542 american sociological review the structure of ......tions of these results are considered and...

542542542542542 AMERICAN SOCIOLOGICAL REVIEWAMERICAN SOCIOLOGICAL REVIEWAMERICAN SOCIOLOGICAL REVIEWAMERICAN SOCIOLOGICAL REVIEWAMERICAN SOCIOLOGICAL REVIEW

This study is motivated by the idea that the racial gap in earnings is generated not

only by individual differences but also by systematic variation in the occupationalstructure that attenuates or exacerbates the effects of race. Using data from the1990 census and the Dictionary of Occupational Titles, a hierarchical linear model-

ing approach is employed that allows the simultaneous exploration of the mecha-nisms of income inequality operating both within and between occupations. Amongprivate-sector employees, striking evidence shows that racial disparities increase in

both absolute and percentage terms as one moves up the occupational earningshierarchy. The association between average occupational earnings and within-occu-pation racial disadvantage reveals an overlooked source of racial earnings inequal-

ity which constrains the opportunities available to upwardly mobile black men in theprivate sector. This association cannot be explained by measured individual charac-teristics, or by the status, demographic composition, or skill demands of occupa-

tions. In the public sector, on the other hand, racial inequality in earnings is notsystematically associated with average occupational earnings, and is instead moreclosely tied to individual human capital and occupational placement. The implica-

tions of these results are considered and directions for future research are sug-gested.

542 American Sociological Review, 2001, Vol. 66 (August:542–567)

The Structure of Disadvantage:

Individual and Occupational

Determinants of the

Black-White Wage Gap

decades, allowing black men to enter eliteeconomic sectors previously dominated bywhites (King 1992). Despite the gains madeby blacks in overcoming occupational seg-regation, however, black men’s earningscontinued to fall far short of the earnings oftheir white peers at all levels of economicattainment (Harrison and Bennett 1995). Ofgreater concern, this gap in earnings hadwidened substantially over the 1980s(Bound and Freeman 1992), despite narrow-

t the start of the 1990s, the eco-nomic status of black men was char-A

Eric Grodsky

University of Wisconsin, MadisonDevah Pager

University of Wisconsin, Madison

Direct all correspondence to Eric Grodsky,Department of Sociology, 1180 ObservatoryDrive, Madison, WI 53706 ([email protected]). We thank Bill Carbonaro, AdamGamoran, David Grusky, Robert M. Hauser, JohnAllen Logan, Lincoln Quillian, Rob Warren, andfour anonymous ASR reviewers for their helpfulcomments on early drafts. Our research was sup-ported in part by a grant to the University of Wis-consin-Madison Center for Demography andEcology (NICHD HD 05876). The second author

gratefully acknowledges support from the JacobK. Javits Fellowship. Order of authorship is al-phabetical and does not reflect differences in thecontributions of the authors; this is a collabora-tive project. This paper was awarded the JamesA. Thompson Award from the ASA section onOrganizations, Occupations, and Work, August2001.

acterized by two opposing trends. On onehand, unprecedented numbers of black menwere employed in high-level professional,managerial, and technical occupations(Farley 1996). Occupational segregation haddeclined appreciably over the preceding two

BLACK-WHITE EARNINGS INEQUALITY ACROSS OCCUPATIONSBLACK-WHITE EARNINGS INEQUALITY ACROSS OCCUPATIONSBLACK-WHITE EARNINGS INEQUALITY ACROSS OCCUPATIONSBLACK-WHITE EARNINGS INEQUALITY ACROSS OCCUPATIONSBLACK-WHITE EARNINGS INEQUALITY ACROSS OCCUPATIONS 543543543543543

ing black-white gaps in both educational at-tainment and cognitive test scores (Jencksand Phillips 1998; Mare 1995).

The existing research on racial earningsinequality, broadly divided along two linesof inquiry, does little to reconcile these op-posing trends. The occupational segregationapproach emphasizes the importance of oc-cupational placement and mobility in theearnings attainment process (Hout 1984;Stolzenberg 1975). This structural approachhighlights the disproportionate representa-tion of blacks in occupations of low status,skill, and earnings (Braddock and McPart-land 1987; Parcel and Mueller 1983) withthe implicit assumption that most of the ra-cial wage gap can be overcome through pro-gressive occupational redistribution (e.g.,Tomoskovic-Devey 1993). And yet, as notedabove, the black-white gap in earnings hasincreased over recent years despite advancesin black occupational mobility. Clearly, thereis more to the story than occupational place-ment alone.

The second line of inquiry analyzes earn-ings inequality across the labor market,demonstrating the persistence of wage dis-parities between blacks and whites net ofextensive statistical controls (Bound andFreeman 1992; Cain 1986; England et al.1988). Research in this tradition empha-sizes global factors in earnings attainmentrather than factors specific to occupationsor labor markets. When labor market vari-ables are brought into such analyses, theyare typically introduced as a series of dum-my variables representing broad occupa-tional or industrial categories (e.g., Kil-bourne, England, and Beron 1994), oftenignoring the potential variation in racialearnings inequality at different points in theoccupational structure.1

Although both these approaches offer use-ful insights into the factors underlying per-vasive racial disparities, neither offers an in-tegrated perspective on how labor marketplacement may mediate the emergence of ra-cial wage disparities. Understanding how lo-cation in the occupational structure shapesthe nature of disparities in earnings by raceis fundamental to gaining an accurate pictureof how earnings inequality develops. If cer-tain positions in the labor market are associ-ated with a more severe racial penalty thanothers (i.e., if there is an interaction betweenoccupation and race), then treating these in-dicators separately overlooks a key elementof racial stratification.

The importance of this relationship hasbeen highlighted in the work of Kaufman(1983). Using data from the 1970 census,Kaufman demonstrates that black men facethe greatest disadvantage in labor market di-visions at the high end of the earnings hier-archy. The implication of this finding is thatan equalization of the racial distributionacross labor market divisions would moveblacks from low-paying jobs with a smallracial gap to higher-paying jobs with a largerracial gap. While improving the absoluteearnings of blacks, this shift would increaseblack disadvantage relative to their white co-workers and widen levels of inequalityacross comparable employees. Given recentempirical trends toward greater equality inoccupational attainment and greater inequal-ity in earnings, the relationship betweenthese processes merits further investigation.

The present study builds on the importantinsights offered by Kaufman’s work. Usingmore contemporary data, we investigate therelationship between occupations and racialearnings inequality, explicitly investigatingvariation in the severity of the race penaltyacross the occupational hierarchy. We thengo beyond this descriptive decomposition toprovide an explanatory model of occupa-tional earnings inequality, looking to thecharacteristics of occupations that may gen-erate the observed patterns of racial dispari-ties.

1 One noteworthy exception is the dual labormarket literature, which explicitly investigatesvariation in the racial earnings penalty across la-bor market sectors (Beck, Horan, and Tolbert1978; Dickens and Lang 1985; Doeringer andPiore 1971; Sakamoto and Chen 1991). Whileproviding valuable insights, the broad labor mar-ket distinctions employed in this literature leavea tremendous amount of internal heterogeneityunexplored. This line of research has receivedextensive criticism concerning the imprecisionand inconsistency with which labor market dis-

tinctions are made and the oversimplification ofits dualist construction (for extensive reviews,see Cain 1976; Hauser 1980; Hodson andKaufman 1982).


A STRUCTURAL MODEL OF

RACIAL EARNINGS INEQUALITY

The notion that rewards inhere in jobs or la-bor market positions rather than individualassets has been fundamental in motivatingthe sociological understanding of the earn-ings attainment process. In particular, therole of occupations in shaping employmentexperiences is well established (Grusky andSørensen 1998; Kohn 1977; Sørensen 1996),and the direct association between occupa-tions and earnings is demonstrably strong(Featherman and Hauser 1978; Sewell andHauser 1975; Stolzenberg 1975). Althoughwe do not preclude the possible influencesof other labor market structures (e.g., indus-try, firm, job), we view the occupationalstructure as a central mediating mechanismin the process of earnings allocation andworker differentiation.

Building on this understanding, we de-velop a structural model of racial earningsinequality that distinguishes among the oc-cupational mechanisms that may contributeto the black-white gap in earnings. Ourmodel evaluates three potential sources ofearnings inequality at the occupational level:between-occupation sources, within-occupa-tion sources, and the interaction of the two.

The first of the three sources, between-oc-cupation earnings inequality, developsthrough a process of occupational sortingwhereby certain occupations enjoy higherwage rates than others. To the extent thatblacks are disproportionately concentrated inlower paying occupations net of their ownindividual attributes, racial disparities inearnings inevitably emerge. We do not di-rectly model the process of occupationalsorting, but view the observed matches ofindividuals to occupations as the outcome ofthat process.

The second mechanism operates withinoccupations, whereby blacks and whites inthe same occupation are offered differentwage rates. Certain occupations may demon-strate more severe penalties to blacks thanothers, leading to variation in racial earningsinequality across the occupational structure.2

To the extent that this variation is associatedwith observable characteristics of occupa-tions, we can develop causal explanationsregarding the differences between the earn-ings of black men and white men within thesame occupation.

The third mechanism can be thought of asan interaction of the between- and within-occupation sources of inequality. Thismechanism is present only if racial dispari-ties within occupations vary systematicallyaccording to the average earnings across (be-tween) occupations. This mechanism is apotentially important and much overlookedsource of racial earnings inequality. As dis-cussed above, Kaufman’s (1983) analysis ofthe 1970 census suggests a positive relation-ship between average earnings and racialearnings inequality, such that black disad-vantage grew larger as the average earningsof a labor market division increased. Sincethat time, however, the American economyhas changed a great deal. Some would arguethat today’s economy demands more skillsof its workers than ever before (Murphy andWelch 1994), with high-earning employeesincreasingly recruited on the basis of indi-vidual achievement rather than group ascrip-tion. If the competitiveness of today’s econ-omy leaves less room for discrimination,then we would expect a reversal in the rela-tionship between earnings and inequality,such that the black-white gap should narrow(net of other characteristics) as the earningsof an occupation increase.

Without taking into account the interactionof average occupational earnings and themagnitude of within-occupation earningsdisparities by race, we could under- (orover-) estimate the degree to which redistri-bution of blacks into higher-paying occupa-tions affects earnings inequality by raceoverall—the move into higher-paying occu-pations may be accompanied by a lower (orhigher) racial earnings penalty. Using ourstructural model of racial earnings inequal-ity, we can simultaneously identify thesources of racial earnings inequality thatemerge between occupations, within occupa-tions, and through the interaction of the two.

wages for manufacturing occupations demon-strate less of a race effect (Cotton 1989; Mossand Tilly 1996).

2 For example, there is some indication thatblacks in service sector occupations suffer agreater-than-average racial penalty, while the


Our Approach

In the first part of our paper we present a de-composition of the racial gap in earningsinto its three constituent parts. This analysisallows us to assess the relative influence ofindividual versus occupational effects on theblack-white wage gap, as well as to provideestimates of the between- versus within-oc-cupation sources of inequality. Next, weseek to explain each of the three mechanismsthat operate at the occupational level (be-tween-, within-, and the interaction of thetwo). We consider a variety of occupationalcharacteristics (discussed below) that maycontribute to the observed pattern of earn-ings inequality from each source. Finally, weprovide a qualitative analysis of those occu-pations with the most (and least) severe ra-cial wage gap, identifying potential mecha-nisms not captured by standard quantitativeanalyses. With this approach, we hope toprovide new insight into the labor marketprocesses leading to persistent racial dispari-ties in wages.

DATA AND VARIABLES

Data for these analyses come from the 1990Public Use Microdata Samples (PUMS) ofthe decennial census. We restricted our sam-ple to noninstitutionalized civilian men be-tween the ages of 25 and 64 who were em-ployed in nonfarm occupations at the timeof the decennial census and who had posi-tive earned income for 1989. Public- and pri-vate-sector samples were drawn separatelyto allow all parameters to vary by sector andin recognition of the fact that there is a smallsubset of occupations unique to each sectorthat makes the fit statistics and parametersfor each sector not strictly comparable.3 Forthe private-sector file, all nonwhites and arandom 25-percent sample of whites wereextracted from the 5-percent PUMS; for thepublic-sector file, the full 5-percent PUMSsample was used in order to provide suffi-cient cell counts in our occupation-levelanalyses. The samples include over 1 million

African American, Hispanic, Asian, andwhite men, about three-quarters of whom areemployed in the private sector. Appendix Apresents details concerning the sample selec-tion. Table 1 describes each variable and pre-sents means and standard deviations for allmeasures by sector of employment.

To understand the characteristics of occu-pations that determine wage rates, we con-sider a variety of compositional and requis-itional factors that have demonstrated impor-tant effects in previous research. We focuson three sets of occupational characteristicsthat may contribute to within- and between-occupation earnings inequality: occupationalprestige, composition, and skill require-ments.4 These components may operate dif-ferently in the public and private sectors, andfor men and women, so we estimate modelsseparately by sector and limit our analysisto men.5

Occupational Prestige

A long history of research has addressed therelationship between occupational standingand earnings, demonstrating the sizable pre-mium for employment in prestigious orhigh-status positions, net of individual back-ground characteristics (Duncan 1961;Featherman and Hauser 1978; Sewell andHauser 1975). Additional evidence suggests

3 Examples of such occupations include, in thepublic sector, legislators and air traffic control-lers, and in the private sector, private householdworkers.

4 Note that all labor market variables discussedhere refer to the prestige, composition, and skillrequirements of a national pool of occupations,rather than to the characteristics of an indiv-idual’s job. To the extent that this operation-alization is incorrect (i.e., to the extent that labormarket effects obtain at the job or firm level andnot at the occupation level), our estimates ofthese effects may be attenuated by aggregationerror. We expect that wage valuation is affectedby the general characteristics of an occupation,but should wages depend more on local labormarket factors, this variation will not be ac-counted for in the present analyses.

5 We have initiated a parallel set of analysesfor women, which we intend to pursue in a sepa-rate paper. The complexities which emerged inour model specification for women (related toadditional family structure variables and correc-tions for racial variation in women’s labor forceparticipation) prevent direct comparability andare thus more satisfactorily examined in an inde-pendent analysis.


Table 1. Definitions and Descriptive Statistics for Individual and Occupational Variables

Private Sector Public Sector

Variable Mean S.D. Mean S.D.

INDIVIDUAL VARIABLESlog of hourly earningsa 2.48 .72 2.55 .60

EducationNo school .01 .09 .00 .06

Less than 8th grade .05 .22 .02 .16

Some high school .13 .33 .07 .25

High school diploma/GED b .31 .46 .24 .43

Some college .20 .40 .22 .41

Associate’s degree .06 .24 .07 .26

Bachelor’s degree .16 .37 .19 .39

College + .09 .28 .19 .39

Race/EthnicityWhite (not Hispanic) .82 .49 .78 .50

African American (not Hispanic) .08 .27 .14 .34

Hispanic .08 .26 .06 .24

Asian .03 .16 .03 .16

Years of Work ExperienceWork experience c 22.49 11.12 22.86 10.70

Work experience squared 629.27 581.41 637.14 553.37

RegionNortheast .06 .23 .05 .22

Midlle Atlantic .15 .36 .15 .36

East north central .18 .38 .13 .34

West north central .07 .25 .07 .25

South Atlantic .17 .38 .21 .41

East south central .06 .23 .06 .23

West south central .10 .30 .10 .30

Mountain .05 .22 .06 .25

Pacific .16 .36 .16 .37

Marital StatusMarried .72 .45 .75 .44

Widowed .01 .08 .01 .08

Divorced .09 .29 .09 .28

Separated .02 .15 .02 .15

Never married .16 .37 .14 .34

Spouse absent (Yes/No) .02 .14 .02 .15

(Table 1 continued on next page)

that occupational standing may be positivelyassociated with racial disparities (Telles1994; Tienda and Lii 1987), making it aprime candidate for explaining both within-and between-occupation wage inequalities.

Following this argument, we expect thatwhile overall wages will rise with occupa-tional standing, so will the racial wage gap,leaving high-status blacks at a greater rela-tive earnings disadvantage than their lower


status peers. In these analyses, we use theNakao-Treas prestige score as a proxy foroccupational standing (Nakao and Treas 1994).6

Occupational Composition

Although prestige is an important dimensionof occupational standing, attributes of occu-pational incumbents may also contribute tothe desirability of an occupation independentof prestige. Tomaskovic-Devey (1993) dis-cusses the process of status composition,whereby the typical race or gender of an oc-cupation “becomes a fundamental aspect ofthe job, influencing the work done as wellas the organizational evaluation of the worth

of the work” (p. 6).7 All else being equal,therefore, the higher the concentration ofminority and female workers, the less thework will be rewarded (England et al. 1988;Tienda and Lii 1987). With respect to com-positional effects on the within-occupationracial gap in earnings, we take the view thatan occupation’s racial and gender composi-

OCCUPATIONAL VARIABLESOccupational Standing

Percent some college .51 .28 .52 .28

Prestige 43.63 14.53 44.34 14.58

Occupational CompositionPercent black .10 .06 .10 .06

Percent female .37 .30 .36 .29

Occupational Skill RequirementsCognitive skills d –.01 .92 .02 .91

Interpersonal skills e –.02 .88 .01 .88

Manual skills f .02 .79 –.01 .79

a Hourly earnings is estimated by dividing total earnings in 1989 by the product of weeks worked in 1989and average hours worked per week in 1989.

b Unfortunately, census data do not allow us to distinguish between high school graduates and those whoobtain a GED, a distinction that has important implications for wages (Murnane et al. 1995).

c Work experience is defined as age minus years of education minus 5. Years of education were assumedto be 0 for no school, 12 for high school diploma/GED, 14 for an associate degree, 13 for some college, 16for college graduate, 18 for a masters degree, 19 for a professional degree, and 21 for a Ph.D.

d Additive composite, including indicators of complexity in working with data, complexity working withpeople, general educational development, intellectual aptitude, verbal skills, and numerical aptitude.

e Additive composite, including indicators of adaptability to dealing with people, demand for talking orhearing, verbal skills and complexity in dealing with people.

f Additive composite, including indicators of manual dexterity, and three separate requirements of reach-ing, climbing and stooping.

(Table 1 continued)


Variable Mean S.D. Mean S.D.

6 We also estimated models that included oc-cupational education as a proxy for occupationalstatus (Hauser and Warren 1997). In these mod-els, the coefficient for occupational education isin the same direction as that for occupationalprestige, but is not statistically significant.

7 There is some circularity here which is diffi-cult to reconcile. Tomaskovic-Devey (1993) as-serts that occupational composition determinesearnings, but it may be the case that earnings ac-tually determine occupational composition. Thelatter argument would be consistent with the eth-nic queuing perspective in which members oflow-status minority groups are relegated to theleast desirable positions (Lieberson 1980; Model1997; Waldinger 1989). Our research cannot con-clusively adjudicate between these competingexplanations. Our main interest is in testing forthe presence of such an association (regardlessof causal direction) and examining its implica-tions for racial earnings inequality.


tion serve as status markers, with minority-and female-dominated occupations havinglower standing than white- or male-domi-nated occupations. Following our predic-tions for occupational prestige, we expectthat high concentrations of blacks and/orwomen will have a negative effect on theaverage earnings for an occupation and willattenuate racial wage inequality within oc-cupations. Our measures of racial and gen-der occupational composition are straight-forward: the percentage of workers in eachoccupation who are black or female.

Occupational Skills

The substantive requirements of occupationshave frequently been cited as a potentialsource of earnings inequality within and be-tween groups (England et al. 1988; Spenner1983). We consider three types of skill de-mands as possible sources of wage inequal-ity: cognitive skills, interpersonal skills, andmanual skills.8

The effects of cognitive skills have re-ceived a great deal of attention in prior lit-erature, as the rapid development of technol-ogy, increases in international trade, and thehuge growth in white-collar employmenthave contributed to a rising premium on in-tellectual aptitude and ability (Freeman1996; Murnane, Willett, and Levy 1995;Murphy and Welch 1994). Heightened com-petition in the economy has increased the in-centive for employers to weigh individualcompetence over ascribed characteristicssuch as race.9 We expect, therefore, that cog-nitive skill demands will be associated with

higher average occupational earnings andlower within-occupation earnings disparitiesby race.

Interpersonal skills represent a secondskill dimension of growing importance, par-ticularly given the rapidly expanding servicesector. Moss and Tilly (1996) cite interper-sonal (or soft) skills as an important factorin racial wage disparities, arguing that em-ployers tend to devalue the communicationskills and personality traits of blacks relativeto whites who have equivalent formal cre-dentials. Likewise, we expect that while in-terpersonal skills are characteristic of low-earning occupations overall, they will be as-sociated with greater earnings inequality be-tween blacks and whites.

Finally, manual skills represent an impor-tant third dimension of occupational differ-entiation which may shape the income pro-files of black and white incumbents. Manualskills, including physical strength and dex-terity, may be rewarded in the market whencognitive or interpersonal skills are not. Ifthis is the case, and if blacks are discouragedfrom entering occupations with an emphasison analytic skills, differentiation along thelines of manual skills may help explain an-other facet of racial earnings inequality. Fur-thermore, the products of occupations em-phasizing manual skills may be more con-crete and thus easier for a supervisor toevaluate. This may lead to a more merit-ocratic basis for decisions regarding em-ployee compensation. We think that occupa-tions that require manual skills, while offer-ing a lower average rate of pay, will tend tohave a lower racial gap in earnings than oc-cupations that do not emphasize manual skills.

Our scales for cognitive, interpersonal,and manual skills are derived from measuresincluded in the Dictionary of OccupationalTitles (DOT). The DOT is one of the fewdata sets to offer measures of jobs that arenot based on reports of job holders or on ag-gregations of job holder attributes. Thus, anyerrors in measures derived from the DOT arelikely not related to biases or errors associ-ated with individual job holders.

On the other hand, the Dictionary of Oc-cupational Titles was last updated in 1977,over a decade prior to the collection of thedata used in these analyses. Furthermore,many job titles were not updated for the

8 Unfortunately, we have no comparable mea-sure at the individual level, and therefore, to theextent that individual skill and occupational skillare correlated net of individual predictors, ourestimates of the effects of occupational skill de-mands on earnings may be upwardly biased.

9 Alternatively, however, some argue that theincentives for statistical discrimination rise withthe quality of the job, as the costs of training andforgone productivity are higher in such markets(Tomaskovic-Devey and Skaggs 1999). Employ-ers facing poor information about productivitydifferences among applicants may increasinglybase their decisions on group averages (or ste-reotypes), particularly when the costs of a faultyselection are high.


1977 edition. Finally, data for the DOT werecollected at the job level (on 12,099 jobs)rather than the occupation level (501 occu-pations for 1990). To create measures thatmap onto census occupation codes, research-ers have aggregated measures across jobs,often summing scales of items that were or-dinal but not interval (England and Kil-bourne 1988). This adds an unknownamount of error to DOT measures.10

Despite these limitations, the DOT offersthe best available data on the characteristicsof occupations and is well-suited for theevaluation of occupational attributes. Noother data source provides measures of suchan extensive range of occupational charac-teristics, particularly with respect to specificskill dimensions. Scales we estimate fromthe DOT have good face validity and reli-ability (Cronbach’s alpha ranges from .80 to.96). Furthermore, the cognitive-skill-de-mands scale correlates at .90 at the occupa-tional level with the Hauser and Warren(1997) measure of occupational education.This represents fairly strong construct valid-ity for an independent scale estimate.

We use a simple additive model to con-struct the three skill factors.11 The cognitiveskills factor is a linear composite of six in-dicators: complexity in working with data,complexity in working with people, generaleducational development, intellectual apti-tude, verbal skills, and numerical aptitude.The interpersonal skills factor is based onindicators of adaptability to dealing withpeople, demand for talking or hearing, ver-bal skills and complexity in dealing withpeople.12 Finally, the manual skills indicator

is an additive function of manual dexterityand three separate requirements of reaching,climbing and stooping. We would have pre-ferred to include measures more closelyaligned to manual skill and craftsmanship,but such measures were not available in theDOT. Nonetheless, we believe that the man-ual skills factor should be moderately corre-lated with true manual skills.

Each scale has a mean of 0, with standarddeviations of .92, .88, and .78 for cognitive,interpersonal, and manual skills, respec-tively. These scales are not constrained to beorthogonal, and in fact, the correlation be-tween cognitive and interpersonal skill de-mands is substantial (.81). These scale char-acteristics should be kept in mind when in-terpreting the skills coefficients; skill de-mand effects are estimated net of other skillsand relative to other occupations. We cannotconceive of a job characterized by the ab-sence of skills, only jobs with relativelystrong or weak demands for each of theskills specified.

Occupational Sector

One final feature of the labor market associ-ated with the magnitude of earnings dispari-ties by race is the distinction between publicand private sectors. The public sector haslong been regarded as the “vanguard of equalopportunity” (Krislov 1967), closely ap-proximating Weber’s ideal-type bureaucracywith its highly rationalized system of hiring,promotion, and remuneration (Grandjean1981). The established bureaucratic proce-dures that direct all stages of employmentdecisions in the public sector are thought toshield against forms of discrimination thatmay prevail in private-sector firms (DiPreteand Soule 1986; Moulton 1990).13 Indeed,empirical evidence suggests that the wage

10 Cain and Treiman (1981) offer a more de-tailed discussion of the strengths and weaknessesof the DOT as a data source for sociologicalanalyses.

11 Before settling on this additive model, wealso estimated a measurement model (with errorsin indicators and the latent factor) and a prin-ciple-components factor model. The factor mea-sures derived from the three models were corre-lated with one another above the r = .95 level,and results of our multilevel models using differ-ent factor indicators are similar in substance tothe findings we report here, although the size ofthe skills coefficients vary slightly.

12 Note that verbal skills and complexity indealing with people are included in both the in-

terpersonal and cognitive skills indicators. In ourmeasurement models of these factors (not re-ported here), we found that these indicators dem-onstrated strong loadings on both skill types, sug-gesting that they were important indicators ofboth cognitive and interpersonal skill.

13 But see Bridges and Nelson (1989) for evi-dence that the bureaucratic procedures that de-termine wage rates in government positions mayin fact produce greater wage disparities for lowerstatus workers.


gaps by race and gender are substantiallylower in the public sector than they are incorresponding private-sector occupations,and that a disproportionate number of blacksand women are employed in public-sectorpositions (Ehrenberg and Schwarz 1986). Weconsider the relationship between occupa-tional sector and earnings disparities by race,assessing the extent to which the public sec-tor effectively attenuates the negative rela-tionship between race and earnings across theoccupational distribution.

METHODS

We first estimate conventional OLS modelsto assess the contribution of individual-levelvariables to racial earnings inequality. Thesemodels provide a baseline estimate of racialearnings inequality net of individual charac-teristics. The strength of our approach, how-ever, emerges when we move to a two-levelframework. Using this approach, we can di-rectly test the hypotheses that result from ourstructural model of earnings inequality.14 Inthese models, the level-1 unit of analysis isthe individual, while the level-2 unit ofanalysis is the occupation.15 A single errorterm is estimated at the individual level,while separate error terms are estimated foreach occupational outcome (occupationalearnings and the racial gap in earnings).

The estimation of separate occupationaldisturbances offers several analytic advan-

tages. By partitioning earnings variance intoits within- and between-occupation compo-nents, the two-level model allows us to testempirically whether there is significantvariation across occupations in average earn-ings, as well as in the relationship betweenrace and earnings. Partitioning variance alsolets us assess the extent to which occupa-tional earnings and racial earnings inequal-ity are correlated net of individual-level pre-dictors.

If there is meaningful variation betweenoccupations in some individual-level out-come or predictor (i.e., earnings or race), wecan model this variation at the occupationlevel. The intercept or slopes from the indi-vidual-level equation thus become outcomesat the occupation level of analysis, each withits own disturbance. The occupation-levelmodel has two components—a fixed-effectscomponent that is a function of occupationalattributes, and a random component that rep-resents unmeasured occupational attributesand random error. Correlation among ran-dom components reflects the relationshipbetween occupational outcomes net of ob-served individual and occupational charac-teristics.

MODELS

At the individual level, we estimate the logof hourly earnings as a function of indi-vidual human capital, race and ethnicity, re-gion of residence, marital status, and a ran-domly distributed disturbance.16 Formally,the individual-level model is:

14 Many of the analyses we conduct using mul-tilevel models could be executed in a single-levelframework. The fixed-effects portions of ourmodels are simply complex interaction terms. Forexample, the within-occupation racial earningsdifference in an OLS model could be evaluatedwith a dummy variable for each j – 1 of j occu-pations and an interaction of the j – 1 dummyvariables with the indicator for blacks. Similarinteractions could be added to estimate the ef-fects of each of the occupational characteristicsincluded in our models. The standard errorsaround these level-1 parameters would then haveto be corrected by allowing for the correlation ofdisturbances within occupations. The cumber-some nature of these procedures, however, in ad-dition to the advantages of a multilevel modelingapproach outlined below, make individual-levelapproaches less desirable.

15 Occupations are coded according to the1990 three-digit detailed census classification.

16 We are conscious of the problems of scalingearnings in the loglinear form discussed byHauser (1980), Hodson (1985) and recently re-vived by Peterson (1999). We favor the loglineartransformation for both technical and rhetoricalreasons. Technically, we need to correct forheteroskedastic variation across the earnings dis-tribution in order to meet the standard assump-tions of our modeling approach. Rhetorically, weare interested in talking about relative earningsdifferences within occupations. The semilogform of the earnings equation allows us to do soin a straightforward manner. To assess the effectof this transformation, we ran a parallel set ofanalyses using untransformed hourly earnings asour dependent variable. The results provided sub-stantially stronger evidence of the effects we re-


Yij = β0j + β1-7j(Education) + β8j(Black)

+ β9j(Hispanic) + β10j(Asian)+ β11j (Experience)

+ β12j(Experience2)

+ β13-20j (Region)+ β21-26j (Marital Status) + rij .

where i indexes individuals and j indexesoccupations. The disturbance rij is assumedto be random normal with a mean of 0 andvariance of σ2.

If we impose the assumption that occupa-tions are identical in their wage functionsand that, net of observed variables, individu-als are randomly assigned to occupations(formally, that the r ij are independent withinoccupations), the individual-level model isidentical to a simple OLS regression modelthat excludes dummy variables for occupa-tion.

We challenge these assumptions in ourtwo-level models, in which we allow the in-tercept term (which represents mean occu-pational earnings for white men) and therace coefficient (which reflects the within-occupation racial gap in earnings) to varyfreely across occupations. This approach al-lows us to directly assess the extent to whichblack disadvantage is generated through dis-proportionate concentration of blacks inlower-paying occupations (reflected in theintercept), their concentration in occupationsin which they receive less pay than theirwhite counterparts (reflected in the race pa-rameter), or both.

To explore the effects of occupationalcharacteristics on the process of earnings al-location, we estimate the intercept term andthe race parameter for black men from theindividual-level equation as dependent vari-ables at the occupational level.17 Formally,the model for the intercept is:

β0j =γ00 + γ01(Prestige)+ γ02-03(Composition)

+ γ04-06(Skill Demands) + u0j .

where j indexes occupations, β0j is the inter-cept term from the individual-level equation(representing average occupational earningsadjusted for individual attributes), and u0j isan occupation-specific disturbance assumedto be normally distributed with a mean of 0and variance τ00.

Similarly, the formal model for the effectsof occupational characteristics on the racialgap in earnings within occupation j is:

β8j = γ80 + γ81(Prestige)+ γ82-83(Composition)

+ γ84-86(Skill) + u8j .

where β8j is the race coefficient for blacks inoccupation j and u8j is an occupation-spe-cific disturbance in the association betweenrace and earnings assumed to be normallydistributed with a mean of 0 and variance τ88.

Under this model specification, the inter-cept term represents the average earnings ofwhite men in occupation j, while the racecoefficient represents the deviation of theaverage earnings of black men in occupationj from the average earnings of white men inthe same occupation. Our interest is in racialearnings inequality (the race coefficient), butin order to accurately assess the role of oc-cupation-level variables in generating racialearnings inequality, we must estimate boththe intercept and the race coefficients simul-taneously. If we estimated only the race co-efficient at the occupation level, our esti-mates of γ80-86 as well as the variance of u8jwould be upwardly biased to the extent thatthese occupational factors (and unobservedsources of variation in occupational earn-ings) affect both white and black workers.

Before moving on, we wish to cautionreaders regarding the interpretation of occu-pation-level coefficients and standard errors.In most applications of the two-level randomcoefficients model, researchers have samplesof units at both levels of analysis. For ex-ample, in research on school effects re-searchers might have a sample of studentstaken from a sample of schools. Ideally, thesampling probabilities for both levels ofanalysis will be known, and standard errors(and perhaps point estimates) can be ad-justed accordingly. In the present study, wehave a sample of individuals but a census ofoccupations (with the exception of those oc-

port here, leading us to conclude that our use ofthe semilog specification represents a conserva-tive estimate of the relationships we observe.

17 This modeling approach is also known as aslopes-as-outcomes model or random coefficientsmodel.


cupations excluded because of sample re-strictions). Our final sample represents over90 percent of the employed civilian nonfarmmale population between the ages of 25 and64 and all of the occupations in which theyare employed. Although standard errors forindividual-level predictors can be interpretedin the usual fashion, standard errors at theoccupation level, because of the nature ofthe occupation-level sample, have a moreambiguous interpretation. We recommendthat the occupation-level standard errors beviewed as estimates of parameter dispersioncontaminated by measurement error ratherthan as the traditional measure of samplingerror. The smaller the standard error, themore consistent the effects of that measureat the occupation level.

RESULTS

Individual-Level Variation

Our initial (single-level) estimates of theearnings inequality experienced by blackmen are shown in Table 2a. The predictedunadjusted difference in log hourly earningsof black men and white men in the privatesector is –.34 log units (a difference of $3.65per hour at the private-sector mean) (seeOLS Model 1). This coefficient becomes thebaseline for the percentage of the black-white gap in earnings that is left unex-plained. By definition, 100 percent of thegap is unexplained in the initial model.

Adding human capital variables (educa-tional attainment and potential years of ex-perience) (OLS Model 2) reduces the coeffi-cient for blacks by 38 percent to –.21. Theregional variables (in OLS Model 3), in-cluded to control for geographic differencesin earnings due to labor supply and demandfactors, reduce the black coefficient by an-other 3 percent to –.20 Finally, includingmarital status and an indicator for whether aspouse is absent (OLS Model 4) reduces thepredicted race gap to –.16.18 In total, the in-

clusion of education and potential experi-ence, region and marital status reduces theassociation between race and log hourlyearnings by one-half for men working in theprivate sector, leaving a substantial wagepenalty for black men net of individual-levelpredictors.19

In the public-sector sample, we find im-portant differences in the nature of racial dis-parities. First, the baseline OLS difference inexpected log hourly earnings for black menand white men in the public sector is appre-ciably smaller than it is in the private sector.Nonetheless, without adjusting for any indi-vidual differences we find black men earnabout 21 percent less than white men in thepublic sector (a difference of $2.85 at thepublic-sector mean). Adding education andpotential experience to the equation halvesthe earnings disadvantage for blacks in thepublic sector from –.24 log units to –.11 logunits. This is a larger proportionate reduc-tion than was associated with the inclusionof human capital measures in the private sec-tor where black-white differences were re-duced by 38 percent. While not perhaps colorblind, the public sector appears to operateunder a more meritocratic system of wageallocation than the private sector, weighingmore heavily the formal credentials of edu-cation and experience.

Adding controls for region has essentiallyno effect on public-sector racial earnings dif-ferences, but including indicators for mari-tal status and spouse absence reduces thepredicted racial earnings difference by anadditional 10 percent relative to the OLSbaseline coefficient for black men. The finaladjusted OLS estimate of black earnings dis-advantage in the public sector is 9.0 percent,a little more than half of the predicted 15.5percent difference found for the private sec-tor. While perhaps not the “vanguard ofequal opportunity,” the public sector comesmuch closer to achieving racial parity inearnings than does the private sector.

data. Rather, we include marital status as a pre-dictor to obtain a conservative estimate for theblack-by-earnings association.

19 These estimates are consistent with previousresearch on black-white wage differences usingdata from similar time periods. For example,Bound and Freeman (1992) estimate an adjustedgap of –.179 using the 1989 CPS earnings data.

18 The role of marital status in earnings equa-tions for men has been the subject of some con-troversy in the literature. Although we are in-clined to follow Korenman and Neumark (1990)in attributing the bulk of the male marriage ef-fect to increased productivity rather than selec-tion, we do not advance that claim with these


Although important differences emergefrom our public- and private-sector analyses,both analyses reveal a substantial race gapleft unexplained by individual-level vari-ables. We thus turn to our occupation-levelanalyses as a means of better understandingthe mechanisms which underlie racial dis-parities in earnings.

Table 2b. Measures of Fit for HLM Models

HLM Model 1 HLM Model 2

Measure of Fit Private Sector Public Sector Private Sector Public Sector

Mean Occupational Earnings of WhitesVariance .050 .043 .052 .044

χ2 95,345 42,191 90,794 36,082

D.f. 463 465 450 431

Reliability .930 .872 .942 .889

Black DeviationVariance .— .— .005 .013

χ2 — — 1,052 1,584

D.f. — — 450 431

Reliability .— .— .354 .497

Overall deviance 1,464,922 533,758 1,464,523 533,230r (intercept, black) .— .— –.551 –.141

Table 2a. Individual-Level (OLS) and Multilevel (HLM) Model Estimates of the Coefficient forRacial Differences in Wages: PUMS, 1990


Estimated Percent Estimated PercentModel Effect Unexplained Effect Unexplained

Individual-Level Models

OLS Model 1 –.338 100.00 –.239 100.00(baseline) (.003) (.003)

OLS Model 2 –.212 62.83 –.114 47.69(adds human capital variables) (.003) (.003)

OLS Model 3 –.201 59.70 –.113 47.38(adds region) (.003) (.003)

OLS Model 4 –.155 46.06 –.090 37.59(adds marital status) (.003) (.003)

Occupation-Level Models

OLS Model 5 –.087 25.86 –.047 19.75(adds occupations) (.003) (.003)

HLM Model 1 –.089 .— –.048 .—(occupation free) (.003) (.003)

HLM Model 2 –.093 .— –.048 .—(occupation and race free) (.005) (.008)

Note: Numbers in parentheses are standard errors.

Between-Occupation Variation

The occupation-level analysis assesses theimportance of each of the three inequalitygenerating mechanisms discussed earlier—variation in earnings between occupations,variation in the within-occupation earningsdisadvantage experienced by black men, and


the interaction of the two. The first mecha-nism operates through the differential con-centration of blacks and whites in high- orlow-paying occupations. The magnitude ofthis source of variation can be measured inthe OLS framework by including dummyvariables for each of the 468 private-sector(or 471 public-sector) occupations, or bymoving to a two-level hierarchical linearmodel in which the intercept (representingaverage occupational earnings for whites)varies freely across occupations.

Models including controls for occupationare shown in the bottom panel of table 2a.The estimates for OLS Model 5 are directlycomparable and quite similar to the HLM re-sults for the model in which only the inter-cept is freed (HLM Model 1).20 Note that byincluding controls for occupation, the inter-pretation of the race coefficient changes. Theindicator for race now represents the ex-pected within-occupation difference in logearnings between white workers and blackworkers (in contrast to the average black-white earnings difference due to both within-and between-occupation differentiation).This allows us to distinguish between earn-ings inequality that emerges as the result ofdifferential placement versus that due to dif-ferential rewards. Under both models, occu-pations mediate approximately 20 percent ofthe black-white gap in earnings.21 Allowingoccupational intercepts to vary (or includingdummy variables for occupation) brings thetotal percentage of the expectation of therace coefficient we have accounted for toroughly 75 percent. Thus, a majority of theracial gap in earnings can be accounted forby individual differences in human capital,region, and marital status (55 to 60 percent)

and by the concentration of blacks in low-paying occupations (20 percent). There re-mains, however, a significant effect of race,even after controlling for individual charac-teristics and occupational sorting.

The race effect in HLM Model 1 representsthe average difference in earnings betweenblack and white workers in the same occupa-tion. If the effect of race were constant acrossthe occupational structure (net of differencesdue to occupational sorting), then this esti-mate would provide an accurate assessmentof the within-occupation racial gap in wages.If, however, the effect of race varies depend-ing on one’s position in the labor market,then this average estimate conceals impor-tant information regarding the role of occu-pations in shaping racial disparities. HLMModel 2, the HLM baseline model, providesan empirical test of this proposition. Thismodel includes all of the individual-levelpredictors and allows both the intercept andthe race coefficient to vary across occupa-tions. Essentially, this amounts to estimatinga separate intercept and slope term for eachoccupation included in our sample.

The results of this model indicate that ra-cial earnings inequalities vary significantlyacross occupations in both the public and pri-vate sectors (χ2 = 1,052, d.f. = 450 in the pri-vate sector, and χ2 = 1,584, d.f. = 431 in thepublic sector).22,23 The hypotheses for homo-

20 These models formally differ with respectto specification of the error term. While the OLSmodel includes one error term that varies acrossindividuals, HLM models include error terms atboth the individual and occupation levels.

21 The test statistics for HLM Model 1 demon-strate the significant improvement in fit that re-sults from allowing average earnings to varyacross occupations (χ2 = 95,345, d.f. = 463 in theprivate sector and χ2 = 42,191, d.f. = 465 in thepublic sector). Significance tests for occupation-level variation test the model specified against amodel in which occupational variation is con-strained to 0.

22 For a model in which a single level-1 slope(or only the intercept) is freely estimated, the re-liability for the level-1 parameter is equal to theparameter variance divided by the sum of the pa-rameter and error variance for a particular occu-pation. In the case of the intercept, that quantityis τ00/ (τ00 + νqqj), where νqqj is the error varianceof the intercept estimate for men in occupation j.νqqj comes from the error variance-covariancematrix for occupation j , and in the case of onerandomly varying coefficient the matrix is scalarand equals σ2/nj, where σ2 is the level-1 errorvariance and nj is the number of observations foroccupation j. However, in the case of two ran-dom coefficients, the covariance of the two coef-ficients must be taken into account. The formulafor νqqj then becomes σ2 (X′j Xj)–1 where in thiscase the matrix Xj includes a column of 1s for theintercept and a column for the race indicator (1for black men, 0 for white men).

23Although the intercept term is estimated quitereliably in each model (with reliability > .90), therace coefficient is not. The average reliability of


geneity in both occupational earnings andwithin-occupation racial earnings differencesare thus soundly rejected. This finding sup-ports two of our basic hypotheses concern-ing the mechanisms by which racial earningsinequalities obtain at the occupation level.24

The Relationship between

Occupational Earnings and Racial

Disadvantage

To evaluate the third possible mechanism ofracial earnings inequality—the interactionbetween occupational earnings and racialearnings differences within occupations—we regressed estimated within-occupationearnings differences on estimated averageoccupational earnings.25 The regression line,along with point estimates for predictedwithin-occupation black earnings differ-

the black-white wage gap estimate remainsaround .35. This may be due in part to the rela-tively low variance in the adjusted wage gapacross occupations (relative to variance in wagesoverall), and may be further compounded by thesubstantial intercept-race correlation.

24 The estimate of the average race effect un-der this model does not differ substantively fromthe preceding HLM model. The slight change inthe size of the coefficient for private-sector mod-els is due to the use of Bayesian estimation pro-cedures that place greater weight on more reli-able estimates. In this model, we are less inter-ested in the fixed-effect estimate presented thanwith the randomly varying estimate produced foreach occupation that serves as one of our depen-dent variables in the following analyses.

25 We used the empirical Bayes estimates ofearnings for the occupational earnings/within-oc-cupation racial earnings differences for theseanalyses, rather than the OLS estimates. Giventhe size of our overall sample and the reliabilityof intercept estimates, the OLS and empiricalBayes estimates for occupational earnings are al-most identical (r = .98). The estimated coeffi-cients for within-occupation earnings differ-ences, however, have a much lower reliabilityand much lower sample sizes in general. To cor-rect for these shortcomings, we use the empiricalBayes estimates for within-occupation earningsdifferences. Results using OLS estimates are inthe same direction (the correlation between thetwo is .54), but the relationship between occupa-tional earnings and within-occupation earningsinequality using OLS estimates is about twice asstrong as what we present here.

ences (plotted along the y-axis) and averageoccupational earnings (plotted along the x-axis) are illustrated in Figure 1. The regres-sion estimate shows a striking relationshipbetween occupational earnings and within-occupation racial earnings differences: Foreach unit increase in the occupational earn-ings of white men, we expect a –.17 unit de-crease in the relative earnings of blacks inthat occupation. In other words, the higherthe average earnings of white men in an oc-cupation, the greater the relative penalty ex-perienced by their black co-workers.26

That blacks in higher-earning occupationsexperience a greater racial penalty than dotheir lower-earning peers reveals an impor-tant and often overlooked source of racialearnings inequality. If we were to constrainthe race effect to be uniform across the oc-cupational distribution (as is conventionalin research using standard analytic tech-niques), we would miss a substantial rangeof variation in within-occupation racial dif-ferences in earnings. While the averageearnings difference between black men andwhite men in the private sector is about 9percent, observed differences vary acrossoccupations from about a 10-percent advan-tage for blacks among clergy to a disadvan-tage of around 22 percent for podiatrists,actuaries, and lawyers. Overlooking thisvariation is particularly consequential forour understanding of black occupationalmobility. Even as black men enjoy higherearnings in an absolute sense as they moveup in the occupational hierarchy, in a rela-tive sense they find themselves ever furtherbehind their white co-workers. This resultreinforces the earlier research of Kaufman

26 The distinction between relative and abso-lute earnings differences is important. If the ab-solute earnings difference were constant acrossoccupations, the relative earnings difference be-tween black men and white men would declineas occupational earnings rose. This is becauseequations for individuals in different occupationshave different intercepts, and in order for the dol-lar amount of a difference to be constant for allmen, the relative difference between men in high-earning occupations would have to be less thanfor men in lower-earning occupations. We findjust the opposite, implying greater racial dispari-ties in higher-earning occupations in both a rela-tive and absolute sense.


Private SectorW

ith

in-O

ccu

pat

ion

Rac

ial E

arn

ing

s D

iffe

ren

ce

Mean Occupational Earnings for Whites1.5 2 2.5 3

–.4

–.2

0

.2

Mean = –.09

(1983) that found that “. . . eliminating un-equal employment opportunities shouldmove blacks into the core [high earnings]sector where they would be facing evengreater wage discrimination” (p. 585).Those occupations with the greatest re-wards are also those in which blacks sufferthe greatest disadvantage relative to theirwhite peers.

In the public sector, we find a different re-lationship. Unlike the private sector whereearnings gains associated with advancementinto higher-earning occupations are in partoffset by the greater relative wage penalty toblacks in such occupations, the public sectordemonstrates no such trend. Figure 2 plotsthe relationship between mean occupationalearnings for whites along the x-axis andwithin-occupation earnings inequality (meanearnings for blacks minus mean earnings forwhites) along the y-axis.27 Although occupa-

tional earnings for whites and within-occu-pation racial earnings inequality are related,this association is weak (with a regressionslope of –.03). Black men working their wayup into higher-paying public-sector jobs,therefore, come closer to achieving earningsparity with white men than they would werethey employed in identical private-sector oc-cupations.

This is not to say that earnings for blacksare equal to earnings for whites across pub-lic-sector occupations. In fact, racial dispari-ties in earnings in the public sector varywidely across occupations. Unlike the pri-vate sector, however, this variation is onlyweakly related to the occupational earningsdistribution and is not always in the direc-tion of black disadvantage. For example,while black public-sector bakers and miscel-laneous woodworking machine operatorssuffer an earnings penalty of more than 30percent, black public textile sewing machineoperators and hand packers and packagers

27 These results correspond to a model inwhich five highly leveraged occupations havebeen deleted. Those occupations are folding ma-chine operators, shaping and joining machine op-erators, crushing and grinding machine operators,dressmakers, and hand packers and packagers.Including these occupations yields a regression

Figure 1. Regression of Within-Occupation Black-White Differences in Earnings on MeanOccupational Earnings for Whites: Private Sector

Note: Models includes no occupation-level predictors.

coefficient –.05 compared with the above esti-mate of –.03. Eliminating highly leveraged occu-pations in the private-sector equation had virtu-ally no effects on our estimates.


enjoy an earnings advantage of nearly 20percent. In fact, fully 18.6 percent of blackmen in the public sector have estimated oc-cupational earnings higher than those of oth-erwise similar white colleagues comparedwith only 2.5 percent of private-sector blackmen. This more randomly distributed racialearnings penalty bodes well for black menworking their way up in the public sector.

Explaining Racial Inequality

in Earnings

The preceding analyses have demonstratedthe importance of the three mechanisms ofracial earnings inequality. In both sectors,we found a substantial impact of the dispro-portionate concentration of blacks in low-paying occupations, and in the private sec-tor we found that the racial gap in earningsgrows wider with average occupational earn-ings for whites. But what factors account forthe differential returns to occupational place-ment? And what explains the remainingrace-by-earnings association? The followinganalysis explores the contributions of occu-pational standing, occupational composition,and occupational skills to the between- and

within-occupation sources of racial earningsinequality.

Our models of inequality in mean occupa-tional earnings for whites are quite success-ful in explaining variation in occupationalearnings and generally support our hypoth-eses (see Table 3). Occupational prestige hasa strong positive effect on occupational earn-ings, while percent black and percent femaleare negative predictors of occupational earn-ings (although the coefficient for percentblack does not reach statistical significance).Likewise, the skill indicators show effects inthe expected direction, with cognitive skilldemands leading to higher average earningsand interpersonal and manual skill require-ments associated with lower occupationalreturns.28

Table 3 also shows estimates for modelspredicting variation in within-occupation ra-

28 Note that the effects of both cognitive andinterpersonal skills on average occupational earn-ings are more than twice as large in the privatesector than in the public sector, suggesting thatremuneration in private-sector occupations ismore closely tied to skill demands relative totheir public-sector counterparts.

Mean Occupational Earnings for Whites

Mean = –.05

1.5 2 2.5 3

–.4

–.2

0

.2

Public Sector

Wit

hin

-Occ

up

atio

n R

acia

l Ear

nin

gs

Dif

fere

nce

Figure 2. Regression of Within-Occupation Black-White Differences in Earnings on MeanOccupational Earnings for Whites: Public Sector

Note: Models includes no occupation-level predictors.


cial earnings inequality. The most strikingresult, in our view, is that few of our occu-pation-level predictors explain variation inthe wage gap between occupations. Al-though the null hypothesis of homogeneityin racial inequality across occupations issoundly rejected in our baseline model andall other models we estimate, our indicatorsof occupational prestige, composition, andskill requirements account for little of thisvariation. In the private sector, only percentfemale is a significant predictor of variationin racial earnings inequality across occupa-tions, with occupations that have high con-centrations of women demonstrating a lowerracial gap in wages. In the public sector, in-terpersonal skills demonstrate a significanteffect: Occupations that emphasize interper-sonal skills appear to be those in whichblack men are relatively closer in earnings

to their white colleagues. If there is a trendto be found in this analysis, it is that blacksfare better relative to their white colleaguesin occupations lower on the status hierarchy(i.e., occupations characterized by high con-centrations of women and/or with a strongemphasis on interpersonal skills). Overall,however, we are unable to explain much ofthe variation in the wage gap across occupa-tions.

The Relationship between the

Occupational Earnings of Whites

and Racial Disadvantage

Does the inclusion of occupation-level pre-dictors help account for the increase in theracial disadvantage at higher levels of occu-pational earnings? Although we expectedthat occupational characteristics would ac-

Table 3. Explaining Between- and Within-Occupation Sources of Racial Earnings Inequality, bySector

Average Within-OccupationOccupational Earnings Racial Earnings Disadvantage

Independent Variable Private Sector Public Sector Private Sector Public Sector

Occupational Status and CompositionIntercept 2.447 2.507 –.090 –.049

(.008) (.008) (.006) (.008)

Occupational prestige/10 .062 .060 –.003 –.003(.011) (.010) (.007) (.010)

Percent black –.213 –.221 –.113 .126(.169) (.168) (.112) (.172)

Percent female –.266 –.249 .048 –.023(.033) (.031) (.023) (.031)

Occupational SkillsCognitive skills .078 .034 .002 –.028

(.025) (.023) (.018) (.024)

Interpersonal skills –.075 –.028 –.019 .039(.019) (.017) (.014) (.018)

Manual skills –.034 –.031 –.002 –.009(.015) (.014) (.010) (.014)

Model FitVariance .027 .021 .005 .012

χ2 40,838 12,825 1,030 1,478

D.f. 444 425 444 425

Reliability .909 .828 .363 .493

Overall deviance .— .— .— .—

r (intercept, black) .— .— –.629 –.093

Note: Numbers in parentheses are standard errors.


count for some of this relationship, we foundinstead that the correlation between occupa-tional earnings and within-occupation earn-ings inequality in the private sector in-creases with the inclusion of occupationalpredictors, from –.55 in the baseline modelto –.63. Figure 3 plots the net relationshipbetween occupational earnings and within-occupation racial earnings inequality esti-mated from a model that includes controlsfor all of our measured individual and occu-pational characteristics. The distributionacross the downward slope is substantiallyless dispersed in the private sector, suggest-ing a strong relationship between occupa-tional earnings for whites and within-occu-pation earnings inequality, net of our ob-served predictors. The increasing black dis-advantage observed higher in the occupa-tional hierarchy is therefore not a functionof occupational status, composition, or skills(despite the fact that these variables predictaverage occupational earnings). Somethingdistinct about the earnings profiles of occu-pations corresponds to the magnitude of ra-cial inequality, apart from the other dimen-sions of occupational characteristics mea-sured here.

In the public sector, including occupa-tional characteristics produces the oppositeeffect, attenuating the residual correlationbetween within- and between-occupationearnings inequality relative to the baselinemodel. Where the relationship betweenearnings inequalities within and betweenoccupations was weak from the start, it be-comes even weaker with the addition of oc-cupation-level predictors. Figure 4 plots therelationship between net occupational earn-ings for whites and net within-occupationracial earnings differences from a model in-cluding all significant predictors reportedabove. After controlling for indicators ofoccupational prestige, composition, andskills, the already modest relationship be-tween racial disparities and occupationalearnings is further attenuated. The regres-sion line, estimated by regressing the re-sidual component of the within-occupationearnings differences on the residual compo-nent of occupational earnings, is virtuallyflat, again indicating the lack of associationbetween these indicators. Unlike the privatesector where racial disparities are system-atically related to the occupational earningsof whites, in the public sector, net of occu-

Net

Wit

hin

-Occ

up

atio

n R

acia

l Ear

nin

gs

Dif

fere

nce

Net Mean Occupational Earnings for Whites–1 –.5 0 .5

–.2

0

.2

.4 Private Sector

Figure 3. Regression of Net Within-Occupation Black-White Differences in Earnings on Net MeanOccupational Earnings for Whites: Private Sector

Note: Models includes all occupation-level predictors.


pational characteristics, we find no suchtrend.

Identifying a Mechanism of Racial

Disadvantage

That average occupational earnings are sys-tematically related to the magnitude of ra-cial disadvantage is a troubling finding withrespect to prospects for upwardly mobileblack men. We tested several plausible ex-planations for this pattern but failed to findsupport for any of our initial hypotheses.What is it about high-earning occupationsthat generates patterns of increasing disad-vantage? What are the attributes of occupa-tions that lead to greater or lesser racial dis-parities?

To answer these questions, we look to thespecific private-sector occupations that dem-onstrate the most and least pronounced ra-cial disparities. By examining the clusters ofoccupations that produce the revealed pat-terns of inequality, we can generate an in-ductive explanation of the attributes of oc-cupations that may yield the observed re-sults. While this qualitative examination isnot conclusive, it should provide a useful set

of hypotheses to be more formally tested infuture research.

Table 4 presents the occupations with thehighest and lowest black-white wage gaps.An interesting pattern emerges: Many of theoccupations with the largest racial gap inwages, such as securities and financial ser-vices, insurance sales, managers in proper-ties and real estate, actuaries, lawyers, andphysicians, are occupations that rely on de-veloping a profitable clientele for success. Ifblacks and whites in the same occupationhave fairly segregated social networks, thenwe would expect whites in these occupationsto benefit from the wealthier pool of poten-tial clients to which they have access. Dif-ferences in the resource base of clients couldtherefore account for the observed disparityin earnings of white and black men in thesame (high-earning) occupations.

Support for this argument is found in theliterature on the occupational mobility ofblack men. Hout (1986), drawing on Lie-berson (1980), develops the concept of“queue jumping” whereby low-status mi-norities gain access to restricted occupa-tions, given a sufficient minority commu-nity size to support such employment. Hout

Net Mean Occupational Earnings for Whites–.5 0 .5

–.4

–.2

0

.2

.4 Public SectorN

et W

ith

in-O

ccu

pat

ion

Rac

ial E

arn

ing

s D

iffe

ren

ce

Figure 4. Regression of Net Within-Occupation Black-White Differences in Earnings on Net MeanOccupational Earnings for Whites: Public Sector

Note: Models includes all occupation-level predictors.


(1986) argues that “ethnic segregation cre-ates ecological niches that tend to be filledby in-group members” such that “a sizeableminority, if sufficiently segregated, can sup-port a number of service professionals, pro-prietors, and tradesmen” (pp. 222, 215).Hout’s emphasis is on occupational oppor-tunities; our extension of this argument isthat while blacks may gain access to theseelite occupations because of the ethnicniche, they are then relegated to a less af-fluent client-base for their services. Blackprofessionals and service providers may

therefore reach nominal parity with whites,but their actual work conditions and re-wards remain far from equal.

These opportunities may likewise beseized by white employers who seek to bestexploit the “black market” by assigningtheir minority employees to serve minoritycommunities. If black real estate agents, forexample, are assigned disproportionately toblack clients and black neighborhoods, thenit follows that their sales commissions willbe significantly lower than otherwise equalwhites (Kiel and Zabel 1996). Evidence ofthis type of employee channeling can befound in the work of Collins (1983 19891993) and Durr and Logan (1997).29

Consistent with this hypothesis, we seethat those occupations with the smallest ra-cial disparities are often those whose salarydepends little on the type of clients served.Upholsterers, bus drivers, hotel clerks, andwoodworking machine operators, for ex-ample, are occupations whose wage ratesare set on the basis of production or laborrather than the demand for service from aparticular clientele. These occupationsleave less room for racial differentiation inearnings, as salaries are determined withoutrespect to the social networks of individualworkers. Of course, the present data do notallow us to conclusively test this argument.Such a test would require data not onlyabout the racial composition of employeeswithin an occupation, but also about the ra-cial composition of those who patronize anoccupation (by race of employee). We do,however, believe that this preliminary in-vestigation reveals an important possiblemechanism of racial stratification not previ-ously examined. We hope that future workwill pay greater attention to sources of oc-cupational differentiation that emerge bothwithin and beyond the workplace. The seg-regated networks of most American workersmay be an important source of earnings dis-

Table 4. Occupations with the Largest andSmallest Racial Gap in Earnings

RacialOccupation Gap

Largest Racial Wage GapSecurities and financial

services sales occupations .722

Podiatrists .771

Insurance sales occupations .778

Longshore equipment operators .786

Lawyers .789

Dentists .798

Managers and administrators, n.e.c. .799

Physicians .803

Actuaries .806

Stevedores .808

Managers, properties and real estate .813

Smallest Racial Wage GapUpholsterers 1.017

Cooks, private household 1.020

Religious workers, n.e.c. 1.025

Taxicab drivers and chauffeurs 1.026

Miscellaneous woodworkingmachine operators 1.031

Teachers, prekindergartenand kindergarten 1.034

Hotel clerks 1.064

Bus drivers 1.071

Family child care providers 1.096

Child care workers, private household 1.107

Clergy 1.134

Note: The race gap is defined as the mean earn-ings of blacks in an occupation divided by the meanearnings of whites in that same occupation.

29 This explanation is also consistent with re-cent allegations made by Frank Warren et. al.against Xerox Corp (Frank Warren et al. v.Xerox Corp., No. 01-CV-2909, E.D. N.Y.). War-ren et al. have alleged that African-American em-ployees “were assigned to sales territories in low-income and minority neighborhoods” and as a re-sult made significantly less money than theirwhite counterparts (“Xerox . . .” 2001).


parities by race, not only in terms of gain-ing access to elite occupations (Mouw2000), but also in terms of profiting fromone’s labor once there.

DISCUSSION

We have examined the individual and occu-pational characteristics that are associatedwith earnings inequality between black menand white men. Our findings show that justover one-half of the racial gap in earningscan be accounted for by variation in indi-vidual attributes such as human capital, re-gion, and marital status. An additional 20percent of the race gap in earnings is due tothe differential placement of blacks andwhites across the occupational distribu-tion—blacks are concentrated in occupationswith low average earnings, even after con-trolling for individual characteristics. In gen-eral, these lower-paying occupations arecharacterized by low prestige, few hard skillrequirements, an emphasis on soft skills, andhigh proportions of female and black incum-bents.

The extent to which the racial gap in earn-ings is a reflection of individual differencesand occupational segregation is not surpris-ing. Previous research has extensively docu-mented the effects of these variables; suchresearch commonly finds that racial dispari-ties in earnings persist even after accountingfor these factors.

Our study extends prior work on earningsdisparities by race by concentrating onvariation in the effects of race on earningsacross the occupational distribution and theextent to which occupational measures canexplain that variation. While most analysesassume the race gap to be constant for all oc-cupations, our empirical tests lead us to re-ject this assumption. There is significantvariation in the magnitude of racial earningsinequality across occupations in both thepublic and private sectors, even after con-trolling for a host of individual attributes.Recognizing variation in the degree of racialdisparity that emerges at different points inthe occupational structure is critical to gain-ing a comprehensive understanding of theblack-white gap in wages.

For the 23 percent of black men em-ployed in the public sector, we find encour-

aging evidence that occupations confer theirrewards primarily on the basis of individualqualifications, with largely random varia-tion in the magnitude and direction of racialwage inequality. For the 77 percent of blackmen employed in the private sector, how-ever, we are less confident that meritocracyis the driving force behind wage allocation.The strong and systematic relationship be-tween occupational earnings for whites andracial disparities in earnings suggests thatrace remains a salient feature in the occupa-tional hierarchy of the private sector. High-earning private-sector occupations are char-acterized by greater racial inequalities inearnings, tempering the rewards for occupa-tional advancement and widening the gulfbetween high-achieving black and whiteemployees. We were surprised to find thatoccupational standing, composition, andskill requirements were unable to accountfor even part of this relationship. To whatthen do we attribute our findings for the pri-vate sector?

We first consider the possibility of omit-ted-variable bias. Recent research arguesthat previous measures of human capitalhave failed to fully capture the skill differ-entials between blacks and whites andtherefore have overstated the effects of dis-crimination. Several researchers have foundthat including direct measures of cognitiveability (using the Armed Forces QualifyingExam) can substantially (though not fully)attenuate the racial differences in standardearnings equations that remain after con-trolling for education and other related fac-tors (Farkas and Vicknair 1996; Neal andJohnson 1996; O’Neill 1990). Particularlyamong college graduates and women, dif-ferences in skills can explain nearly all ofthe wage gap between blacks and whites.While some of these findings seem compel-ling, they have not gone unchallenged.Spurred on by Herrnstein and Murray’s(1994) The Bell Curve, others have pro-vided evidence that suggests that, even netof substantial controls for both backgroundand ability, racial differences in earnings re-main large and statistically significant(Raudenbush and Kasim 1998). Further-more, recent work by Cawley et al. (1996),Ashenfelter and Rouse (1999), and Cardand Limieux (1994) casts doubt on the as-


sertion that unmeasured skills exert anysubstantial bias on earnings equations thatinclude educational attainment, or that thereturns to educational attainment are biaseddue to unmeasured skills.

More to the point, Raudenbush and Kasim(1998) directly test the hypothesis thatblack-white skill differences are responsiblefor variation in the within-occupation racialgap in earnings. If the racial gap in skills(e.g., the difference in average cognitiveability between blacks and whites) also var-ies according to occupational earnings suchthat the black-white skill gap is larger inhigh-paying occupations than in low-payingoccupations, then variation in racial earningsinequality may be an artifact of actual skilldifferences between black and white work-ers within the same occupation. However, ina model that includes individual- and occu-pation-level indicators of literacy skills (ameasure highly correlated with conventionalmeasures of cognitive ability), Raudenbushand Kasim find that a substantial portion ofthe occupational variation in black-whiteearnings inequality remains unexplained.There is little indication from previous re-search, therefore, that unmeasured skills arethe driving force behind our findings.

We have further explored the possibilitythat unmeasured skill differences are affect-ing our results through two indirect tests.First, if skill differences were the drivingforce behind racial wage inequalities, wewould expect at least some of this effect tobe picked up by our measures of occupa-tional skill requirements. Given that indi-vidual skill and occupational skill require-ments are likely correlated, in the absenceof a direct measure of individual skill theoccupational variable should provide amodest (if noisy) proxy. Our results do notsupport this argument—while occupationalskills are a strong indicator of occupationalearnings overall, they explain none of thevariation in racial earnings inequalities.30

Second, we have tested for the presence ofa race-by-education interaction in our indi-vidual-level model. If black-white skill gapsare greater at higher levels of educational at-tainment, we would expect the race-by-edu-cation interaction to be negative and increas-ing in magnitude across levels of educationalattainment. If this interaction term addedsubstantial explanatory power to our model,we would also expect a lower adjusted meanlevel of racial earnings inequality and morerestricted variance in racial inequality acrossoccupations. None of these results obtains,further weakening concerns over the poten-tial skill bias. However, future researchshould pursue this line of inquiry using di-rect measures of cognitive skill as an indi-vidual-level attribute.

A second possible source of spuriousnessemerges from the pattern of variation inearnings across the occupational distribu-tion. For example, if there is greater varia-tion in log occupational earnings in high-earning occupations compared with low-earning occupations, then the observed pat-tern of racial disparities in earnings maymerely reflect greater earnings variationoverall. While plausible, in this case a directtest bears evidence to the contrary. We ex-amined the association between the varianceand mean of the log of occupational earningsand did not find a positive relationship. Infact, we find that the relationship betweenearnings variance and average occupationalearnings is relatively flat across the occupa-tional distribution. Thus, we can safely re-ject the concern that general patterns of vari-ance in occupational earnings are driving ourreported results.

Having investigated the most plausiblesources of spuriousness, we conclude thatthe relationship between the average earn-ings of an occupation and the magnitude ofracial disparities must be explained on moresubstantive grounds. Our exploratory analy-sis demonstrated a tendency for occupationswith large racial disparities to be client-based professions that rely on social net-works for success. We believe that segre-gated social networks combined with thedisparity in assets possessed by blacks andwhites may be an important source of earn-ings inequality for blacks and whites in thesame occupation. While we do not discount

30 Depending on model specification, cognitiveand interpersonal skills are sometimes significantpredictors of the race gap in the public sector. Inthe private sector, however, where the relation-ship between occupational earnings and racialdisparities is found, occupational skills neverdemonstrate a significant effect.


the influence of within-firm processes—par-ticularly the role of direct race discrimina-tion—we think that social forms of segrega-tion may contribute to the perpetuation ofracial disadvantage among working blackmen.

CONCLUSION

We began this investigation by discussingthe apparent contradiction between recenttrends in occupational mobility and earningsinequality. Our analysis of labor market dis-parities that takes into account the relation-ship between occupational placement andracial earnings inequality reveals one mech-anism that may underlie these opposingtrends. We find that, far from representingindependent processes, occupational mobil-ity and earnings inequality are intimatelylinked such that movement into higher-earn-ing occupations (declining occupational seg-regation) is associated with greater within-occupation wage disparities (increasing ra-cial wage inequality) for private-sectorworkers.

Contrary to theories that predict greaterrationalization and meritocracy in high-pro-file occupations, our results suggest that thevertical differentiation of occupations in theprivate sector is directly associated with themagnitude of observed racial disparities. Asblack men gain entry to the most highlycompensated occupational positions, they si-multaneously become subject to more ex-

treme racial disadvantage. Although we can-not conclusively explain this relationship,we believe an important component may bethe social segregation of black and whiteprofessionals. We hope that future researchwill investigate this claim more thoroughly.If we want to pursue policies which advancethe goal of racial earnings equality, we mustgain a better understanding of the occupa-tional processes which drive the persistentearnings disadvantage experienced by blackmen.

Eric Grodsky is a Ph.D candidate in Sociologyat the University of Wisconsin–Madison. His dis-sertation extends the thesis of maximally main-tained inequality to explore stratification in thepostsecondary opportunities of students in theU.S. across three decades. He is also involved intwo international comparative projects on strati-fication in higher education. In addition to thesociology of education, his research interests in-clude stratification, quantitative methodology,and race and ethnicity.

Devah Pager is a Ph.D candidate in Sociologyat the University of Wisconsin–Madison. Her re-search focuses on racial stratification in educa-tion, employment, and the criminal justice sys-tem. She has an article forthcoming (with LincolnQuillian) in the American Journal of Sociologythat investigates the effect of neighborhood ra-cial composition on perceptions of neighborhoodcrime. She is currently completing her disserta-tion, which is an audit study of the effect of acriminal record on employment opportunities forblack and white job-seekers.

Appendix A. Selection Procedures for Private and Public-Sector Samples: PUMS, 1990


Percent of Percent ofCriterion for Selection Unweighted N Sample Retained Unweighted N Sample Retained

Unconstrained 975,335 100.0 418,904 100.0

Excluding those notAfrican American,white, Hispanic or Asian 953,202 97.7 412,734 98.5

Excluding unemployed 831,526 85.3 390,348 93.2

Excluding earnings < 0 822,631 84.3 388,111 92.6

Excluding military 822,631 84.3 378,260 90.3

Excluding farm workers 781,457 80.1 372,953 89.0

Excluding those in occupationslacking DOT measures 780,236 80.0 372,543 88.9


REFERENCES

Ashenfelter, Orley and Cecilia Rouse. 1999.“Schooling, Intelligence, and Income inAmerica: Cracks in the Bell Curve.” Workingpaper 6902, National Bureau of Economic Re-search, Cambridge, MA.

Beck, Elwood M., Patrick M. Horan, and CharlesM. Tolbert II. 1978. “Stratification in a DualEconomy: A Sectoral Model of Earnings De-termination.” American Sociological Review49:454–73.

Bound, John and Richard Freeman. 1992. “WhatWent Wrong? The Erosion of Relative Earn-ings and Employment among Young BlackMen in the 1980s.” Quarterly Journal of Eco-nomics 107:201–32.

Braddock, Jomills Henry and James M.McPartland. 1987. “How Minorities Continueto Be Excluded from Equal Employment Op-portunities: Research on Labor Market and In-stitutional Barriers.” Journal of Social Issues43:1–78.

Bridges, William and Robert Nelson. 1989.“Markets in Hierarchies: Organizational andMarket Influences on Gender Inequality in aState Pay System.” American Journal of Soci-ology 3:616–58.

Cain, Glen. 1976. “The Challenge of SegmentedLabor Market Theories to Orthodox Theory.”Journal of Economic Literature 14:1215–57.

———.1986. “The Economic Analysis of LaborMarket Discrimination: A Survey.” Pp. 693–785 in Handbook of Labor Economics, editedby O. Ashenfelter and P. Layard. Amsterdam,Netherlands and New York: North-Holland.

Cain, Pamela S. and Donald J. Treiman. 1981.“The Dictionary of Occupational Titles as aSource of Occupational Data.” American So-ciological Review 45:253–78.

Card, David and Thomas Lemieux. 1994.“Changing Wage Structure and Black-WhiteWage Differentials.” American Economic Re-view 84:29–33.

Cawley, John, Karen Connelly, James Heckman,and Edward Vytacil. 1996. “Measuring the Ef-fects of Cognitive Ability.” Working paper5645, National Bureau of Economic Research,Cambridge, MA.

Collins, Sharon. 1983. “The Making of the BlackMiddle Class.” Social Problems 30:369–81.

———. 1989. “The Marginalization of Black Ex-ecutives.” Social Problems 36:317–31.

———. 1993. “Blacks on the Bubble: The Vul-nerability of Black Executives in White Cor-porations.” Sociological Quarterly 34:429–48.

Cotton, Jeremiah. 1989. “Opening the Gap: TheDecline in Black Economic Indicators in the1980s.” Social Science Quarterly 70:803–19.

Dickens, William T. and Kevin Lang. 1985. “A

Test of Dual Labor Market Theory.” AmericanEconomic Review 75:792–805.

DiPrete, Thomas and Whitman Soule. 1986. “TheOrganization of Career Lines: Equal Employ-ment Opportunity and Status Advancement ina Federal Bureaucracy.” American Sociologi-cal Review 51:295–309.

Doeringer, Peter and Piore, Michael. 1971. Inter-nal Labor Markets and Manpower Analysis.Lexington, KY: Lexington Books.

Duncan, Otis Dudley. 1961. “A SocioeconomicIndex for All Occupations.” Pp. 109–38 in Oc-cupations and Social Status, edited by A. J.Reiss. New York: Free.

Durr, Marlese and John R. Logan. 1997. “RacialSubmarkets in Government Employment: Af-rican American Managers in New York State.”Sociological Forum 12:353–70.

Ehrenberg, Ronald, and Joshua Schwarz. 1986.“Public-Sector Labor Markets.” In Pp. 1219–68 in Handbook of Labor Economics, edited byO. Ashenfelter and R. Layard. Amsterdam,Netherlands and New York: North-Holland.

England, Paula, George Farkas, BarbaraKilbourne, and Thomas Dou. 1988. “Explain-ing Occupational Sex Segregation and Wages:Findings from a Model with Fixed Effects.”American Sociological Review 53:544–59.

England, Paula and Barbara Kilbourne. 1988.Occupational Measures from the Dictionary ofOccupational Titles for 1980 Census DetailedOccupations [MRDF]. Ann Arbor, MI: Inter-university Consortium for Political and SocialResearch [distributor].

Farkas, George and Kevin Vicknair. 1996. “Com-ment: The Need to Control for Cognitive Skill.”American Sociological Review 61:557–64.

Farley, Reynolds. 1996. The New American Re-ality: Who We Are, How We Got Here, andWhere We Are Going. New York: Russell SageFoundation.

Featherman, David L. and Robert M. Hauser.1978. Opportunity and Change. New York:Academic Press.

Freeman, Richard. 1996. “When Earnings Di-verge: Causes, Consequences, and Cures to theNew Inequality in the U.S.” Harvard Univer-sity and National Bureau of Economic Re-search, Cambridge, MA. Unpublished manu-script.

Grandjean, Burke. 1981. “History and Career ina Bureaucratic Labor Market.” American Jour-nal of Sociology 86:1057–91.

Grusky, David and Aage Sørensen. 1998. “CanClass Analysis Be Salvaged?” American Jour-nal of Sociology 103:1187–234.

Harrison, Roderick J. and Claudette E. Bennett.1995. “Racial and Ethnic Diversity.” Pp. 141–210 in State of the Union: America in the1990s, vol. 2, Social Trends, edited by R.


Farley. New York: Russell Sage Foundation.Hauser, Robert M. 1980. “Comment on Beck,

Horan, and Tolbert, ‘On Stratification in aDual Economy.’” American Sociological Re-view 45:702–12.

Hauser, Robert M. and John Robert Warren.1997. “Socioeconomic Indexes of Occupa-tional Status: A Review, Update, and Cri-tique.” Sociological Methodology 27:17–298.

Herrnstein, Richard J. and Charles Murray. 1994.The Bell Curve: Intelligence and Class Struc-ture in American Life. New York: Free Press.

Hodson, Randy. 1985. “Some ConsiderationsConcerning the Functional Form of Earnings.”Social Science Research 14:374–94.

Hodson, Randy and Robert L. Kaufman. 1982.“Economic Dualism: A Critical Review.”American Sociological Review 47:727–39.

Hout, Michael. 1984. “Occupational Mobility ofBlack Men, 1962 to 1973.” American Socio-logical Review 49:308–22.

———. 1986. “Opportunity and the MinorityMiddle Class: A Comparison of Blacks in theUnited States and Catholics in Northern Ire-land.” American Sociological Review 51:214–23.

Jencks, Christopher and Meredith Phillips, eds.1998. The Black-White Test Score Gap. Wash-ington, DC: Brookings Institution Press.

Kaufman, Robert. 1983. “A Structural Decompo-sition of Black-White Earnings Differentials.”American Journal of Sociology 89:585–611.

Kiel, Katherine and Jeffrey Zabel. 1996. “HousePrice Differential in U.S. Cities: Householdand Neighborhood Racial Effects.” Journal ofHousing Economics 5:143–65.

Kilbourne, Barbara, Paula England, and KurtBeron. 1994. “Effects of Individual Character-istics on Earnings: Intersections of Race andGender.” Social Forces 72:1149–76.

King, Mary. 1992. “Occupational Segregation byRace and Sex, 1940–88.” Monthly Labor Re-view 115:30–37.

Kohn, Melvin. 1977. Class and conformity: AStudy in Values, with a Reassessment. Chi-cago, IL: University of Chicago Press.

Korenman, Sanders and David Neumark. 1990.“Does Marriage Really Make Men More Pro-ductive?” Journal of Human Resources 26:282–307.

Krislov, Samuel. 1967. The Negro in FederalEmployment. Minneapolis, MN: University ofMinnesota Press.

Lieberson, Stanley. 1980. A Piece of the Pie:Blacks and White Immigrants Since 1880. Ber-keley, CA: University of California Press.

Mare, Robert D. 1995. “Changes in EducationalAttainment and School Enrollment.” Pp. 155–213 in State of the Union: America in the1990s, vol. 1, Economic Trends, edited by R.

Farley. New York: Russell Sage Foundation.Model, Suzanne. 1997. “An Occupational Tale of

Two Cities: Minorities in London and NewYork.” Demography 34:539–50.

Moss, Philip and Chris Tilly. 1996. “ ‘Soft’ Skillsand Race: An Investigation of Black Men’sEmployment Problems.” Work and Occupa-tions 23:252–76.

Moulton, Brent. 1990. “A Reexamination of theFederal-Private Wage Differential in theUnited States.” Journal of Labor Economics8:270–93.

Mouw, Theodore. 2000. Four Essays on the So-cial Structure of Urban Labor Markets. De-partment of Sociology, University of Michi-gan, Ann Arbor, MI.

Murnane, Richard, John Willett, and Frank Levy.1995. “The Growing Importance of CognitiveSkills in Wage Determination.” Review of Eco-nomics and Statistics 77:251–66.

Murphy, Kevin and Finis Welch. 1994. “Indus-trial Change and the Rising Importance ofSkill.” Pp. 101–32 in Uneven Tides: Rising In-equality in America, edited by S. Danziger andP. Gottschalk. New York: Russell Sage Foun-dation.

Nakao, Keiko and Judith Treas. 1994. “UpdatingOccupational Prestige and SocioeconomicScores: How the New Measures Measure Up.”Sociological Methodology 24:1–72.

Neal, Derek and William Johnson. 1996. “TheRole of Premarket Factors in Black-WhiteWage Differences.” Journal of PoliticalEconomy, 104:869–95.

O’Neill, June. 1990. “The Role of Human Capi-tal in Earnings Differences between White andBlack Men.” Journal of Economic Perspec-tives 4:25–45.

Parcel, Toby L. and Charles W. Mueller. 1983.Ascription and Labor Markets: Race and SexDifferences in Earnings. New York: Aca-demic.

Peterson, Trond. 1999. “Functional Form forContinuous Dependent Variables: Raw VersusLogged Form.” Paper presented at the AnnualMeetings of the American Sociological Asso-ciation, August, Chicago, IL.

Raudenbush, Stephen and Rafa Kasim. 1998.“Cognitive Skill and Economic Inequality:Findings from the National Adult Literacy Sur-vey.” Harvard Educational Review 68:33–79.

Sakamoto, Arthur and Meichu D. Chen. 1991.“Inequality and Attainment in a Dual LaborMarket.” American Sociological Review 56:295–308.

Sewell, William H. and Robert M. Hauser. 1975.Education, Occupation, and Earnings:Achievement in the Early Career. New York:Academic Press.

Sørensen, Aage. 1996. “The Structural Basis of


Social Inequality.” American Journal of Soci-ology 101:1333–65.

Spenner, Kenneth I. 1983. “Deciphering Pro-metheus: Temporal Change in the Skill Levelof Work.” American Sociological Review48:824–37.

Stolzenberg, Ross. 1975. “Education, Occupa-tion, and Wage Differences between White andBlack Men.” American Journal of Sociology81:299–323.

Telles, Edward E. 1994. “Industrialization andRacial Inequality in Employment: The Brazil-ian Example.” American Sociological Review59:46–63.

Tienda, Marta and Ding-Tzann Lii. 1987. “Mi-nority Concentration and Earnings Inequality:Blacks, Hispanics, and Asians Compared.”

American Journal of Sociology, 93:141–65.Tomaskovic-Devey, Donald. 1993. Gender and

Racial Inequality at Work: The Sources andConsequences of Job Segregation. Ithaca, NY:ILR Press.

Tomaskovic-Devey, Donald and Sheryl Skaggs.1999. “An Establishment-Level Test of theStatistical Discrimination Hypothesis.” Workand Occupations 26:420–43.

Waldinger, Roger. 1989. “Immigration and Ur-ban Change.” Annual Review of Sociology15:211–32.

“Xerox Sales Employees File Racial Discrimina-tion Lawsuit.” 2001. Bloomsberg News, May10. Retrieved July 11, 2001 (http://news.cnet.com/investor/news/newsitem/0-9900-1028-5889757-0.html).

542 american sociological review the structure of ......tions of these results are considered and...

Documents