Discrimination Over the Life Course: A Synthetic Cohort Analysis of Earnings DifferencesBetween Black and White Males, 1940-1990* MELVIN E. THOMAS, LehighUniversity CEDRIC HERRING, University ofIllinoisat Chicago HAYWARD DERRICK HORTON, Iowa StateUniversity United States Microdata Extract FileandtheCurrent SurPopulation Usingdatafromthe1940-1980 examines three in File,thisstudy competing explanations ofthedisparity vey's1990AnnualDemographic overthelifecourse.The"legacy that blackandwhite maleearnings ofdiscrimination" explanation suggests in earnings olderblacks, and morethanpastdiscrimination racialdisparities current reflect against nothing blackandwhitemalesshouldbesimilar overthelifecourse.The"cumulative thattheearnings ofyounger thattheblack-white overthelifecourse, earnings gapincreases explanation suggests effects ofdiscrimination" The"vintage inearnings exists cohorts andthatthisdivergence ofmalesas wellas oldercohorts. foryounger in self-investments in human earnings gap reflects differences arguesthatthenetblack-white hypothesis" constant overtimeandoverthelifecourse forall capitalandthattheracialearnings gapshouldbevirtually a synthetic cohort Thisstudy maleworkers. cohorts presents analysis oftheeffects (vintages) ofblackandwhite malesfrom1940-1990.Itshowsthataginghasa in earnings forblackandwhite ofagingon thedisparity were blackmalesineachyearandcohort ontheblack-white curvilinear analyzed earnings gap. Younger effect blacks.However, therewasa convergence in the thanmiddle-aged closertotheirwhitemalecounterparts withanyoftheformulations, the males.While notcompletely consistent blackandwhite earnings ofelderly tothepredictions mostclosely ofdiscrimination explanation. ofthecumulative effect conform findings Introduction one of the mostdisadvantaged Black Americanscontinueto represent groupsin the in UnitedStates.Despitethewidelyheldviewthattheyhave shownsteadyimprovements ofthemeatheirearningsrelativeto thoseofwhites,recentstudiesreportthatirrespective blackshaveyetto achieveparitywithwhites sureor standardofsocioeconomic well-being, ofpresentblackdisadvantage suggestthatit (Jaynesand Williams1989). Someexplanations is nothingmorethanthe"legacyofpastdiscrimination" (e.g.,Wilson1978). Othersexplain in earningsbetweencohortsthatcan be racialinequalitiesin termsof"vintage"differences and humancapitalprofiles in racial differences (e.g., experience group-specific explainedby oftheblack-white Welch1973). Stillothersproposean explanation earningsgap thatsugofdiseffects" becauseofthe"cumulative have notdisappeared geststhatracialdifferences Thomas crimination 1993). againstblackworkersoverthelifecourse(e.g., on a draft comments * Wewouldliketo thankBillBridges, Jonathan Crane,andMelvinOliverfortheirhelpful ofSociology and Anthropology, to MelvinE. Thomas,Department Lehigh ofthispaper. Addressall correspondence PA 18015. 681 TaylorSt.,Bethlehem, University, 608 1994 Vol.41,No.4, November SOCIALPROBLEMS, Discrimination OvertheLifeCourse the"vintagehypotheclaimsofthe"legacyofpastdiscrimination," To testthecompeting ofdiscrimination" ofracialdisparities in earneffects sis,"and the"cumulative explanations the how of have affected of race and a cross-time earnings blacks age analysis ingsrequires ofage,timeperiod,and thecohort and whitesovertheyears.Thisrequiresa disentangling cohortsofblackand whiteworkers. effects ofraceon earningsfordifferent racialdisparities in earnings In thisstudy,we providesuch an analysisof age-related usingdata spanningthesix decadesfrom1940 to 1990. The studyhas twoparts.The first of racialdifferences examination by age in 10 yearintervalsfrom partis a cross-sectional thenatureoftherace-age-period interaction. In the 1940 to 1990. We use thisto illustrate secondpartofthestudy,we providean analysisoffiveage cohorts.We do thisto provide insightinto the natureof changesin racialinequalityin earningsover timeand the life course. samThe sampleswe use arenotfroma panelstudy.Theyare,however,representative pointsin time;thus,theyprovidean appropriate ples of the same populationat different ofthestatusofeachcohortat eachpointin time,and theyareappropriate forour estimation In cohort both of the we examine differences besynthetic analysis. parts analysis, earnings tween black and white males while applyingstatisticalcontrolsfor relevantsocicharacteristics. odemographic Models of Black-WhiteEarningsOver the Life Course The LegacyofPast Discrimination The legacyof past discrimination explanationforcurrentblack-white earningsdifferofWilliamJuliusWilson'sbook,TheDeences gainedprominence throughthepublication clining Significance ofRace(1978, 1980). In thiswork,Wilson(1980) arguedthatclasshad factorexplaining thesituationofblacksin America. supersededraceas themostimportant He contendsthatin themodernindustrial and stateintervention in period,economicgrowth theformofcivilrights administrative actiontoimprovecivilrights, and theimplelegislation, mentationof affirmative actionpolicieshave createda largeblackmiddleclass primarily and theremovalofracialbarriers in educationand throughthecreationofjob opportunities in government and corporateemployment. At thesame time,segmentation oflabormarout ofcentralcities,and thedeclinein theproduction of kets,the movementofindustries of relative to the services have a social created and economic situation that goods production blackpoverty. perpetuates These economicfactors, accordingto Wilson,have createda largeand growingblack underclassthatlacksthe cultural,social,and economicresourcesto takeadvantageof the new opportunities Wilson(1980, 1987)argues,is not discrimination, (Wilson1987). Current the primary cause of theproblemof theblackunderclass.Pastdiscrimination createdthe blackunderclasswhichcontinuestodayprimarily becauseofeconomicand othernon-racial factors.Moreover,Wilson(1978:17) contends: Thereisstilla significant all college andallcollege income educated whites educated gapbetween blacksbecauseofthesubstantially lowerincome ofoldereducated blacks.Butis thismainly a . . .? No, the comparatively discrimination low incomesof older consequenceof present-day blacksis one ofthelegaciesofpastdiscrimination. Thus for Wilson and other proponentsof the legacy of past discrimination,differencesin measures of well-beingacross social classes forblacks and whitesare attributableprimarilyto the low life chances of older blacks who were victimsof past discrimination.Thus, froma cohort analysis framework,the legacy of past discriminationperspectiveparallels a period effects-driven modelofchangein racialdifferences in earnings. 609 610 THOMAS/HERRING/HORTON The Cumulative Effects ofDiscrimination In answeringWilson's (1978:17) contentionthat"thereis stillan income gap between all college educated blacks and all college educated whites because of the substantiallylower income of older educated blacks . . . because of a legacy of past discrimination,"Willie (1979:55) utilizesstatisticsprovidedby the U.S. CivilRightsCommissiontitledSocialIndicators and Women.This studypresentsstatisticaladjustmentsusing multiple ofEqualityforMinorities for each regression minoritygroups' educational level,job prestige,income level in the state of residence,weeks worked,hours recentlyworked,and age. Accordingto Willie (1979:55), based on this study, one mayconcludethatevenwhenall thingsare equal (including age,educationand occupation), receivean annualincomethatis 15to20 percent blacksandotherminorities lessthanthatreceived males.Thesefindings indicatethatracismis aliveand wellin theUnitedStates. bymajority Morerecentstudieshavealso challenged the"legacyofpastdiscrimination" explanation of earningsdifferences betweenblacksand whites. Cotton(1990), Thomas (1993), and an alternative Thomasand Horton(1992) have offered forthenarrowing ofthe explanation racialgap fortheyoungeragedgroup.Theyproposethatyoungerblackshave alwaysdone betterthan older blacks when compared with similarwhitesbecause the negative impactof discriminationis cumulativeover the lifecourse. Thomas (1993) found thatyoungerblacks werebetteroffcomparedwithsimilarwhitesthanolderblacksin both1968 and 1988. He is relatively smallfor youngerblacks, suggeststhat"thenegativeimpactof discrimination and olderblacks"(1993:613). butovertimeincreasesand becomeslargeformiddle-aged in familyincome, Thomasand Horton(1992), in a similarstudyof racialdifferences withyoungerreference on family incomewas smallerforfamilies foundthattheraceeffect persons than those with older referencepersons. They conclude that ratherthan reflecting the data suggestthatthe effectsof race increased over the the effectsof past discrimination, thatlabormarket lifecourse. Cotton'sanalysisofblack-white earningsdifferences suggests as blacksand womenwithexperienceand discrimination increasesovertime,"particularly slotsin the higherreachesof becomeeligibleforthepolicyand decision-making seniority businessand theprofessions" (1990:25). abouta color-blind ofthecumulaSo despiteoptimistic society, proponents projections modelwould suggestthatblacksin each era are confronted tiveeffects ofdiscrimination by discrimination,the effectsof which accumulate over the lifecourse. It is theoreticallypossiwill cease to existunder periodsof economic ble thatthe cumulativeeffectsof discrimination discrimination. The empiricalrealityaccordingto racial and and declining prejudice growth the cumulativeeffectsof discrimination model, however,is thatblacks in Americahave hisin which temporary toricallybeen facedwithseemingprogressand subsequentretrenchment economic gains have been misappropriated.Thus, froma cohort analysis framework,the model cumulativeeffectsof discriminationexplanationis analogous to an age effects-driven in earnings. of change in racial differences The VintageHypothesis Anotherexplanationofthepresentgap in earningsbetweenblacksand whitesis what has become known as the "vintage hypothesis" (e.g., Welch 1973, 1975, 1980; Link, Ratledge,and Lewis 1976, 1980; Smithand Welch 1977; Hoffman1979; Akin and Garfinkel 1980; and Duncan and Hoffman1983). Accordingto thisview, labor marketearningsreflect levels of productivityforindividualsat different stages of theirlives. Differencesin producin human capital such as education in self-investments tivityin turnresultfromdifferences and on-the-jobtraining. Discrimination OvertheLifeCourse in thequality ofthisperspective, Welch(1973), theoriginator suggeststhatdifferences ofeducationhaveplayeda majorrolein theobservedgapbetweenblacksand whites.Thus, to schooling attendedlowerqualityschools,incomereturns becauseblackshave historically forblackshave been lessthanforwhites.'Recently, however,theblack-white earningsgap has begunto close because the returnsto educationhave risenfasterforblacksthanfor whitesas thequalityofeducationforblackshas improvedformorerecentcohorts.Among to schooling have begunto be as high thosewithat leastsomecollege,forexample,returns of forblacksas theyare forwhites(Welch1973; Smithand Welch1977). Thus,proponents in earnings: racialvariations thevintagehypothesis arguethattherearecohort-differentiated to theireduOldervintagesofblackswho receivedpoorereducationsreceivedlowerreturns Theseracialdifferences thantheirwhitecounterparts. have beencomcationalinvestments ofearningsoverthelifecycle.Newer poundedovertheyearsand have led to a divergence to their vintagesofmoreequallyeducatedyoungblacksexperiencesimilarearningsprofiles in theearnings ofblacksand whitesover Thereshouldbe no divergence whitecounterparts. for shouldbe stronger theyearsforthesenewercohorts.Moreover, vintage(cohort)effects blacksthantheyare forwhites. in earnings lifecycledifferences The important questionswithrespectto black-white forblacksand whites,whether includewhetherlabormarketdynamicsoperatedifferently overtheyears,and whethersuchracialdifferences havebeenconstant suchracialdifferences thevintagehypothesis thatthelabormarare constantoverthelifecourse.Clearly, specifies ketnow operatesthesameforblacksand whites,thatthissimilaroperationis a changefrom and thatnow thatdiscrimination has discrimination, previouseraswhenblacksexperienced been diminished youngerblacksshouldexperiencethe same labormarketdynamicsover the vintage theirlives thatyoungerwhitesdo. Thus,froma cohortanalysisframework, modelof changein racialdifferences in is equivalentto a cohorteffects-driven hypothesis earnings. SomeCompeting Hypotheses The threemodelsof black-white overthelifecoursemake similar earningsdisparities ofyoungblacksrelativeto youngwhitesand olderblacks aboutthewell-being predictions relativeto olderwhites:The earningsgap shouldbe smallestfortheyoungestworkersand netofrelevantworkcharacteristics. Thesemodelsdiffer, howlargestfortheoldestworkers, in thegapsovertimeand overthelife aboutthenatureofchange ever,in theirpredictions modelpredicts thatnetof course.As DiagramslA and IB show,thelegacyofdiscrimination should relevantcharacteristics, theblack-white become smaller over time, earnings gap esperacialdiscrimination), butanysuchdissimilarities ciallyforyoungercohorts(due todeclining in earningsshouldremainrelatively constantoverthelifecourseforyounger(post-1960s) is themostsignificant its cohorts.In otherwords,ifthelegacyofpastdiscrimination factor, out behindtheir shouldbe observedovertimein the formof olderblacksstarting effects whitecounterparts beduringtheperiodbeforetheCivilRightsMovementand remaining hindthroughthepresent, and youngerblacksstarting out on moreequal footing withtheir theirlifetimes. whitecounterparts and maintaining thatparitythroughout ofdiscrimination thepredictions ofthecumulative effects model. Diagram2 illustrates thatnetofrelevantcharacteristics, Thismodelpredicts gapsin earningsat anypointin time buttheblack-white shouldreflect levelsofdiscrimination againstblackworkers, earnings gap shouldalso becomeprogressively largerforeach cohortas theyprogressthroughthe life to argue that differencesin the qualityof educa1. As one of the anonymous reviewerspoints out, it is difficult tion, per se, are reflectionsof self-investments. Rather,it is perhaps more accurate to say that such differencesreflect social investmentsin individuals. Still,the centralpoint is that accordingto the vintage hypothesis,as education becomes more equal, earningsshould also become more equal. 611 612 THOMAS/HERRING/HORTON Legacy of Past Discrimination Overall Pattern r- C 1940 1960 1950 - Whites 1980 1970 - 1990 Blacks Diagram la Legacy of Past Discrimination Young Cohorts Only -E Cn 1940 ?-, low d~op dso 1960 1950 - Diagram lb d~op 400p 10?, 400p 400 1980 1970 -IBlacks SWhites Whites - Blacks 1990 OvertheLifeCourse Discrimination Cumulative Effects of Discrimination ogo r-p ._g --- 1 .L Whites Blacks g 20-29 30-39 40-49 50-59 60-69 Age Diagram2 ofdiscrimination is accurate,the effects explanation cycle.In otherwords,ifthecumulative of discrimination shouldincreaseoverthe lifecourseof blackAmericans. negativeeffects in age cohortsovertime. racialdisparities Thisshouldbe observedin theformofincreasing thatblacksand Thismodelsuggests thevintagehypothesis. Diagrams3A and 3B illustrate iftheyhad similarlevelsofhumancapital.Therefore, as whiteswouldhavesimilarearnings blacks'levelsofhumancapitalbegintomorecloselyapproximate thoseofwhites,theirearncharacthatoncework-relevant predicts ingswilldo likewise.Thus,thevintagehypothesis teristics (such as education)have been takenintoaccount,the black-white earningsgap shouldbe virtually constantovertimeforall cohorts(vintages).To thedegreethatblackwhitehumancapitalgapshaveclosedsincethe1960s,Diagrams3Aand 3B accurately representthevintagemodel. First, however,we providea brief Below,we examinetheaccuracyofthesepredictions. ofthedata and methodsthatwe used to testthesepredictions. description The Cohort AnalysisFramework is its emphasison of the cohortanalysisframework The mostsignificant contribution substantive iswhenapproaching simultaneously considering age,period,and cohorteffects without to the others is a one of these effects sues. In mostinstances, regard any considering theidentificaassessment.Whena standardcohorttablehas been constructed, misleading wouldappearto be ratherstraightforward. tionofage,period,and cohorteffects Ageeffects differences intracohort can be determined (i.e., by readingthe cohorttable by examining of downand to theright).Thisallowsone to compare,forexample,theearnings diagonally ofthatsamecohort10 yearslater(in 1960) thecohortage 30-39in 1950withtheearnings whenitsmemberswouldthenbe age 40-49,and to comparethembothwiththesamecowouldthenbe age 50-59. Thisprocesscould hort10 yearslater(in 1970)whenitsmembers 613 614 THOMAS/HERRING/HORTON Vintage Hypothesis Pre 1960s C-p v-p C"p -- Whites C 0p- 20-29 30-39 40-49 50-59 Blacks 60-69 Age Diagram 3a Vintage Hypothesis Post 1960s -p - l" : 20-29 -- Whites Blacks S- 30-39 40-49 Age Diagram 3b 50-59 60-69 OvertheLifeCourse Discrimination thenbe repeatedfortheremaining birthcohorts(e.g.,thosewho were20-29 in 1950),with to those thatwereconsistently attention across agingeffects particular beinggiven replicated the different birthcohorts. thesameage category at one pointin time can be determined Periodeffects bycomparing at anotherpointin time(i.e.,byreadingacrosstherows). This withthatsameage category allowsone to compare,forexample,theearningsofthoseage 30-39in 1950 withthoseage 30-39 in 1960,and to comparethembothwiththoseage 30-39in 1970. Hereagain,greater thatwereconsistently acrossall attentionwouldbe givento thoseperiodeffects replicated age groups. can be determined Finally,cohort membership effects by examiningintercohort changes oftheearnings (i.e.,byreadingdownthecolumns).This,forexample,allowsa comparison of thoseage 30-39 in 1950 withthoseage 20-29,40-49, 50-59,and 60-69 in 1950. Once wouldbe givento thosecohorteffects thatwereconsistently again,greaterattention replitoconstructing standard catedacrossall periods.An alternative cohorttablesinvolvesgraphexamplediscussedabove, the graphic ing the relationships.Usingthe same substantive methodwould place the dependentvariable(in thiscase earnings)on the Y axis,and the on theX axis. The valuesforeach cohort(i.e., the diagonalstrings yearof measurement obtainedbyreadingdownand to theright)wouldthenbe plotted(usingdistinguishing symbols) and labeled. Iftherewereno age,period,or cohorteffects occurring, onlyone flatline wereoperative, wouldappearin thegraph.Ifonlycohorteffects parallelflatlineswouldbe ifonly observed.If onlyagingeffects existed,parallelslopeswouldbe seen. And finally, a singlejaggedlinewouldbe found. wereoccurring, periodeffects ofage,period,and cohorteffects in standard As simpleas itmayseem,theinterpretation In particular, the problemofstatistical cohorttablesand graphsis actuallyratherdifficult. in cohortanalysis.Sta(i.e.,theidentification problem)remainsa complication confounding in whichthebasic resultsfromthefactthatthereis a lineardependency tisticalconfounding oftwofactors(i.e.,age and period,or age and cohort,or periodand cohort)are both effects involvedin each diagonal,row,or columncomparison (Glenn1977). Forexample,age and in columncomparisons are represented becausethe cohortsto be compared cohorteffects are apparentin rowcomcohortand periodeffects have attaineddifferent ages. Similarly, birthcohortsare beingcomparedat different pointsin history. parisonsbecause different are involvedin diagonalcomparisons becausethe cohortsnot And age and periodeffects ofage,period,and cohort periods.Thus,theseparation onlyage,theyage intonewhistorical ofstandard cohorttablesis difficult unlesstheobservedeffects effects are byvisualinspection and consistent bothpronounced Forthisreason,theinterpretation acrossall comparisons. of and withoutambiguity.Some standardcohorttablesis seldomcompletely straightforward is to be expected. magnitudeofage and periodand cohorteffects Workhas been done to developstatistical ofcotechniquesto overcomethelimitations and Mason 1978; Smith,Mason,and hortanalysis(e.g.,Schaie 1965;Baltes1968;Fienberg frameFienberg1982; Mason and Fienberg1984; and Wolinsky1990). The "accounting work"of Mason and his colleaguesessentially employsa multipleclassification analysis framework.Here the dependentvariable,forexampleearnings,is predictedby a set of the age, period,and cohorteffects.Expressedin dummy(binary)variablesrepresenting form: mathematical notationin Equation1, thisapproachtakesthefollowing I-1 Y= k + d i=1 J-I K-I aiAi + d pjPj + d ckCk+ e k=1 j=1 (1) A is thesetofdummyvariablesrepresenting theage whereY is earnings;k is theintercept; P is thesetofdummyvariablesrepreand theai are theirregression coefficients; categories coefficients; C is thesetofdummy sentingthedifferent periodsand thepj aretheirregression 615 616 THOMAS/HERRING/HORTON thevariousbirthcohortsand theckare theirregression variablesrepresenting coefficients; and e is theerrorterm. Noticethatthisapproachbeginsbyomitting one dummyvariablefromeach oftheage, and period,and cohortparentterms.As a consequence, Equation1 remainsunderidentified cannotbe estimated.It is at thispointthatMason et al. (1973) introducetwo statistical involvedin theiraccountingframework: The firstrestriction restrictions is thatone additionaldummyvariablemustalso be omitted.Equation2 presentsan exampleof such a in whichtheadditional fromthesetofvariablesreprerestriction dummyvariableis omitted sentingtheage categories: 1-2 Y=k+d i=1 J-1 aiai+d j=1 K-1 pjpj+d k=1 ckck+e (2) willhave equivalentbehaviors(i.e., that Here,it is assumedthattwooftheage categories thereis no difference betweenthem,all otherthingsbeingequal). thatmustbe made in orderforEquation2 to workis The secondstatistical restriction thattheeffects ofage,period,and cohortmustonlybe additive.Thatis,theeffects ofeach not the of mustbe consistent the others. linear)throughout range (though necessarily Ifbothofthesestatistical restrictions can be made,thentheaccounting framework will the effects of age, period,and cohort.Unfortunately, thereis much succeedin separating andabouttheproblems arerealistic, ofimplementing debateaboutwhethertheserestrictions thewell-documented them(e.g.,Glenn1976,1977,1989;and Palmore1978). Clearly, relanot to and between do conform these For this we reason, earnings age tionship assumptions. tobe theunrealistic and restrictions of chosenotto relyon whatwe considered assumptions we reliedon a moregeneralmultiple theaccounting framework. classification Rather, analyand clearlystatedpredictions fromthecompeting theoframework, sis,a graphical analytical ofthepatterns in thedata. reticalmodelsto guideour dataanalysisand interpretations to resolvetheidentification Thereis generalconsensusthattheonlywayultimately or statistical problemis throughsoundtheory(Glenn1989). The cohortanalyst confounding must,therefore, statistical) (andnotstrictly groundsto explainthegeneral relyon theoretical of or cohort. The of identifying or effects the additive age, period, importance main) (i.e., of the tablescannotbe emphasized thosetheoretical groundsbeforethe visualinspection fromthatinvolvedin any enough. Althoughthisa priorirelianceon theoryis no different ofcohortanalysismakesit all formofanalysis(Masonand Fienberg1984),thecomplexity outlined themoreimportant (Glenn1989). Forthisreason,we use thethreeperspectives oftheresults:As mentioned the aboveto guideouranalysisand interpretations previously, discrimination model;thecumulaparallelsa periodeffects-driven explanation legacyof past, is analogousto an age effects-driven ofdiscrimination tiveeffect model;and the explanation a effects-driven model. to cohort is vintagehypothesis equivalent Data and Methods The dataforthisstudycomefromtwosources:theCurrent PopulationSurvey'sAnnual Extract File (March)for1990and theUnitedStatesMicrodata File,1940-1980: Demographic betweenthe ofAging.Thisstudyincludesonlythosemale respondents The Demographics ages of 20 and 69 who werein thelaborforceand workedformoneythepreviousyear.2 The above criteriayieldeda samplesize of 36,850 valid cases (30,956 whitesand 5,894 2. Because levels of labor forceparticipationamong black males tend to be lower than those forwhite males, an analysis based only on labor forceparticipants(who presumablyhave higherearningsthan non-participants)will provide a conservativeestimateof the black-whiteearningsdisparity. Discrimination OvertheLifeCourse blacks).3The dependentvariablein thisstudyis personalearningsin thepreviousyear. Independentvariablesin theanalysisincluderace (coded 1 forwhitesand 2 forblacks),age, status(coded 1 forself-emsector,employment occupationalstatus(lastyear),industrial yearsofeducation(coded0 through20 foractualyearsof ployed,0 fornotself-employed), 2 forMidwest,3 forSouth,and 4 forWest),city region(coded 1 forNortheast, education),4 for 0 for residence, 1 (coded city non-city residence),hoursworkedperweek (averagelast year)and weeksworkedlastyear.Ageis groupedfortheanalysisas follows:1) 20-29;2) 3039; 3) 40-49; 4) 50-59 and; 5) 60-69. The occupationalcategoriesused in each yearexamineddiffer fromeach otherslightly; however,forthisanalysis,theyhavebeengroupedas follows:1) professional, and proprietors; technical, managers, 2) administrative supportand and laborers;and 6) farming. sales; 3) craftspersons; fabricators, 4) service;5) operatives, withtheidea oflabormarketsegmentation, Consistent classifications were job industrial andgovernment industrial sectorsusinga systemdevelopedby groupedintocore,peripheral, Beck,Horan,and Tolbert(1978). Coreindustries (coded1) includemostdurablegoodsmanand wholesaletrades.Peripheral(coded2) inufacturers, mining,utilities, transportation, dustries includemostnondurable retailsales,entertainment industries, goodsmanufacturers, and personalservices.UnlikeBeckand hisassociates(1978) we have maintained as a separate categorythe government sector(coded 3), whichincludesjobs in federal,state,and local agencies. We used theregression-based MultipleClassification Analysis(MCA)5statistical procedureto analyzethedataforeach yearin thisstudy(i.e., 1940,1950,1960, 1970,1980,and 1990). We used earningsas the dependentvariable,and we used race,age, occupational status,yearsof education,industrial sector,employment status,city,region,hoursworked variables.In thecross-sectional perweek,and weeksworkedlastyearas independent analysis,in each year,we used MCA to computetheunadjustedand adjustedmean earningsof blacksand whitesacrossthe fiveage groupswhilecontrolling forthe otherindependent variables(see Andrews, Morgan,and Sonquist1973). Theadjustedmeanscan be interpreted as theexpectedmeansifblacksandwhiteswereidentical on all oftheindependent variables. The difference betweenblacks'and whites'adjustedmeansis the net race effectafterall otherindependent variableshave been takenintoaccount.The difference, is the therefore, net"cost"ofbeingblackas opposedtobeingwhiteinactualearnings.In thesynthetic cohort analysis,we examinedtheunadjustedand adjustedmeanearningsofblacksand whitesfor each cohortovertheyearsstudied.We identified fiveage cohorts:CohortA was in the2029 age groupin 1940 and wouldbe in the60-69 age category in 1980. CohortB was in the 20-29 age groupin 1950and wouldbe in the60-69 age groupin 1990. CohortC was in the 20-29 age groupin 1960and wouldbe in the50-59age groupin 1990. CohortD was in the 20-29 age groupin 1970and wouldbe in the40-49 age groupin 1990. CohortE was in the 20-29 age groupin 1980 and wouldbe in the30-39 age groupin 1990. 3. Blacks were over-sampledin the United StatesMicrodataExtractFile, 1940-1980: The Demographicsof Aging Study. 4. While thisanalysis does not employ an indicatorof qualityof education,it should be noted thatneitherhave many of the proponentsof the vintagehypothesis(e.g., Welch 1973, 1975; and Smithand Welch 1977). Instead, these formulationshave assumed thatschool qualitywas poorerforoldervintages,especiallyforblacks (see Hoffman1979 for an elaboration of this point). 5. MCA can be consideredthe equivalent of regressionanalysis using dummyvariables (Andrews,Morgan, and Sonquist 1973). However, MCA is more convenientthan dummy regressionanalysis because: 1) there is no need to recode all of the independentvariablesinto sets of dummyvariables; 2) the coefficients forall categoriesare expressed as deviations fromthe mean ratherthan fromsome unknown "reference"category;3) MCA does not sufferfromthe and, 4) MCA treatseach categoryof the independentvariable in a way that is similarto problem of multicollinearity; how a dummy variable is treatedin a regressionanalysis. Therefore,nominal,ordinal,and intervalscale variablesare appropriateas independentvariables in the analysis. See Andrews,Morgan, and Sonquist (1973) fora more detailed discussion of MCA. 617 618 THOMAS/HERRING/HORTON Findings Cross-Sectional Analysis,1940-1990 Table 1 presentsthe unadjusted mean earningsforblacks and whites forthe five age groups in each of the surveyedyears,and Figure 1 illustratesthese patterns. As Figure 1 shows, in all years except 1940, the youngestblacks (20-29) had unadjusted mean earnings closer to theirwhite age mates than blacks in any other age group. In fact,the percentage Mean EarningsofBlackand WhiteMalesbyAge,1940-1990 Table 1 * Unadjusted 1940 1950 1960 1970 1980 1990 Whites20-29 Blacks20-29 Black % of White $1,177 $421 35.8% $2,621 $1,951 74.4% $4,601 $3,043 66.1% $7,502 $5,635 75.1% $14,729 $12,634 85.8% $21,735 $18,646 85.8% Whites30-39 Blacks30-39 Black % of White $1,582 $600 37.9% $3,476 $2,157 62.1% $6,211 $3,666 59.0% $10,576 $6,558 62.0% $20,528 $15,079 73.5% $31,443 $23,683 75.3% Whites40-49 Blacks40-49 Black % of White $1,677 $672 40.1% $3,733 $2,150 57.6% $6,224 $3,170 50.9% $10,444 $6,618 63.4% $23,175 $15,089 65.1% $36,300 $26,788 73.8% Whites50-59 Blacks 50-59 Black % of White $1,562 $625 40.0% $3,410 $2,049 60.1% $6,138 $2,946 48.0% $9,805 $5,691 58.0% $20,732 $14,343 69.2% $34,595 $25,684 74.2% Whites60-69 Blacks60-69 Black % of White $1,452 $522 35.9% $3,003 $1,713 57.1% $4,825 $2,648 54.9% $8,266 $4,379 53.0% $17,458 $11,688 66.9% $28,689 $21,371 74.5% 1980 1990 100 - p 80- R 60 - E 40 - T 20 N . ... ..... -,. 1940 1950 20 to 29 S60 to 59 1960 1970 Year 30 to 39 40 to 49 60 to 69 Mean EarningsbyAge,1940-1990 Figure 1 * Percent ofBlackto WhiteUnadjusted DiscriminationOver the Life Course Blackand WhiteMales'Earnings Table 2 * MultipleClassification AnalysisModelsPredicting ofEarnings,1940-1990" byAgeand Decade,NetofOtherDeterminants Variables Independent GrandMeans: Race by Age Betab: Whites20-29 Blacks20-29 Whites30-39 Blacks 30-39 Whites40-49 Blacks40-49 Whites50-59 Blacks 50-59 Whites60-69 Blacks 60-69 Occupation Beta: Professional Administrative Crafts Service Operatives Farm Sector Beta: Core Periphery Government Region Beta: Northeast Midwest South West Education HoursPer Week WeeksPer Year Self-Employed City AdjustedR-Square N 1940 1950 1960 1970 1980 1990 18206.52 30884.16 8030.81 4610.58 2730.55 .18** .25** .27** .19"* .19"* -7220.29 -4662.25 -1504.39 -798.33 -497.82 -4112.31 -8149.37 -1019.94 -1675.17 -574.85 -516.74 15.58 1049.65 586.10 328.77 -4578.64 -2150.68 -934.36 -684.52 -316.61 2578.34 3353.71 1375.55 1046.37 607.73 -1619.74 -1216.97 -627.90 -634.81 -310.04 2179.99 3932.52 966.10 479.78 1012.79 -315.12 200.39 -624.32 -618.96 -293.96 -301.95 1215.57 79.86 -115.78 237.88 -846.55 -1513.57 -864.09 -952.52 -503.38 .22** .25** .27** .20** .20** 3454.71 5499.45 2347.45 1696.97 1050.04 -24.99 -556.15 22.68 -192.37 118.90 -57.16 -162.85 -1319.77 83.28 118.44 -3862.09 -5967.14 -942.88 -1477.31 -484.37 -1467.06 -3366.66 -313.84 -643.89 -172.51 -1693.52 -2679.02 -8422.77 -1102.23 -450.52 .08** .10** .08** .09** .08** 337.99 621.92 1417.41 176.53 95.49 -1549.39 -2989.63 -174.50 -375.16 -813.50 -274.29 124.86 -231.03 329.84 -42.07 .06** .12** .14** .08** .11** .11** 243.82 2935.93 191.98 774.32 6.26 149.87 694.13 403.16 -780.63 387.55 30.53 155.81 -879.74 -1610.28 -471.09 -1015.56 -252.98 -101.99 622.72 670.31 56.27 268.35 430.72 41.86 991.85** 2525.14** 331.34** 524.27** 99.08** 130.61** 53.62** 133.91** 223.85** -2.82 14.04* -6.48** 361.98** 129.07** 631.02** 25.21** 78.97** 53.57** 911.76** 4812.41** 5258.22** -356.36 -506.08 -184.21 -127.40 -473.07* -1541.28** -240.37 140.61** -86.52 .301** .316** .440** .425** .541** .341** 2971 2639 29159 2944 1978 2820 1121.86 .21** -135.49 -298.44 126.15 -180.92 251.35 -160.63 232.67 -203.21 103.87 -222.63 .32** 728.69 173.52 139.26 -251.64 -122.97 -304.22 .11** 78.65 -112.75 203.67 ** p < 01 * p < .05 coefficients arepresented as are unstandardized a. Coefficients (dollaramounts).TheMCAcategorical arepresented as multiple forthecovariates fromthegrandmean. Coefficients deviations regression coefficients. in the ofthepredictor, theability b. Betaindicates given,to explainvariation usingall thecategories forall of in OLS regression coefficient variable(earnings).It is analogousto thecorrelation dependent variable. oftheindependent thecategories drop in black-to-whiteearningsfromthe youngestgroup to the next youngestgroup was quite dramatic (down 11 percent on average). Young blacks' earningsas a percentage of young whites' earningswere greatestin 1980 (85.8 percent) and 1990 (85.8 percent) followed by 1970 (75.1 percent). Declines in the ratio of black-to-whiteearnings fromthe youngestage group to the next youngestage group were greatestin 1970 (-13.1 percent), 1980 (-12.3 percent),and 1950 (-12.3 percent). There was increasingracial disparityin mean earningswith age-especially in 1990, 1980, 1970, and 1950. This patternis mostconsistentwiththe "cumulativeeffectof discrimination" perspectiveof the racial earningsgap. In 1960, however,therewas a curvilinearage effectcaused by a modest convergencein mean earningsof black and white males in the 6069 age group. 619 620 THOMAS/HERRING/HORTON Table 2 presentsthe multivariate resultsfrommultipleclassification analysismodels ofblackand whitemalesbyage and year,netofotherdetertheearnings (MCA) predicting minantsofearnings.Withonlyone exception, thistableshowsthatblacksin each age catenetoftheothercorrelates of goryand in eachyearearnedlessthantheirwhitecounterparts, also suggestthatvirtually all age groupsof earnings.Moreover,the negativecoefficients blacksin everyyear(except50-59 in 1990) receivedless thanaverageearnings.Thisis in to thepatternforwhites,whereonlytheyoungest contrast and theoldestrerespondents from1970-1990hadearnings thatwerebelowaverage.Morecritical forthethree spondents theoretical modelsat hand,however,is thefactthatforeach year,theblack-white earnings moreclearlyin gap increasedwithage (untiltheoldestage group).Thispointis illustrated aremostconsistent withthecumulative Table3 and in Figure2. Again,thesepatterns effect of discrimination perspective. in themodelwerealso relatedto earnings.Foreach Table2 also showsthatotherfactors attainedthe highestearnings, year,occupationevidenceda patternin whichprofessionals serviceworkers, followedbyadministrators, and farmworkers craftsmen, operatives, respecin thecoresectorhad greaterearningsthanthose tively.Theresultsalso showthatworkers in thegovernment sector,withthosein theperiphery earningtheleaston average.Regional in theSouthconsistently differences variedbyyear,butrespondents hadthelowestearnings. Theresultsalso showearnings increased as educationincreased, as numberofwork generally as numberofworkweeksperyearincreased, whenone was selfhoursperweekincreased, factors employed,and when one livedoutsidethe city. In each model,the independent accountedformorethan30 percentbutlessthan55 percentofthevariancein earnings. theresultsfromtheMCA,andFigure2 graphically Table3 summarizes depictsthemean for of white mean each as a of blacks yearexamined. earningsbyage percentage earnings controlforother Here,the resultsare presentedas adjustedmeansthatagain statistically reflect whatwe wouldexpectblackand ofearnings.Theseadjustedmeanearnings correlates whitemalesto earniftheywereidenticalin termsofoccupation, education,etc. Thegapin thenegativeeffect of"beingblack"on earningsunmediated adjustedmeanearningsreflects variablesin theanalysis. bytheotherindependent Theresultsshowthatyoungerblackswereagaincloserto theirwhitecounterparts than wereolderblacksto theirwhitepeers.As Figure2 shows,blacksin the20-29age grouphad adjustedmean earningsthatwerecloserto theirwhiteage equals thanblacksin otherage increasein theracialgapbetweenthe groupsin all oftheyears.Therewas also a significant earnings youngestto secondyoungestage groupsin everyyearexamined.The black-white gap increasedbetweenages 30-39 and 40-49 in all yearsexcept1970 and 1990. However, that oftheracialgap fortheoldestage category.Thisagainsuggests therewas a narrowing was curvilinear.Youngerblacksin each year the effectof age on the net racialdisparity Theracialgapthenwidenedsignif(including1940) werecloserto theirwhitecounterparts. in all yearsexbutnarrowedagainfortheolderage categories icantlyforthemiddle-aged, 1950. cept CohortAnalysis as suggested is curvilinear, Iftheeffect ofagingon theracialgapin earnings bythecrossin theage cohortsas theymoveovertime.Figure thisshouldbe reflected sectionalanalysis, of whites'unadjustedmean blacks'unadjustedmean earningsas a percentage 3 illustrates theblack-toin Table1). As Figure3 illustrates, E (presented earningsforCohortsA through E werehighestfortheyoungest whiteearningsratiosforCohortsB through age group.But reachedthe30-39age point.For whenworkers foreach cohort,theydeclinedsignificantly CohortA, the racialgap in unadjustedmean earningsdecreasedbetweenages 20-29 (in 1940) to ages 30-39 (in 1950),butincreasedat age 40-49 (in 1960). ForCohortsA through Discrimination OvertheLifeCourse Table 3 * AdjustedMean EarningsofBlackand WhiteMalesbyAge,1940-1990 1940 1950 1960 1970 1980 1990 Whites20-29 Blacks20-29 Black % of White $986 $823 83.5% $2,233 $2,156 96.5% $3,812 $3,591 94.2% $6,526 $6,356 97.4% $13,544 $14,094 104.1% $23,664 $22,735 96.1% Whites30-39 Blacks30-39 Black % of White $1,248 $941 75.4% $3,059 $2,414 78.9% $5,197 $3,926 75.5% $9,080 $7,096 78.2% $18,222 $16,056 88.1% $30,367 $26,306 86.6% Whites40-49 Blacks40-49 Black % of White Whites50-59 Blacks 50-59 Black % of White $1,373 $961 70.0% $3,338 $2,421 72.5% $5,657 $3,976 70.3% $9,406 $7,403 78.7% $20,785 $16,587 79.8% $34,238 $29,667 86.7% $1,355 $919 67.8% $3,210 $2,437 75.9% $5,623 $3,992 71.0% $8,997 $7,406 82.3% $20,387 $17,891 87.8% $34,817 $31,085 89.3% Whites60-69 Blacks60-69 Black % of White $1,226 $899 73.4% $2,968 $2,227 75.0% $4,690 $3,746 79.9% $7,915 $7,078 89.4% $17,905 $17,360 97.0% $32,100 $29,371 91.5% 1980 1990 120 100- E 80- C 60- E N T 40.. 20 0 i 1940 1950 1960 20 to 29 5 50 to 59 Year 30 to 39 L 1970 40 to 49 60 to 69 MeanEarningsbyAge,1940-1990 Figure2 * Percent ofBlackto WhiteAdjusted C, therewas a modest trendtowardconvergenceof black and white unadjusted mean earnings at the older age points. Figure 4 charts black-to-whiteearnings for the five cohorts,controllingfor the independentvariablesin thisstudy(as presentedin Tables 2 and 3). As withthe cross-sectional portionof the study,controllingforthese variablesincreasesthe curvilineareffectof agingon the racial gap in earnings. In all fivecohorts,therewas a wideningof the racial gap in adjusted mean earningsbetween ages 20-29 and 30-39. This widening was particularlydramatic forCohortsB throughE. The convergencein adjusted mean earningsforolder blacks and whites was also evidentaftercontrols. 621 100 80604020- 1940 1950 1970 1960 1 YEAR Cohort A Cohort B Cohort D Cohort E of WhiteMale Mean EarningsFor FiveAgeCohorts Figure3 * BlackMale Percentage Coh 120 100 80 6040201940 1 1970 1960 1950 YEAR Cohort A Cohort D - Cohort B Cohort E of WhiteMale AdjustedMean EarningsFor FiveAge Cohorts Figure4 * BlackMale Percentage ?" Coho 624 THOMAS/HERRING/HORTON An important is thedramatic declinein blackearningscomparedwiththosefor finding whitesfromtheyoungest (20-29 age category years)to thenext(30-39 years).Thisdecline is consistent foreveryperiodand cohort(see Tables2 and 3). Thisresultis similarto those reportedin priorstudies(Farley1984; Farleyand Allen 1987; Thomasand Horton1992; Thomas1993). Attheyoungest bothblacksand whitesare likelyto holdlowage category, of lack of work because their experience.However,as theymoveintotheir30s payingjobs an important occurs. Whitesbeginto gain not onlyfromthe amountof time divergence thatlimitopportunities for spentin the laborforce,but fromthe absenceof the barriers blacks(Duncan 1968; Lieberson1980). Similarfindings have been notedin studiesofselfofwell-being and otherindicators (Hortonand DeJong1991;Horton1992). In employment have beenprevented frommaximizing theirhumancapitalrelaessence,blackshistorically tiveto whites(Franklinand Moss 1988), and thispatternpersistsover timeand the life course(Thomas1993). effects Theseresultsare consistent withthecumulative ofdiscrimination as explanation, theblack-white earningsgap didbecomeprogressively largerforeach cohortas theymoved withthismodel(noranyother),however,is thefindthroughthelifecycle.Notconsistent ofblack-white foreach cohort patterns ingthattherewereuniform earningsconvergence in earningsin the aftertheir"primeearningsage" (afterage 50). Thisracialconvergence ofreducedworkeffort lateryearsmaybe a reflection bywhitemaleswho havemorewealth builtup thanblackmales,have othersourcesofincomebesidesearnings, and have entered ofearningsbetween As Tables1-3 show,theconvergence and semi-retirement. retirement olderblacksand whiteswas causedbythedropin earningsbetweenages 50-59 and 60-69 thatwas moredramaticforwhitesthanforblacks.Thatis,theearningsofolderblacksdid theearnings ofolderwhites notimproveand becomesimilartothoseofolderwhites;rather, declinedsharplyand becamemoresimilarto thoseofolderblacks.6 Discussion and Conclusions racialdisparity in The "legacyof past discrimination" explanationforthe continuing the here. blacks no from cohort were receives support analysispresented Younger earnings in all oftheyears blackscomparedwiththeirwhitecounterparts betteroffthanmiddle-aged analyzed.Thiswas notdue to theirbeingable to takeadvantageofnewlydevelopedopporofthe to theprojections tunitiesbutto theirpositionin thelifecycle.Therefore, according cohortanalysis,withinthenext10 yearstheearningsofblacksin theyoungest age groups and willremainbehindforthenext10 behindtheirwhitecounterparts willfalldramatically will"catchup,"due to more theirearnings to 20 years.However,whentheybecomeelderly, sharplydecliningwhiteearnings. Thenetblackforthevintagehypothesis. Theresultsprovidelittlein thewayofsupport whiteearningsgap was not constantovertimenoroverthelifecycle. Indeed,the results grewand thendeclinedwithincohorts. suggestthatsuchdisparities over the life of the cumulativediscrimination The studyprovidessome confirmation ofraceon blackmalesdidincreaseas theymoved coursethesis.Whilethenegativeeffects decreasedas blackmalesmovedintothesenior intothemiddlelifestages,theracialeffects to notethatblacksin the30 to 50 age grouparein their"prime"in lifestages.Itis important termsof raisingchildren, careers,and buildingwealth. Thisanalysissuggests establishing in termsofearnings. thatitis blackmalesin thisage rangewhowerethemostdisadvantaged Our analysisalso showsa patternthatwas unanticipated byany ofthemodels:There oftheearningsofblackand whitemalesin theupperage groups.It is was a convergence 6. In other analyses not presentedhere, the above analysis was repeated forcollege educated and non-college fromthe patternfound in the overall sample. educated subsamples. Neithersubsample differedsignificantly Discrimination OvertheLifeCourse likelythatwage earningsbecomemoresimilarat thisstagein thelifecyclebecauseelderly whitemales beginto dependmoreon accumulatedwealththantheirblackcounterparts. in wealthis muchgreater thanracialinequalityin Studieshave shownthatracialinequality earningsand income(e.g.,Oliverand Shapiro1989). Elderlywhitemalesmay,therefore, assets. becauseofearnings fromincome-producing de-emphasize earningsfromemployment Blackelderlymales,on theotherhand,mayhave to worklongerdue to inadequateretireolderwhitemales mentbenefits.Because of this,theremaybe a selectionbias involving to retireearlier(in whitescan perhapsafford who are stillin thelaborforce.Moreaffluent theirlate 50s to mid-60s)iftheychoose,and thusnotbe includedin thisstudy.Thiswould theluxury whitesstillin thelaborforcewhocouldnotafford leavebehindthoselessaffluent of early retirement-a situationtheywould share with theirblack counterparts.This could account forthe convergencebetween black and white earningsin the 60-69 age group.7 The empiricalfindingsallude to a cogent explanation of why young blacks relativeto while olderblacksrelativeto older youngwhiteswould have smallearningsdifferences are notguidedby purerationality whiteswouldhave largerones: Labormarketdynamics to and perfectcompetition; rather,a numberof concreteprocessesoperatesystematically forblacksand whites. For example,segmentedlabor remuneration generatedifferential different markets offer jobs (Cain 1976;Beck,Horan,andTolbert1978;Dickens qualitatively in theprimary labormarketoffer and Lang1988;and Lichter1988). Positions relatively high foron-the-job and thepossibility advancement; jobs in the pay,good benefits, job security, seasonalor contingent emlow pay,poorbenefits, secondarylabormarketofferrelatively ployment,and littleupward mobilityon the job. Blacks are under-representedin the primary labor marketand over-representedin the secondarylabor market (Lichter 1988). In addition, normative beliefs about "black jobs" and "white jobs" (Neckerman and Kirschenman 1991), employerreliance on "softskills"ratherthan tangibleones (Kirschenman networks(Braddockand McPartland recruitment "tastes 1987),employers' 1992), informal Neckerman and and Kirschenman and Sassen-Koob fordiscrimination," 1987; 1991) (Portes associations and Oliver and exclusionary (Johnson practices bylaborunionsand professional labormarketin lower 1992) all act to steerblacksintoraciallytypedjobs in thesecondary and opportunities forupwardmobility.Butmoreimportant payingjobs withfewerbenefits is the factthatin the secondarylabormarket,additionalyearsof workexperiencedo not labormarket)do notimprovemuch countformuch;thus,blacks'earnings(in thesecondary withadditionalyearsofexperience, behindwhites'earnings(in the and thustheyfallfarther labormarket)witheach passingyear. primary In addition to these segmentedlabor marketdynamics,thereare also firm-specific practices that come into play. In particular,when young blacks and whites initiallybegin to work forcompanies,neitherrace group has much work-relevantexperience. There is, however, a tendencyforthe same credentialsforblacksto resultin less access to professionaljobs than forwhites (Shelton 1985; Landry 1987). Thus, smallerproportionsof college-educated blacks occupy positions of authority. So, fewerblacks are able to influencedecisions con- of subordinates. blacksat different and promotion cerninghiring,retention, Consequently, oftheirwhitecounterparts. careerstageslacktheobjectivesponsorship Moreover,as they withinfirms is morelikelyto inhibittheupwardmobility ofemployment age, thestructure aregivenfeweropporofblacks.As Collins(1989) pointsout,blacksin managerial positions to enhancetheirprofessional tunitiesto accrueadditionalskillsand responsibilities credentialsforfutureroundsof competition. Also,as Johnsonand Herring(1989) demonstrate, 7. In otheranalyses not presentedhere, whiteswere foundto be more likelyto be workingthan blacks in all age bracketsexcept the 60-69 category. Blacks in the 60-69 age bracketwere more likelyto be working(50.2 percent)than theirwhite counterparts(46.7 percent). This suggeststhat whites may be retiringearlierthan blacks - probablybecause it is economicallymore feasibleforthem to do so. 625 626 THOMAS/HERRING/HORTON individualswho do not experienceearlycareersuccessesare less likelyto experiencepromoless likelyever to be promoted. This [tentionsin subsequent years. "Blacksare significantly dency] continues to be exacerbatedover time" and over the lifecycle (Johnsonand Herring 1989:22). In addition,blacksand individualsfromlow socioeconomicbackgroundsare more likelyto be dismissedfromtheirjobs, especiallyearly in theircareers. Again, the career effectsof such dismissalspersistover the lifecourse (Johnsonand Herring1989). Thus, fewer blacks benefitfromthe higherearningsand rewardsthatare associated with promotionand job tenure. These patternsare consistentwith the argumentsof the cumulative effectsof discriminationmodel. In sum, thisanalysis raises anew questions about whetherrace is decliningin its significance in the work place, and whether studies that optimisticallypoint to gains made by younger,more educated blacks are accurate. If historyis any predictor,and if currentpatternspersist,presumedgains made by young black males may in factturnout to be illusory. References Akin,JohnS., and IrvGarfinkel in black-white 1980 "Thequalityofeducationand cohortvariation earningsdifferentials: American EconomicReview 70:186-191. Comment." FrankM., JamesMorgan,and JohnSonquist Andrews, forSocialResearch. 1973 MultipleClassification Analysis.AnnArbor,Mich.:Institute Baltes,Paul B. and cross-sectional 1968 "Longitudinal effects." sequencesin thestudyofage and generation HumanDevelopment 11:145-171. II M. Horan,and CharlesM. Tolbert, Beck,E.M., Patrick in a dual economy:A sectoralmodelofearningdetermination." American 1978 "Stratification Review43:704-720. Sociological Braddock, Jomills Henry,II, and JamesM. McPartland continueto be excludedfromequal employment 1987 "How minorities opportunities: and institutional barriers." Journalof SocialIssues43:5-39. Researchon labormarkets Cain,GlenG. labormarkettheoriesto orthodoxtheory:A survey." 1976 "Thechallengeofsegmented 14:1215-1257. JournalofEconomicLiterature Collins,Sharon ofblackexecutives." SocialProblems 36:317-331. 1989 "Themarginalization Cotton,Jeremiah oftheblackmiddle 1990 "Thegap at thetop: Relativeoccupational earningsdisadvantages class." ReviewofBlackPoliticalEconomy18:21-37. Dickens,WilliamT., and KevinLang American EconomicReview of segmented labormarkettheory." 1988 "Thereemergence 78:129-134. Duncan,OtisD. or inheritance ed. Daniel ofpoverty ofrace?"In On Understanding 1968 "Inheritance Poverty, 85-110.New York:BasicBooks. Moynihan, Duncan,Greg,and Saul D. Hoffman Journal economicstatusofblackworkers." 1983 "A new lookat thecausesoftheimproved ofHumanResources17:268-282. Farley,Reynolds Mass.:HarvardUniversity Press. theGap? Cambridge, 1984 Blacksand Whites:Narrowing and WalterAllen Farley,Reynolds, 1987 The ColorLineand theQualityofLifein America.New York:RussellSage Foundation. DiscriminationOver the Life Course Fienberg,Steven E., and William M. Mason 1978 "Identificationand estimationof age-period-cohortmodels in the analysis of discrete archival data." Sociological Methodology 19:1-67. Franklin,John Hope, and AlfredA. Moss 1988 From Slaveryto Freedom, 6th edition. New York: Knopf. Glenn, Norval D. 1976 "Cohort analysts' futilequest: Statisticalattemptsto separate age, period, and cohort effects."American Sociological Review 41:900-904. 1977 Cohort Analysis.BeverlyHills: Sage. "A flawed approach to solving the identificationproblemin the estimationof mobility 1989 effectmodels: A commenton Brody and McRae." Social Forces 67:789-795. Hoffman,Saul D. 1979 "Black-whitelifecycle earningsdifferencesand the vintagehypothesis:A longitudinal analysis." AmericanEconomic Review 69:855-867. Horton, Hayward Derrick 1992 "Race and wealth: A demographicanalysis of black homeownership." Sociological Inquiry 62:480-489. Horton, Hayward Derrick,and Gordon DeJong 1991 "Black entrepreneurs:A sociodemographicanalysis." Research in Race and Ethnic Relations 6:105-120. Jaynes,Gerald, and Robin Williams,eds. 1989 A Common Destiny:Blacks and American Society.Washington,D.C.: National Academy Press. Johnson,James H., and Melvin L. Oliver 1992 "Structuralchanges in the U.S. economy and black male joblessness: A reassessment." In Urban Labor Marketsand Job Opportunity,eds. G. Petersonand W. Vroman, 113147. Washington,D.C.: Urban InstitutePress. Johnson, RobertJ., and Cedric Herring 1989 "Labor marketparticipationamong young adults: An event historyanalysis." Youth & Society 21:3-31. Kirschenman,Joleen 1992 "Low-wage labor marketsin a restructured economy: Skill requirementsand black men." Paper presentedat Race, Ethnicityand Urban PovertyWorkshopsponsored by the Universityof Chicago and NorthwesternUniversity. Landry,Bart 1987 The New Black Middle Class. Berkeley,Calif.:Universityof CaliforniaPress. Link, Charles, Edward Ratledge,and Kenneth Lewis 1976 "Black-whitedifferencesin returnsto schooling: Some new evidence." American Economic Review 66:221-223. 1980 "The quality of education and cohortvariationin black-whiteearningsdifferentials: Reply." AmericanEconomic Review 70:196-203. Lichter,Daniel T. 1988 "Racial differencesin underemploymentin Americancities."AmericanJournalof Sociology 93:771-792. Lieberson, Stanley A Piece of the Pie: Blacks and White ImmigrantsSince 1880. Berkeley:Universityof 1980 CaliforniaPress. Mason, Karen O., WilliamM. Mason, H.H. Winsborough,and W. Kenneth Poole 1973 "Some methodologicalissues in cohortanalysis of archival data." American Sociological Review 38:242-258. Mason, William M., and Stephen E. Fienberg,eds. 1984 Cohort Analysisin Social Research: Beyond the IdentificationProblem.New York: Springer-Verlag. Neckerman,KathrynM., and Joleen Kirschenman 1991 "Hiringstrategies,racial bias and inner-cityworkers:An investigationof employers' hiringdecisions." Social Problems38:433-447. 627 628 THOMAS/HERRING/HORTON Oliver, Melvin, and Thomas M. Shapiro 1989 "Race and wealth." Review of Black PoliticalEconomy 17:5-25. Palmore, Erdmann 1978 "When can age, period,and cohortbe separated?" Social Forces 57:285-295. Portes,Alejandro,and Saskia Sassen-Koob 1987 "Making it underground: Comparativematerialon the informalsectorin western marketeconomies." AmericanJournalof Sociology93:30-61. Schaie, K. Warner 1965 "A general model forthe studyof developmentalproblems."PsychologicalBulletin 64:92-107. Shelton, Beth Anne 1985 "Racial discriminationat initiallabor marketaccess." Paper presentedat the Annual Meeting of the American SociologicalAssociation,Washington,D.C. Smith,HerbertL., William M. Mason, and Stephen E. Fienberg 1982 "More chimerasof the age-period-cohortaccountingframework:Commentson Rodgers."American Sociological Review 47:787-793. Smith,James,and Finis Welch 1977 "Black-whitemale wage ratios, 1960-1970." AmericanEconomic Review 67:323-338. Thomas, Melvin E. 1993 "Race, class and personal income: An empiricaltest of the decliningsignificanceof race thesis, 1968-1988." Social Problems40:601-614. Thomas, Melvin E., and Hayward D. Horton 1992 "Race, class and familystructure:The case of familyincome, 1968-1988." Sociological Perspectives35:433-450. U.S. Commissionon Civil Rights 1978 Social Indicatorsof Equality forMen and Women. Washington,D.C.: A Reportof the U.S. Civil RightsCommission. Welch, Finis 1973 "Black-whitedifferencesin returnsto schooling."AmericanEconomic Review 63:893907. 1975 "Human capital theory:Education, discrimination, and life-cycles."American Economic Review 65:63-75. 1980 "The quality of education and cohortvariationsin back-whiteearningsdifferentials: Reply." AmericanEconomic Review 70:186-191. Willie, Charles V. 1979 The Caste and Class Controversy.Bayside,N.Y.: General Hall Inc. Wilson, WilliamJulius 1978 "The decliningsignificanceof race: Revisitedbut not revised." Society 15:11, 16-17, 1921. 1980 The Declining Significanceof Race: Blacks and ChangingAmericanInstitutions,2nd edition. Chicago: Universityof Chicago Press. 1987 The TrulyDisadvantaged:The Inner City,the Underclass,and Public Policy. Chicago: Universityof Chicago Press. Wolinsky,FredricD. 1990 Health and Health Behavior Among ElderlyAmericans:An Age-Stratification Perspective.New York: Springer.