340-47.
Lynch,Lisa. 1988. "Race and GenderDifferencesin Private-SectorTrainingfor Young
Workers."IndustrialRelationsResearchAssociation,Proceedings41:557-66.
Malkiel,Burton,and JudithMalkiel. 1973. "Male-FemalePay Differentialsin Professional
Employment." American Economic Review 83(4):693-705.
Marshall,RobertC., and GaryA. Zarkin.1987. "The Effect of Job Tenureon Wage Of-
fers." Journal of Labor Economics 5(3):301-24.
Mincer, Jacob. 1974. Schooling, Experience, and Earnings. New York: National Bureau of
EconomicResearch.
Murphy,Kevin M., and RobertTopel. 1985. "Estimationand Inferencein Two-Step
Econometric Models." Journal of Business and Economic Statistics 3(4):370-79.
Murphy,Kevin M., and Finis Welch. 1990. "EmpiricalAge-EarningsProfiles,"Journalof
Labor Economics 8(2): 202-29.
. 1992. "The Structure of Wages." Quarterly Journal of Economics 107(1):285-
326.
Neal, Derek. 1995. "Industry-Specific
HumanCapital:Evidencefrom DisplacedWork-
ers." Journal of Labor Economics 13(4):653-77.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions
682 The Journal of Human Resources

Oettinger,GeraldS. 1996. "StatisticalDiscriminationand the EarlyCareerEvolutionof


the Black-White Wage Gap." Journal of Labor Economics 14(1):52-78.
O'Neill, June. 1990. "The Role of HumanCapitalin EarningsDifferencesbetweenBlack
and White Men." Journal of Economic Perspectives 4(4):25-45.
Capitaland the Wage Profile:Evidencefrom the
Parent,Daniel. 1996. "Industry-Specific
NLSY and the PSID." Universitede Sherbrooke,August.
Reich, Michael. 1988. "PostwarRacialIncomeDifferences:Trendsand Theories."In
Three Worlds of Labor Economics, ed. Garth Mangum and Peter Philips, 144-67.
Armonk,NY: M.E. Sharpe,Inc.
Shapiro,David. 1984. "Wage DifferentialsamongBlack, Hispanic,and White Young
Men." Industrial and Labor Relations Review 37(4):570-81.
Smith,JamesP., and Finis Welch. 1988. "RacialDiscrimination:A HumanCapitalPer-
spective." In Three Worlds of Labor Economics, ed. Garth Mangum and Peter Philips,
95-116. Armonk,N.Y.: M.E. Sharpe,Inc.
- . 1989. "Black EconomicProgressafterMyrdal."Journalof EconomicLiterature
27(2):519-94.
Topel, Robert.1986. "Job Mobility,Search,and EarningsGrowth:A Reinterpretation of
HumanCapitalEarningsFunctions."In Researchin LaborEconomics,vol. 8, pt. A, ed.
RonaldG. Ehrenberg.Greenwich,Conn.:JAI.
. 1991. "Specific Capital,Mobility,and Wages:Wages Rise with Job Seniority."
Journal of Political Economy 99(1): 145-76.
Williams,Nicolas. 1991. "Reexaminingthe Wage, Tenure,and ExperienceRelationship."
Review of Economics and Statistics 73(3):512-17.
-- . 1994. "EmploymentDuringCollege and Post-CollegeWages." Universityof Cin-
cinnati.
Wolpin,Kenneth.1992. "The Determinantsof Black-WhiteDifferencesin EarlyEmploy-
ment Careers:Search,Layoffs, Quits,andEndogenousWage Growth."Journalof Politi-
cal Economy 100(3):535-60.

This content downloaded from 134.129.120.3 on Thu, 07 Jan 2016 07:44:34 UTC
All use subject to JSTOR Terms and Conditions

You might also like