Litigation Costs and Returns to Experience...Litigation Costs and Returns to Experience ByPAULOYERANDSCOTTSCHAEFER* We develop a model linking maximum damage awards available to plaintiffs

Litigation Costs and Returns to Experience

By PAUL OYER AND SCOTT SCHAEFER

We develop a model linking maximum damage awards available to plaintiffs inwrongful termination lawsuits workersrsquo propensity to sue as a function of experi-ence and returns to experience Using Equal Employment Opportunity Commissiondata on protected-workerdiscrimination complaints and labor-market data from theCurrent Population Survey we examine how returns to experience among protectedworkers changed around the passage of the Civil Rights Act of 1991 We show thatemployersrsquo reactions to employment protections may induce redistributive effectsFurthermore these effects operate not merely across groups of differing protectedstatus but also within groups of identical protected status (JEL D21 J31 J71 K31)

Recent work studying the effects of antidis-crimination protections has focused on the pos-sibility that laws aimed at protecting jobs andpreventing employment discrimination may ac-tually work against protected workers by raisingthe expected costs of employing these workersTheoretical and empirical analyses of such leg-islation usually treat members of protectedgroups as homogeneous and examine whetherand if so how the protections redistribute em-ployment outcomes across groups of work-ers (see eg Thomas DeLeire 2000 DaronAcemoglu and Joshua D Angrist 2001) Ig-nored in this literature however is the possi-bility that antidiscrimination protections mayredistribute employment outcomes amongmembers of protected groups making someprotected workers better off and others worseoff In this paper we examine this question bystudying the relationship between litigation

costs and returns to experience for members ofprotected groups

We begin by developing a stylized model ofthe labor-market effects of employment dis-crimination legislation In our model rms andworkers are initially symmetrically uninformedregarding workersrsquo abilities and public signalsregarding ability are revealed over time Suchlearning gives rise to for-cause rings in whichworkers whose abilities are revealed to be loware terminated However some managers dis-criminate against certain workers and re themeven if ability is not revealed to be low As thelegal system can imperfectly distinguish be-tween for-cause and discriminatory rings bothsets of workers face the option of ling anunlawful termination lawsuit Increases in max-imum damage awards associated with employ-ment discrimination lawsuits affect wages bymaking workers who are likely to sue morecostly to employ

The key issue in linking increases in litigationcosts to changes in returns to experience istherefore whether the increase in litigation costsis greater for experienced or inexperiencedworkers If for example inexperienced workersare more likely to be red for cause (as is thecase in our learning model) then these workersmay be more likely to sue conditional on beingemployed Increases in litigation costs maytherefore make inexperienced workers rela-tively more costly to employ and employerswill discount wage offers accordingly Poten-tially offsetting this effect however is the factthat back (and some future) wages often com-prise one part of damage awards The prospect

Oyer Graduate School of Business Stanford Univer-sity 518 Memorial Way Stanford CA 94305 SchaeferKellogg School of Management Northwestern University2001 Sheridan Road Evanston IL 60208 We thank JoshuaAngrist David Besanko David Butz William CarringtonJim Dana Kathleen Engel Jane Goodman-DelahuntyRachel Hayes Thomas Klein Canice Prendergast ChrisTaber two referees and participants in seminars at theUniversity of Chicago Princeton University NorthwesternUniversity Stanford University the 1999 AEA Meetingsthe 1999 NBER Summer Institute and the 2000 SOLEEALE Meetings for comments We thank Daron AcemogluJosh Angrist and Chris Mazingo for providing a researchextract of EEOC complaints data We received generous nancial support from the Institute for Policy Research atNorthwestern University (Oyer) and the Searle Fund(Schaefer)

683

of higher damages may make experiencedworkers more likely to sue conditional on be-ing red which may imply that increases inlitigation costs will reduce rmsrsquo demand forthese workers Developing these effects furtherwe demonstrate that the effect of increases inlitigation costs on returns to experience de-pends crucially on (i) how employeesrsquo propen-sity to le employment-discrimination litigationconditional on employment varies with experi-ence and (ii) how the increase in propensity tosue (stemming from the increase in litigationcosts) conditional on employment varies withexperience

The passage of the Civil Rights Act of 1991(CRA91) provides an opportunity to study thisrelationship empirically This Act contains anumber of provisions that increased the ex-pected costs to rms of displacing protectedemployees The group protected by CRA91 isbroad and includes racial minorities femalesand those with disabilities While previous fed-eral employment discrimination legislation typ-ically limited plaintiff recovery to lost wagesCRA91 allows employees to sue for up to$300000 in punitive damages By extendingthe Civil Rights Act of 1866 CRA91 allowsemployees claiming unlawful termination onthe basis of race to sue for unlimited punitivedamages CRA91 also gives either side in asuit the right to a jury trial this presumablyfavors plaintiffs as juries are thought to bepartial to claims of individuals over those of rms

We proceed by identifying relationships be-tween the propensity to sue as a function of ageand changes in the returns to experience amongprotected workers around the time of the pas-sage of CRA91 Using data on complaints ledwith the Equal Employment Opportunity Com-mission (EEOC) we nd that wrongful termi-nation complaints drop sharply with age forwomen but rise steadily with age for blacksWe tie this nding into an analysis of the returnsto experience for these protected groups Usingdata from the 1988ndash1996 annual demographics le of the Current Population Survey (CPS) we nd that CRA91 had relatively minor aggregateemployment and wage effects However thelaw does appear to have affected returns toexperience in the way suggested by our modelWe show that returns to experience increasedfor women but not for blacks shortly after the

passage of the Act This nding is consistentwith the experiencepropensity-to-sue relation-ships found in the EEOC data Together these ndings offer a pattern that ts with our modelof litigation costs and returns to experience Wetake these results as evidence that antidiscrimi-nation protections (and employment protectionsin general) have redistributive effects Theseeffects appear to operate not merely acrossgroups of differing protected status but alsowithin groups of identical protected status

Although our empirical analysis focusessolely on CRA91 our model applies equallywell to any erosion of employment-at-willWhile CRA91 did aid the growth in employ-ment discrimination litigation the number ofsuch suits had been increasing steadily for atleast two decades before the passage of CRA91Our analysis therefore suggests that this increas-ing tide in litigation (or more generally con-tinued erosions in employment-at-will) mayhave been a contributing factor in the observedincrease in returns to experience over thatperiod1

I The Civil Rights Act of 1991and the Legal Environment

The Civil Rights Act of 1991 which tookeffect on November 21 1991 strengthened sev-eral prior pieces of employment-discriminationlegislation including the Civil Rights Act of1866 the Civil Rights Act of 1964 (Title VII)the Age Discrimination in Employment Act(ADEA) and the Americans with Disabilities

1 John J Donohue and Peter Siegelman (1991) docu-ment the growth in employment-discrimination litigationthroughout the 1970rsquos and 1980rsquos Lawrence F Katz andKevin M Murphy (1992) and John Bound and GeorgeJohnson (1992) show that the returns to experience in-creased for all workers over this period Bound and RichardB Freeman (1992) show this effect was stronger for blacksthan for whites while June OrsquoNeill and Solomon Polachek(1993) and Francine D Blau and Lawrence M Kahn (1997)show the effect was stronger for women than for menDavid Neumark and Wendy A Stock (1999) consider theeffects of employment protections on returns to experienceover the 1980rsquos but from a very different perspective Theystudy age-discrimination laws and suggest that by provid-ing an enforcement mechanism for implicit contractsthese laws increase the steepness of wage pro les for allworkers Unlike ours their analysis does not explicitlylink wage pro les to litigation costs imposed by protectedworkers

684 THE AMERICAN ECONOMIC REVIEW JUNE 2002

Act (ADA)2 The Act also counteracted several1989 Supreme Court interpretations of earlierantidiscrimination legislation notably WardsCove Packing Co v Atonio and Patterson vMcLean Credit Union CRA91 contained threeprovisions that may have affected the willing-ness of displaced employees to le race- andgender-based discrimination lawsuits it in-creased damage awards available to plaintiffsallowed plaintiffs to ask for jury trials andmade it somewhat easier for plaintiffs to usestatistical evidence to prove discrimination Wediscuss each provision in turn

CRA91 affected available damage awards byboth amending Title VII and extending theCRA of 1866 CRA91 amends Title VII whichhad limited damage awards to back pay only byallowing plaintiffs who allege intentional race-or gender-based discrimination to sue for puni-tive and compensatory damages Maximumdamages under CRA91 vary by employer sizeranging from $0 for rms with fewer than 15employees to $300000 for rms with more than500 employees In addition CRA91rsquos extensionof the CRA of 1866 removed all limits ondamages in cases of racial discrimination intermination The 1866 Act which forbids dis-crimination on the basis of race in the ldquomakingand enforcement of contractsrdquo allows plaintiffsto sue for unlimited punitive and compensatorydamages While the Supreme Courtrsquos Johnsonv Railway Express Agency (1975) and Runyanv McCrary (1976) decisions clearly interpretedthis Act as applying to employment contractsthese decisions did not clarify whether it alsoapplies to the ending of employment contracts(Note that the language of the Act quotedabove is somewhat ambiguous on this point)Throughout the 1980rsquos different federal courtsapplied somewhat different interpretations onthis point which meant that some plaintiffsalleging racial discrimination in ring were al-lowed to proceed under the CRA of 1866 whileothers could sue under Title VII only In the1989 Patterson decision the Supreme Court

ruled that the CRA of 1866 did not apply to thetermination of employment contracts Thus be-tween 1989 and 1991 plaintiffs in such casescould sue under Title VII only and thus couldclaim only back wages as damages CRA91explicitly extends the CRA of 1866 to the ter-mination of contracts thereby removing all lim-its on potential damage awards in race-basedcases Because suits led under the CRA of1866 go directly to federal court (instead ofgoing through the EEOC as Title VII claimsmust) and there is no detailed data source on thenumber of such cases it is impossible to fullyquantify the effects of Patterson and CRA91 onthe litigiousness of black men

CRA91 also gives plaintiffs who seek puni-tive damages the right to a jury trial This mayincrease the costs of displacing workers for tworeasons (i) juries are perceived to favor claimsof individuals rather than corporations and (ii)jury trials increase the legal costs associatedwith defending against employment-discrimina-tion lawsuits

Finally CRA91 strengthened a plaintiffrsquosability to use statistical evidence to prove un-lawful discrimination on the basis of ldquodisparateimpactrdquo A series of in uential 1970rsquos SupremeCourt rulings (starting with Griggs v DukePower Co) had allowed plaintiffs to show un-lawful discrimination by demonstrating that anemployerrsquos practices led to a disparate impacton protected groups even if there was no dis-criminatory intent on the part of the employerThe 1989 Wards Cove decision made it moredif cult to prove disparate impact by requiringplaintiffs to identify a particular employmentpractice leading to the disparate impact Thisdecision also partially reversed Griggs by re-quiring plaintiffs to show that the employmentpractice being challenged was not ldquonecessaryrdquoto the defendantrsquos business (see Abram 1993)CRA91 weakens this standard somewhat al-lowing plaintiffs to use statistical evidence incases where the plaintiff can show the employ-errsquos decision-making process cannot be sepa-rated into speci c practices3

2 Robert K Robinson et al (1992) provide a more de-tailed description of the Actrsquos provisions while Thomas GAbram (1993) assesses its likely impact We focus ourdiscussion and our empirical analysis on provisions ofCRA91 affecting race- and gender-based discriminationcases Changes to the ADEA as a result of CRA91 wererelatively minor

3 It does not appear that plaintiffs have increased claimsof discrimination on the basis of disparate impact since theActrsquos passage (see Abram 1993 Glen D Nager and JuliaM Broas 1994) Because plaintiffs must show disparatetreatment in order to earn punitive and compensatory dam-ages and because CRA91 greatly increased the potential

685VOL 92 NO 3 OYER AND SCHAEFER LITIGATION COSTS AND RETURNS TO EXPERIENCE

CRA91 appears to have had a signi canteffect on the litigiousness of displaced employ-ees Our analysis of lawsuits led in federalcourt shows that the number of cases allegingemployment discrimination more than doubledfrom 1991 to 1995 Similarly the number ofrace-based (gender-based) complaints led withthe EEOC increased by 13 percent (46 percent)from 1991 to 1994 and the monetary bene tsawarded in cases resolved by the EEOC in-creased by 47 percent (87 percent) over thatperiod The large increase in gender-based com-plaints relative to race-based complaints proba-bly re ects the fact that CRA91 gave womentheir rst chance to seek punitive and compen-satory damages and the fact that race-basedsuits seeking redress under the CRA of 1866bypass the EEOC and proceed directly to fed-eral court

While our analysis focuses on suits allegingwrongful termination CRA91 applies broadlyto hiring termination and many on-the-job ac-tivities Our model with slight modi cationswould apply equally well to litigation surround-ing on-the-job activities but explicit consider-ation of hiring-based protections would yieldquite different results As Richard A Posner(1987) and Donohue and Siegelman (1991)(among others) have argued the labor-marketimplications of hiring protections are very dif-ferent from those of protections against discrim-ination in termination or on-the-job activitiesTermination-based protections increase thecosts associated with hiring a protected em-ployee as the increased costs are felt only if theemployee is hired and then terminated Hiring-based protections on the other hand increasethe costs to employers associated with failing tohire a protected employee We limit our analy-sis in this way because of the dramatic shift(documented by Donohue and Siegelman1991) in the 1980rsquos away from hiring-basedemployment-discrimination litigation and to-ward termination-based suits Our own exami-nation of EEOC data from the 1990rsquos reveals acontinuation of this trend Of all gender- andrace-based complaints led with the EEOC be-tween 1992 and 1996 58 percent claimedwrongful termination 56 percent were for dis-

criminatory hiring with the rest based on on-the-job practices such as unequal pay denial ofpromotion or harassment4

II A Model of Litigation Costsand Returns to Experience

In this section we develop a stylized modelof the relationship between employment-discrimination litigation costs and the returns toexperience for protected workers and use itto examine how changes in the legal environ-ment of the type associated with the CivilRights Act of 1991 affect returns to experienceOur model indicates that the effect of CRA91on returns to experience depends crucially on (i)how employeesrsquo propensity to le employment-discrimination litigation conditional on employ-ment varies with experience and (ii) how theCRA91-induced increase in propensity to sueconditional on employment varies with experi-ence These insights suggest a link betweenreturns to experience and the age pro le ofworkers ling unlawful termination claims withthe EEOC This link then forms the basis of ourempirical analysis below

A Workers and Firms

We consider a discrete-time overlapping-generations setting in which potential workerslive for two periods A worker can be employedin both periods of his life by any of M in nitelylived rms We refer to workers in their rst andsecond periods of life as inexperienced and ex-perienced respectively Firms and potentialworkers are risk neutral and discount the futureat rate b We normalize the measure of eachcohort of workers to one

At the beginning of the rst period of a work-errsquos life the worker receives job offers from rms compares these offers to his reservationwage (which we assume to be zero) and de-cides whether to accept a job A worker whoaccepts employment may be red for either oftwo reasons First the worker may be discrim-inated against by his supervisor We assumethat while rms hire with no intention of dis-

sizes of these awards relatively few suits led since 1991have alleged disparate impact

4 These gures actually overstate the importance of hir-ing cases as approximately 10 percent of hiring-based com-plaints also charge wrongful termination and may not havebeen registered had it not been for termination


criminating against protected employees it iscostly to prevent biases of supervisors fromaffecting employment outcomes for individualworkers We argue that supervisors can easilyclaim a dismissal is performance based when inreality the supervisor is simply biased againstcertain types of employees A worker in the rstperiod of life is discriminated against and redwith probability d1

5

Second a worker can be red ldquofor causerdquo ifhis productivity is revealed to be suf cientlylow To model this we assume workers haveeither high or low ability and that workers and rms are initially symmetrically uninformedabout worker ability A worker employed inperiod i [ 1 2 of his life generates signalci which is jointly observed by the worker andthe rm and takes a value from the set L HWe suppose

Probci 5 Lzlow ability 5 a

Probci 5 Lzhigh ability 5 0

In words conditional on a worker being of lowability the signal ci is indicative of this withprobability a We x the fraction of low-abilityworkers in each cohort at 1 2 f1 0 and letthe marginal productivity of low-productivityworkers be zero6 Wages are suf ciently high sothat any worker for whom c1 5 L is redimmediately To simplify our exposition weassume that if the worker is discriminatedagainst then he is red prior to the realizationof c1 Hence the probability that an inexperi-enced worker is red due to discrimination isgiven by d1 while the probability he is red forcause is a(1 2 f1)(1 2 d1) Wages are paid at

the end of the period so that a worker who is red (for either reason) earns no wages in theperiod he is red7

Workers who are not red continue to workthroughout the rst period of life and are paidthe period t inexperienced-worker wage w1tWorkers who are not red in the rst period oflife participate in the labor market again in thesecond period while workers who are red inthe rst period do not work in the second If anexperienced worker accepts a job then he isdiscriminated against and red with probabilityd2 If the worker is not discriminated againstthen a second signal of ability c2 is generatedAs was the case in the rst period workers whoare revealed to be of low ability are red whileany worker who is retained earns wage w2t

A red worker may elect to le suit againsthis former employer A worker red in periodi [ 1 2 draws a personal cost of suing sfrom a density characterized by the cumulativedistribution function Gi

8 The worker sues if theexpected damage award conditional on ling asuit exceeds s We assume each red workerperfectly observes the reasons underlying hisdismissal but that courts observe these reasonsimperfectly Hence workers red for causesometimes win employment-discrimination law-suits while workers who were discriminatedagainst sometimes lose We let qd represent theprobability a worker who was discriminatedagainst wins a lawsuit and let qc (where qc qd)be the probability that a worker red for causewins Fired workers may sue for wages in theperiod they are red (wit) and after CRA91 somepunitive and compensatory damages (D) Condi-tional on winning a lawsuit a worker earns dam-ages of amount wit 1 D Workers red for causetherefore sue with probability Gi[qc(wit 1 D)]while workers who were discriminated against suewith probability Gi[qd(wit 1 D)]

A rmrsquos revenue in a given period depends

5 Of course rms may invest in monitoring or trainingprograms that reduce the likelihood protected workers arediscriminated against One may view CRA91 and similarlegislation as intended to force rms to make such invest-ments We discuss implications of endogenizing rmsrsquochoices over how much discrimination to permit below SeeDebra A Barbezat and James W Hughes (1990) for asimple model of endogenous discrimination

6 The assumptions that workers are either high or lowability and that low-ability workers have zero productivityare not crucial here Comparable results can be obtainedfrom a model in which employeesrsquo abilities are continuousand rms cannot adjust wages downward to match produc-tivity We can also extend our model to allow the uncer-tainty to relate to the quality of the match between theworkerrsquos skills and employerrsquos needs

7 Alternatively we could allow workers to be red afterfraction r of the rst period has elapsed These workerswould then earn wages rw1t This would make the modelmore complex without offering additional insights Morerealistic assumptions could be applied elsewhere withoutchanging our main ndingsmdashwe could for example allowthe realizations of discrimination and ability to occur simul-taneously or allow workers who are red in the rst periodto work in the second period

8 Our modeling of workersrsquo litigation choices is similarto that of Acemoglu and Angrist (2001)


on the number of high-ability experienced andinexperienced workers it employs Denote themeasure of the set of inexperienced (experi-enced) workers employed by rm m in period tas Imt (Emt) A rm employing Imt inexperi-enced workers in period t will nd that fractionf1 have high ability Because low-ability work-ers have a higher likelihood of being red intheir rst period of employment experiencedworkers are more likely to have high abilitythan inexperienced We apply Bayesrsquo rule and nd the probability an experienced worker hashigh ability is f2 5 f1(1 2 a(1 2 f1)) Wedenote rm mrsquos revenue in period t as R(f1Imt 1gf2Emt) where R is increasing and strictly con-cave We allow g $ 1 to capture the possibilitythat experience improves productivity

In making employment decisions rms con-sider all employment-related costs includingwages and the potential costs stemming fromlitigation Consider the Imt inexperienced work-ers hired by rm m in period t Fraction d1 1a(1 2 f1)(1 2 d1) of these workers are redduring period t while the remainder are re-tained through the period and paid wage w1tFraction G1[qc(w1t 1 D)] of the red-for-causeworkers le suits for unlawful termination whilefraction G1[qf(w1t 1 D)] of discriminated-againstworkers le We assume the cost to a rm ofdefending against a suit is k so the total ex-pected cost to the rm from a suit (which in-cludes expected damages plus direct costs ofpreparing a defense) is given by qj(w1t 1 D) 1 kwhere j [ c d depending on the actual reasonfor termination

Firms take current and future wages as exog-enous and choose employment levels to maxi-mize the net present value of pro ts Assumingan interior optimum the rmrsquos period t employ-ment decisions are characterized by the follow-ing rst-order conditions

(1) f1 R9

5 1 2 d1 2 a~1 2 f1 ~1 2 d1 w1t

1 d1G1 qd ~w1t 1 Dqd ~w1t 1 D 1 k

1 a~1 2 f1 ~1 2 d1 G1 qc ~w1t 1 D

3 qc ~w1t 1 D 1 k

(2) gf2R9

5 1 2 d2 2 a~1 2 f2 ~1 2 d2 w2t

1 d2G2 qd ~w2t 1 Dqd ~w2t 1 D 1 k

1 a~1 2 f2 ~1 2 d2 G2 qc ~w2t 1 D

3 qc ~w2t 1 D 1 k

In a steady-state equilibrium all workers exceptthose who were red in their rst period areemployed so we require that MImt 5 1 andMEmt 5 1 2 a(1 2 f1)(1 2 d1)9

B Factors Affecting Returns to Experience

We now address factors affecting returns toexperiencemdashthat is w2t 2 w1tmdashin this labormarket The rst-order conditions in equa-tions (1) and (2) equate the marginal produc-tivity of each cohort to the marginal cost ofemploying a worker in that cohort Threefactors may lead to differences in wages paidto experienced and inexperienced workers (i)differences in productivity (if g 1) (ii)differences in the fraction of high-abilityworkers and (iii) differences in the expectedcosts of litigation We discuss each in turnand then ask how changes in employment-discrimination law may affect the returns toexperience

First note that R9 0 is the marginal pro-ductivity of a high-ability inexperienced workerwhile gR9 is the marginal productivity of ahigh-ability experienced worker If g is strictlygreater than one then experience results ingreater productivity and hence higher wages

Second the rmrsquos expectation of a workerrsquosability depends on the workerrsquos experienceFirms have no information regarding abilitylevels of inexperienced workers hence theprobability that an inexperienced worker hashigh ability is f1 However experienced work-ers remain in the labor force only if they werenot red in their rst period of employment

9 Note that in order for both types of workers to beemployed it must be that wages are such that the right-hand side of (1) is equal to f1gf2 times the right-handside of (2)


Because low-ability workers are more likely tobe red in the rst period than high-abilityworkers (as long as a 0) the share of expe-rienced workers with high ability is greater thanf1 This means greater demand and higherwages for these workers

Third experienced and inexperienced work-ers differ in the expected costs they impose onthe rm from employment-discrimination liti-gation For a worker in period i of his life theexpected cost to the rm from litigation on thepart of that worker is

(3) d i G i qd ~w it 1 Dqd ~w it 1 D 1 k

1 a~1 2 fi ~1 2 di G i qc ~wit 1 D

3 ~qc ~wit 1 D 1 k

A number of potentially opposing effects are inplace here Expected litigation costs are increas-ing in wit as workers earning higher wages earnhigher damage awards and are as a result morelikely to sue conditional on being displacedBecause returns to experience are positive thiseffect works in the direction of higher expectedlitigation costs for experienced workers How-ever litigation costs are also decreasing in fi the likelihood a worker has high ability Be-cause inexperienced workers are more likely tohave low ability and hence more likely to be red for cause this effect works in the directionof higher expected litigation costs for inexperi-enced workers Expected litigation costs alsodepend on di the likelihood a worker is dis-criminated against and G i the distribution ofpersonal costs of litigating If di dj or if GjGi (in the sense of rst-order stochastic domi-nance) then these effects work in the directionof higher expected litigation costs for workersin period i of life As we have no a prioriexpectation as to how d and G vary withexperience we conclude that these two ef-fects could work in the direction of higherlitigation costs for either experienced or in-experienced workers

C Effects of Changes in theLegal Environment

We next attempt to incorporate the effects ofCRA91 into our model While damages avail-

able to pre-1991 plaintiffs were limited to backpay (implying D 5 0) post-1991 plaintiffs canearn both punitive and compensatory damagesWe therefore model CRA91 as increasing D10

This increase in potential damage awards clearlyraises the cost of employing both inexperiencedand experienced workers In order to determinehow the returns to experience are affected weask where the cost increase is larger as wagesfor this group will be depressed relative to theother

To examine this issue we differentiate ex-pected litigation costs in (3) with respect to Dand ask how the resulting expression varies withi The derivative is given by

(4) d i qd G i qd ~w it 1 D

1 ~di qd gi qd ~w it 1 D

3 qd ~w it 1 D 1 k)

1 a~1 2 fi ~1 2 di qc G i qc ~w it 1 D

1 ~a1 2 fi 1 2 di qc g i qc~w it 1 D

3 qc ~w it 1 D 1 k)

where gi is the probability density function as-sociated with Gi Increases in D affect rmsrsquoexpected litigation costs in two ways Firstemployees who sue successfully impose highercosts on the rm in the form of higher damageawards Mathematically this effect is embodiedin the rst and third terms of (4) which are theprobability an employee is displaced and suestimes the derivative of the expected cost to the rm conditional on being sued Second theprospect of higher damage awards induces moredisplaced workers to le suit The second andfourth terms of (4) are the product of the like-lihood of displacement the increase in the like-lihood of suing conditional on displacementand the expected cost to the rm conditional onbeing sued

Clearly the increase in expected litigationcosts associated with CRA91 could be larger for

10 We can obtain similar results focusing on the provi-sion of CRA91 that allows either side to seek a jury trial Asjuries are perceived to favor the claims of individuals overthose of corporations we model this as an increase in bothqd and qc the likelihoods that suits are successful


either experienced or inexperienced workersThe higher wage for experienced workers im-plies that any increase in the likelihood of suingconditional on displacement is more costly forthese workers However the higher likelihoodof displacement for inexperienced workersmeans that the increase in damages is morecostly for these workers

While our model does not yield an unambig-uous comparative static regarding the link be-tween CRA91 and returns to experience it doesallow us to make two observations First it isapparent from (4) that one key determinant ofthe link between CRA91 and returns to experi-ence is how the propensity to sue conditional onemployment varies with experience If inexpe-rienced workers are considerably more likely to le suit conditional on being employedmdashthat isif

(5) d i G i qd ~w it 1 D

1 a~1 2 fi ~1 2 di G i qc ~w it 1 D

is decreasing in imdashthen the increase in damagesassociated with CRA91 is more costly for theseworkers If inexperienced workers are morelikely to be discriminated against or face lowerpersonal costs of ling suit then employers willdiscount wages for inexperienced workers rela-tive to experienced after 1991 and the returns toexperience should increase11

Second another key determinant of this linkis how the increase in propensity to sue con-ditional on employment varies with experi-ence If the increase in suits by inexperiencedworkers is greater than that for experiencedmdashthat is if dig i[qd(wit 1 D)] 1 a(1 2 fi)(1 2 di) gi[qc(wit 1 D)] is decreasing in imdashthen this also favors a larger increase in litiga-tion costs for inexperienced workers and anincrease in returns to experience Our modeltherefore suggests that in order to understandhow CRA91 affects returns to experience wemust rst examine how rates of employment-discrimination litigation vary with age amongprotected workers and how the response of

litigation rates to the Civil Rights Act of 1991varied with age

D Extensions

Before turning to our empirical analysis webrie y describe implications of two simple en-richments of our model First our analysis sug-gests two ways a rm may respond to increasesin potential costs of employment-discriminationlitigation It may adjust its demand for protectedworkers (leading to the changes in wages weillustrate above) but it may also invest in mon-itoring or training programs that reduce the like-lihood protected workers are discriminatedagainst Our model emphasizes the rst andsuggests that CRA91 should negatively impactthe employment of protected workers How-ever rms may also adjust their monitoringefforts in a way that introduces an offsettingpositive effect on employment If we let di be adecreasing function of the rmrsquos level of mon-itoring then passage of CRA91 would cause rms to revisit monitoring decisions Increasedmonitoring could cause discrimination-basedterminations to fall after the passage of the Actwhich would yield con icting effects on overalllevels of protected-worker employment How-ever as long as increased monitoring does notcompletely offset rmsrsquo exposure to increasedlitigation costs our predictions regarding changesin relative wages are not affected

Second while we have for ease of presenta-tion assumed labor supply is completely inelas-tic removal of this restriction does yield oneadditional implication Under the assumptionsthat (i) labor supply is somewhat elastic and (ii)the increase in expected litigation costs associ-ated with an increase in D is larger for inexpe-rienced workers then it is possible to constructexamples in which w2t increases in response tothe increase in D This effect arises as rmssubstitute away from inexperienced workers be-cause of the high potential costs of litigationand bid up the wages of experienced workersThis observation implies that average wageswithin a given period may not be greatly af-fected by increases in D However this does notmean that protected workers are not harmedbecause wages are redistributed from youngerto older workers the present value of a workerrsquoslifetime earnings falls This suggests studiesexamining the effects of employment protec-

11 As we discuss below there is evidence to suggest thatprotected workersrsquo perceptions of employment discrimina-tion vary considerably with age


tions on average wages without also consider-ing how protections may redistribute wagesamong protected workers may miss part of theeffect

III Age Patterns in WrongfulTermination Complaints

Our model suggests that the effect of CRA91on returns to experience is partially determinedby the relationship between experience and thepropensity to sue We therefore begin our em-pirical analysis by examining the age distribu-tion of employees making discriminationclaims Using data from the EEOC and the CPSwe measure the share of employed protectedworkers who le wrongful termination com-plaints and compute how this share varies withage12

Our EEOC data set lists a range of factsregarding each complaint including the date thecomplaint was rst led the ldquobasisrdquo of thecomplaint (eg race gender disability) andthe ldquoissuerdquo (eg hiring discharge harassment)The data also include demographic informationsuch as the plaintiffrsquos state of residence genderrace and (for 70 percent of plaintiffs) age Weanalyze gender-based cases brought by womenand race-based cases brought by black men thatwere rst led with the EEOC between 1988and 199513 To eliminate age-based cases andconcentrate on workers likely to be attached tothe labor force we look exclusively at plain-tiffs aged 20 to 40 at the time of complaintAlso because our model focuses explicitly ontermination-based litigation we consider onlytermination-based complaints There were a to-tal of 113283 gender-based cases brought bywhite women aged 20 to 40 and 118779 race-based cases brought by black men aged 20 to

40 Of these a total of 149489 (644 percent)were wrongful termination cases and compriseour nal sample14

We use the Annual Demographic File of theMarch CPS to estimate the number of employedwhite women and employed black men of eachage between 20 and 40 in each year between1988 and 1995 (where a worker is employed ifhe or she reported working at least 1000 hoursduring the year) We create counts of workersby ageyearprotected group and use thesecounts and the number of complaints in eachageyeargroup cell to determine by cell thepercentage of employees who le a complaintwith the EEOC These complaint rates indi-cate the approximate probability that a personof a given age who works at least 1000 hoursin a given year les a wrongful terminationcomplaint

Figures 1(a) and 1(b) show the complaintrates by age for white women and black menrespectively during 1990 and 1993 We chosethese years as representative pre-CRA91 andpost-CRA91 years the agecomplaint patternsare similar in every year from 1988ndash1995 soexamining these two years is suf cient Thecomplaint rate is much higher for black menthan for white women Each year the EEOCreceived a gender-based wrongful terminationclaim from approximately one out of every2500 to 3500 employed white women but theproportion is one out of 400 to 600 for blackmen Also as suggested in Section I the rate ofcomplaint for both groups is noticeably higherin 1993 than in 1990 The increase in com-plaints is more dramatic for women than forblacks which could be related to the attentiondrawn to gender-based discrimination by the1991 Clarence Thomas con rmation hearingsAlternatively the smaller increase in the blackEEOC complaint rate may be due to the fact that

12 Except when ling under the CRA of 1866 all work-ers seeking redress using CRA91 must start by ling acomplaint with the EEOC

13 Approximately 18 percent of gender-based cases arebrought by men Approximately 80 percent of race-basedcases are brought by blacks 10 percent by whites and therest are split among Asians Native Americans and othersSome complaints allege more than one basis (that is aperson may claim both age and gender discrimination) butover 95 percent of the complaints in the age and basisgroups that we analyze claim a single basis Our results arenot altered if we include complaints with multiple bases orif we examine all nonhiring-based complaints

14 There are at least two limitations of this EEOC dataFirst the age of the plaintiff is missing for approximately 30percent of the observations While we have no reason tobelieve there is any systematic difference between the agesof the complete sample and the missing age sample wewant to be careful about drawing conclusions from a samplelimited in this way Second as discussed above race-basedcomplaints can be led under the CRA of 1866 directly infederal court CRA91 made this a viable option in termina-tion suits so it is unclear how representative EEOC com-plaints are of all post-CRA91 race-based discriminationcomplaints


CRA91 allows race-based termination cases tobypass the EEOC

The age patterns in EEOC complaints arevery different for the two groups As depicted inFigure 1(a) complaint rates decline steadily andsteeply for white women in their 30rsquos Thecomplaint rates in 1990 and 1993 were 0320and 0420 respectively per 1000 for whitewomen in their 20rsquos but only 0222 and 0328per 1000 for white women in their 30rsquos Overthe full 1988ndash1991 (1992ndash1995) period theyearly complaint rate was 0290 (0389) for25-year-old white women and 0192 (0331) for35-year-old white women This pattern of de-creasing complaint rates with age does not holdhowever for black men As shown in Fig-ure 1(b) complaint rates increase slowly andsteadily as black men age In 1990 and 1993 thecomplaint rates were 202 and 218 respec-

tively per 1000 for black men in their 20rsquos but238 and 256 for those in their 30rsquos Similarlyover the full 1988ndash1991 (1992ndash1995) periodthe EEOC received 174 (179) complaints per1000 25-year-old black men per year and 238(247) per 1000 35-year-old black men peryear

We expect the returns to experience for agiven protected group to increase as a resultof the passage of CRA91 if the increase inlitigation-related costs of employment is smallerfor more experienced workers In Section II wemodeled these costs explicitly and argued that if(i) the likelihood of ling a complaint condi-tional on being employed decreases with expe-rience or (ii) the increase in the propensity tosue conditional on being employed associatedwith increased damage awards decreases withexperience then the increase in litigation-related costs of employment may be decreasingwith experience For white women it appearsthat the rst of these conditions holds There isno evidence that the increase in the propensityto sue associated with CRA91 varies with agefor this group but it is apparent that conditionalon employment younger women are morelikely to le complaints with the EEOC Forblack men it appears that neither conditionholds Older black men are more likely to lewith the EEOC and it does not appear that theincrease in propensity to sue associated withCRA91 varied by age

These differences in the age patterns ofEEOC complaints for white women and blackmen lead us to expect different patterns in re-turns to experience as a result of CRA91 Our nding that young white women are more likelyto le employment-discrimination litigationthan older white women leads us to expect thepassage of CRA91 to result in an increase inthe returns to experience for women Becausepropensity to sue trends upward with age forblack men our analysis does not offer a de n-itive prediction for this group

Though this issue is not central to our anal-ysis Figure 1 does raise the question of why theage trend in EEOC complaints differs so mark-edly across these two groups Our model inSection II suggests three potential answers tothis question First the age pattern of com-plaints may differ across these groups if the agepattern of job displacement differs acrossgroups If the rate of displacement drops more

FIGURE 1 EEOC COMPLAINTS PER 1000 EMPLOYED

WORKERS BY AGE FOR WHITE WOMEN AND BLACK MEN


sharply with age for white women than forblack men then we would expect the rate ofcomplaints to drop more sharply with age aswell Using our CPS data however we foundthe age pattern of displacement rates to be verysimilar across these groups

Second the age pattern of complaints maydiffer across groups if the rates of actual dis-crimination vary differently with age acrossgroups If d1 d2 for white women but not forblack men then displacements among young whitewomen will be disproportionately discrimination-based compared to displacements among olderwhite women This could then generate the pat-tern we observe as discriminated-against em-ployees are more likely to le suit than employees red for cause This may occur if for exampleemployers exaggerate the effect of childbearingon productivity and discriminate against womenof childbearing age (see Michael Selmi 2000)In addition Heather Antecol and Peter Kuhn(2000) nd that womenrsquos perceptions of dis-crimination vary considerably with age Usingdata from a Canadian survey of displaced work-ers they show that women aged 25 to 34 aresigni cantly more likely to feel they were vic-tims of gender-induced harm in the workplacethan women aged 35 to 44

Third the age pattern in complaints may dif-fer across groups if the personal costs of lingsuit vary differently with age across these twogroups While we were unable to nd any re-search in economics exploring determinants ofemployeesrsquo litigation decisions this area hasbeen explored by scholars in the sociology oflaw Michele Hoyman and Lamont Stallworth(1986) and Phoebe A Morgan (1999) nd thatwomen are more likely to seek redress throughthe legal system when they have signi cantparental responsibilities If younger women aremore likely to have dependent children com-pared to older women and thus more likely to le with the EEOC then the pattern we observecould be explained Alternatively gender-baseddifferences in government assistance programsand social norms may cause women who be-come disenchanted as they age to nd nonpar-ticipation to be a more viable option than wouldblack men Hence older women who condi-tional on working would be likely to lecomplaints may instead elect not to seek em-ployment Our EEOC data do not containenough demographic information to allow us to

validate or refute these potential explanationsfor differing age trends in complaint rates

IV Empirical Analysis ofLabor-Market Outcomes

A Data and Methodology

We examine effects of CRA91 on labor-market outcomes for protected workers usingdata from the 1988 through 1996 Annual De-mographic File of the March CPS The MarchCPS like all CPS monthly surveys asks aboutcurrent employment status including hoursworked last week full-timepart-time status in-dustry and occupation The March CPS alsogathers detailed information about the respon-dentrsquos employment in the previous calendaryear such as weeks and hours of work andwages

We focus on three measures of employmentoutcomes whether the respondent worked fulltime how many hours the respondent workedand what hourly wage the respondent earnedFirst we de ne survey respondents who workedat least 35 hours on all jobs in the week beforethe CPS interview as ldquocurrently employedrdquoSecond we measure the total hours worked inthe calendar year before the interview as theproduct of the number of weeks the personreported working in the previous year and thereported hours per week Third we calculatedthe hourly wage for the previous year by divid-ing the reported total wages for the year by thetotal hours worked15 Note that the years wediscuss refer to the year about which the respon-dent was asked and not necessarily the surveyyear Employment in year t refers to employ-ment status reported in March of year t whileyear t wages and hours refer to the wages andhours reported retrospectively for year t inMarch of year t 1 1

We limit our analysis to CPS respondentsbetween the ages of 25 and 39 This restrictionfocuses on those with a strong labor-force at-tachment minimizes the effects of schooling

15 The maximum annual earnings in the CPS is $99999so our wage variable is right-censored However this onlyaffects 15 percent of all observations in our wage regres-sions The rate of top-coding varies by year peaking in the1995 CPS (1994 earnings) where 24 percent of the obser-vations in our wage regressions are top-coded


and keeps the comparison group (white men)clean by removing ADEA-protected workers16

We de ne 1987ndash1991 observations as ldquopre-CRA91rdquo and 1992ndash1996 observations as ldquopost-CRA91rdquo17 Summary statistics in Table 1 showthat just over two-thirds of survey respondentsreported full-time employment and the averagework week was 41 hours From columns (2) and(3) of the table it appears there are no notewor-thy differences between the ldquopre-CRA91rdquo andldquopost-CRA91rdquo samples Columns (4)ndash(6) showthat employment rates hours worked andwages are higher for white men than for eitherof the protected groups

Our primary methodology is to measurehow the differences between protected- andunprotected-worker employment-outcome mea-sures changed from the period shortly beforeCRA91 to the period shortly after That is wemeasure the ldquodifference-in-differencesrdquo of forexample black and white wages before andafter the law The critical assumption neededfor our analysis is that in the absence of

CRA91 wages would not have changed in away that differed by race or gender over the1988ndash1995 period At least two non-CRA91factors may have contributed to changes in em-ployment outcomes for protected workers overthis period and we will consider these factors inframing and interpreting our results First omit-ted variable bias could result from other policychanges particularly the minimum wage in-creases in 1990 and 1991 Second there wereunderlying trends in the returns to skill and theincome distribution and the effects of thesetrends may have differed across protected andunprotected groups

B Wage and Employment Effects of CRA91

We rst estimate the effects of CRA91 on thethree employment-outcome variables (employ-ment hours worked and wages) separately forboth protected groups The comparison groupthroughout is non-Hispanic white men under40 years old White men le discriminationclaims far less frequently than members of ei-ther protected group so we expect the Act tohave a much smaller effect on these workersrsquoemployment outcomes18 We measure the effectof CRA91 by estimating

16 We expanded the upper limit of the age range to 50and obtained essentially the same results

17 This is not a perfect division between pre- and post-CRA91 The 1991 measures of wages and hours include 40days of post-CRA91 time and the data for these observa-tions were recorded in March 1992 after the law wasenacted However our results were basically unaffectedwhen we redid our analysis without the 1992 survey obser-vations

18 The Act did affect the legal status of disabled whitemen However fewer than 5 percent of workers in the agegroup we study are disabled (Acemoglu and Angrist 2001)

TABLE 1mdashSUMMARY STATISTICS FROM THE 1988ndash1996 MARCH CPS

Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)

Total observations 254320 150275 104045 118091 122968 13261Percentage currently employed 672 674 670 799 552 659Hoursweek 4075 4065 4089 4447 3647 4142

(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128

(23) (24) (22) (24) (22) (22)State unemployment rate 59 59 58 59 59 61

(16) (17) (15) (16) (16) (15)

Notes Data are from 1988ndash1996 March CPS limited to black men and non-Hispanic white men and women aged25ndash39 Column (2) [Column (3)] includes observations before 1992 [after 1992] Hoursweek and wages are averagesover all nonzero observations Standard deviations are in parentheses The state unemployment rates are reported aspercentages


(6) y it 5 a 1 apdp

1 Ot 5 87

95

b t ~d t 3 dp 1 xi 1 laquo it

where yit is hours worked or log hourly wagesin year t for person i dp is a protected indicatorvariable dt is an indicator for year t and xi is avector of control variables Examining how thebt coef cients differ for pre- and post-CRA91years tells us how employment outcomes wereaffected by the law To get at the prepost effectmore directly we can adjust equation (6) to

(7) y it 5 a 1 ap dp

1 bpost~dpost 3 dp 1 xi 1 laquo it

where dpost is an indicator for ldquopost-CRA91rdquoWhen we measure the effect of CRA91 onemployment status y is an indicator variableequal to one if the respondent reports full-timeemployment and we estimate the coef cientswith a probit speci cation If employment orhours worked is the dependent variable thevector of control variables xi includes indica-tors for year state high school and collegegraduation ve-year age categories half per-centage point intervals in state unemploymentrates marital status and interactions betweenstate unemployment and protected status indi-cators and between marital status and female Iflog wage is the dependent variable then xi

includes experience experience squared and ed-ucation as well as indicators for year statestate unemployment rate and unemploymentrateprotected status interactions Some speci -cations also interact a linear time trend withprotected status for separate trends in employ-ment outcomes for protected and unprotectedworkers

Panel A of Table 2 and Figure 2(a) displaythe results of estimating equations (6) and (7)using white women under 40 as the protectedgroup and white men under 40 as the com-parison group In the gure we plot the bt

coef cients from estimation of equation (6)while the table reports coef cients from esti-mation of equation (7) Our ndings con rmthe well-established facts that women partic-ipate in the labor market at signi cantly lower

rates than men that they work fewer hoursand that they earn less (about 27 percent lesson a regression-adjusted basis) However asthe table and gure show there was a distincttrend towards increasing female labor-forceparticipation hours and wages during theperiod we study Log wages rose by 4 percentmore for women than men over the pre-CRA91 period to the post-CRA91 periodwhile womenrsquos employment grew 2 percentmore than menrsquos

These effects cannot be attributed to CRA91however In columns (2) (4) and (6) wecontrol for female-speci c trends in employ-ment hours and wages which leaves onlysmall (and insigni cant) prepost differencesin the hours and wage regressions Whilethis difference remains signi cant in the em-ployment regression careful examination ofFigure 2(a) shows the growth in female em-ployment was well underway by March 1991predating CRA91 considerably Visual in-spection of Figures 2(b) and 2(c) con rmsthat any effects of CRA91 are minor com-pared to the ongoing growth in womenrsquoshours and wages relative to those of menOverall Table 2 and Figure 2 indicate thatCRA91 had minor effects on average employ-ment hours and wages for white women

The results look different for black men butthis is primarily due to the different underlyinglabor-market trends Panel B of Table 2 con- rms the signi cant difference between blackand white employment outcomes even control-ling for other observable characteristics Theevidence suggests a mild negative effect ofCRA91 on average black employment andhours The blackpost-interaction term is nega-tive for the employment probit [column (2)] andnegative and signi cant for the hours regression[column (4)] Figures 3(a) and 3(b) also suggestthat black employment and hours were affectedby CRA91mdashafter trending up through 1991black employment and hours worked droppedrelative to whites starting in 1992 The effect isnoteworthy but not large representing about a2-percent decline in black hours worked Thelast two columns of the table and Figure 3(c) donot indicate that black wages changed as a resultof CRA91 Overall the evidence in Table 2 andFigure 3 is consistent with CRA91 having had amild negative effect on the employment pros-pects of black men


C CRA91 and Returns to Experience

While average wages for protected groupsappear to be unchanged as a result of CRA91our analysis in Section II suggests that the Actmay alter the returns to experience and thusredistribute wages within groups of protectedworkers From our analysis of the age distribu-tion of EEOC claims we expect this effect to bestronger for white women than for black menWe expect no change in returns to experience

for unprotected workers as a result of the Actso examining the relative changes in returns toexperience for protected and unprotected work-ers allows us to control for other factors affect-ing returns to experience during the early1990rsquos

In our stylized model employers learn overtime about the productivity of workers In actualwork environments it is unclear whether thislearning by employers occurs only when em-ployees are actually on the job or if information

TABLE 2mdashCHANGES IN PROTECTED WORKERSrsquo EMPLOYMENT OUTCOMES POST-CRA91

A White Women Compared to White Men

Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]

Female 3 linear trend 200001 311 00098(00046) (274) (00021)

[200001]Observations 241059 241059 241059 241059 156552 156552Log-likelihoodR2 2143857 2143857 02109 02109 02485 02486

B Black Men Compared to White Men

Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]

Black 3 linear trend 00153 1541 200023(00103) (576) (00046)[00048]

Observations 131352 131352 131352 131352 100578 100578Log-likelihoodR2 270501 270501 01060 01060 02147 02147

Notes Data from 1988ndash1996 March CPS limited to black men and non-Hispanic white men and women aged 25ndash39 PanelA (B) compares white women to white men (black men to white men) Full-time employment 5 1 if individual reportsworking $ 35 hours in week before CPS interview Columns (1)ndash(2) are probits and include indicators for high-school andcollege graduation ve-year age categories state year marital status half-percentage-point intervals in state unemploymentrate and protected statusstate unemployment interactions and marital statusfemale interactions Bracketed terms areprobability derivatives or estimated change in probability that the dependent variable takes value 1 Hours worked is theproduct of average weekly hours and number of weeks worked over the preceding year Columns (3)ndash(4) are ordinary leastsquares (OLS) and include the same controls as in columns (1)ndash(2) Log wage is the log of average hourly wage for the yearColumns (5)ndash(6) are OLS limited to individuals working 1500 or more hours during the speci ed year Controls includeexperience experience squared education and indicators for state year and half-percentage-point state unemployment rateintervals and protected statusstate unemployment interactions Coef cients on ldquoFemalerdquo and ldquoBlackrdquo are for the pre-CRA91period with state unemployment rate of 6 percent Standard errors are in parentheses


is also revealed from the frequency and lengthof nonemployment spells To allow for bothpossibilities we would ideally like to use bothldquopotential experiencerdquo (which we de ne asage 2 years of education 2 6) and ldquoactualexperiencerdquo as measures of workforce experi-ence Unfortunately the CPS does not contain

enough historical employment information aboutindividual respondents to allow us to measureactual experience

We therefore use historical CPS data to de-rive two proxies for actual experience Our rstproxy applies a method similar to that used byTricia Gladden and Christopher Taber (2000)

FIGURE 2 ESTIMATES OF bt FROM EQUATION (6)WHITE WOMEN COMPARED TO WHITE MEN

FIGURE 3 ESTIMATES OF bt FROM EQUATION (6)BLACK MEN COMPARED TO WHITE MEN


We gather labor-force participation rates fromthe 1964ndash1995 March CPS and divide respon-dents into cells along four dimensions genderrace (white or black only) education (less thanhigh school high-school graduate some col-lege college graduate) and year of birth Wethen sum average work experience across yearsfor each cell and use this as a proxy for actualexperience gathered by workers in this cell Werefer to this proxy for a workerrsquos actual experi-ence as ldquoCell Average Irdquo

A problem with this measure however isthat the individuals we observe to be working1500 or more hours in a year (and who there-fore comprise our sample) are likely to havedifferent experience pro les than the averageCPS respondent in a given cell If there is per-sistence in individual employment status thenaverage actual experience across all individualsin a cell is likely to be lower than the averageactual experience of individuals in a cell whoare currently employed19 We devise a secondproxy for actual experience which we referto as ldquoCell Average IIrdquo in response to thisconcern To construct this proxy we assumethat the individuals in any given gender-race-education-birth-year cell who work the mosthours in year t are also those who worked themost hours in year t 2 1 t 2 2 etc While the rst cell-average actual experience proxy as-sumes each individualrsquos labor-force participa-tion is completely independent from year toyear the second assumes the correlation ofacross-year participation is maximized (See theAppendix for a detailed description of the con-struction of these measures) Neither measure isperfect but we believe they represent reason-able lower and upper bounds respectively onaverage actual experience for employed indi-viduals in each cell20

To examine the effects of CRA91 on returnsto experience we estimate the following

~8 w it 5 a 1 ap dp

1 Ot 5 87

95

b t ~d t 3 dp 3 e it 1 xi 1 laquo it

where wit and eit are log hourly wages andexperience (either potential experience or thecell-average proxy for actual experience) re-spectively of individual i in year t Our vec-tor of control variables (xi) is described inTable 321 As above some speci cations in-clude interactions between a linear time trendprotected status and experience to allow thetrend in returns to experience to differ for pro-tected and unprotected workers The coef cientbt represents the amount by which the returns toexperience for the protected group exceeds thatof the comparison group in year t where the rst yearrsquos bt is normalized to zero Highervalues of bt after CRA91 would imply that theincrease in returns to experience was larger forprotected workers To assess prepost differ-ences more directly we also estimate

(9) w it 5 a 1 ap dp 1 bpost~dpost 3 dp 3 e it

1 xi 1 laquoit

As above we limit the analysis to survey re-spondents between the ages of 25 and 39

WomenmdashWe estimate equations (8) and (9)with white women as the protected group andwhite men as the comparison group In Panel Aof Table 3 we list the results from estimation ofequation (9) while in Figure 4(a) we plot the btcoef cients obtained when estimating equation(8)

In column (1) we use potential experienceand obtain an estimate of bpost of 00031 whichis signi cant at better than the 2-percent levelThe magnitude of this estimate implies thatrelative to white men the wage premium for30-year-old women relative to 25-year-oldwomen was 15 percent higher after CRA91than before Plots of the individual year effects

19 Much of the analysis of labor-force attachment hasfocused on women James J Heckman and Robert J Willis(1977) Claudia Goldin (1990) and Kathryn Shaw (1994)document considerable persistence in individual womenrsquosemployment status

20 Using OLS to regress individual-level wages on cell-average experience would produce understated standard er-rors because cell-average experience is actual experienceplus unobserved noise We therefore follow Brent R Moul-ton (1986) and run GLS assuming a random group effect

21 To insure that our results do not re ect the interactionbetween experience and other variables we reran our anal-ysis with controls for experience interacted with educationand with the state unemployment indicators This had atrivial effect on our estimates


in Figure 4(a) (with 1991 normalized to zero)indicate that this premium started to increase in1992 which coincides with the passage of theAct To verify that this effect is not merely thecontinuation of an upward trend in female re-turns to experience we estimate a speci cationthat includes a linear trend and present results incolumn (2) Our estimate of the effect ofCRA91 remains signi cant and grows slightlyin magnitude The corresponding estimates us-ing either cell-average experience [see columns(3)ndash(6)] also show that womenrsquos returns to ex-perience increased following CRA91 The esti-mates reported in columns (4) and (6) indicatethat the premium to being a woman whose cellaveraged ve years of experience more thananother group increased by 42 percent and 58percent respectively after CRA91 Given thatcell-average experience increases more slowlywith age than potential experience the esti-mates using the potential and cell-average mea-sures are fairly consistent

From the results in subsection B it is appar-

ent that during the time period we study labor-force participation rates were increasing forwomen relative to men This trend in participa-tion implies that a post-CRA91 woman of agiven potential experience level is likely to havemore actual labor-force experience than a pre-CRA91 woman of the same potential experi-ence If employers offer higher wages to womenwith more actual experience then we mightexpect this participation trend to result in in-creasing returns to potential experience over thetime period we study22 While our control for atrend in womenrsquos returns to experience and ourresults using cell-average experience suggestthat this effect is not driving our results weoffer two additional robustness checks hereFirst we perform a ldquoplacebordquo analysis wherewe assess the prepost difference in returns to

22 OrsquoNeill and Polachek (1993) document increases inwomenrsquos relative returns to potential experience in the 1980rsquosand show this can be attributed to increased actual experienceresulting from increased labor-force participation

TABLE 3mdashCHANGES IN PROTECTED WORKERSrsquo RETURNS TO EXPERIENCE POST-CRA91


Experience measurePotential

(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)

Female 3 experience 3 post 00031 00045 00110 00100 00010 00115(00011) (00022) (00024) (00048) (00014) (00027)

Female 3 experience 3 trend 200003 00002 200024(00004) (00009) (00005)

Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)

Black 3 experience 3 post 200042 200012 200011 200046 200044 200024(00025) (00048) (00043) (00084) (00033) (00062)

Black 3 experience 3 trend 200007 00008 200004(00009) (00016) (00012)

Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136

Notes Dependent variable is the log of average hourly wage for the year Data from the 1988ndash1996 March CPS limited toblack men non-Hispanic white men and women aged 25ndash39 and working 1500 hours or more in the speci ed year PanelA (B) compares white women to white men (black men to white men) Controls include experience experience squarededucation indicators for year state and state unemployment rate interactions between protected status and unemploymentrate experience and experience squared and interactions between year and experience experience squared and protectedstatus In columns (1)ndash(2) regressions use OLS and experience is de ned as age 2 education 2 6 In columns (3)ndash(6)regressions use GLS and experience is de ned as the average experience for 1965ndash1995 March CPS respondents with thesame gender race education and year of birth Cell Average I and II measures are computed using different assumptionsabout the persistence of labor-force participation See Appendix for details Standard errors are in parentheses


experience as if the CRA has been passed at theend of 1987 If increasing female labor-forceparticipation leads mechanically to increasingreturns to experience then we would expect tosee a prepost change using 1987 as the breakpoint Second we use the National LongitudinalSurvey of Youth (NLSY) to obtain a smallsample of workers for whom we are able tomeasure actual workforce experience

To perform our placebo analysis we expandour data to include the 1986ndash1991 MarchCPS23 We analyze these data as though a lawthat might have affected womenrsquos returns to

experience was enacted at the end of 1987 thatis we run the regressions presented in Panel Aof Table 3 where the ldquoprerdquo and ldquopostrdquo periodsare 1985ndash1987 and 1988ndash1990 respectively Ifthis analysis yields positive changes in wom-enrsquos returns to experience then we would ques-tion whether the effects we document areattributable to CRA91 We nd using all threemeasures of experience no evidence that therewas an increase in womenrsquos relative returns toexperience around 1987 The analysis yieldsldquopost-1987rdquo coef cients (which we do not re-port) that are negative and statistically insignif-icant The estimated linear trend in womenrsquosreturns to experience is positive and signi cantusing each of the two cell-average experiencemeasures and negative and insigni cant usingpotential experience Furthermore in regres-sions that combine post-CRA91 data with1985ndash1990 data we nd that the ldquopost-1991rdquoeffect is statistically distinguishable from theldquopost-1987rdquo effect That is we can reject thehypothesis that the increase in womenrsquos returnsto experience using 1991 as a break point isequal to that using 1987 as a break These ndings suggest that the positive and signi cantldquopostrdquo coef cients presented in Table 3 are notfollowing mechanically from increasing femalelabor-force participation

As a second check we gather data from theNLSY24 The main advantage of this survey forour purposes is that we can follow individualsthrough their careers and measure actual labor-market experience more accurately This comesat the cost of a much smaller sample TheNLSY sampled fewer than 12000 individualsduring the years we study while the CPS sur-veyed each resident of 60000 different house-holds Also because the NLSY is administeredto the same people each year the sample ageseach year and we have to drop the youngestpre-CRA91 observations and the oldest post-CRA91 observations in order to measure thereturns to experience over comparable ranges ofexperience

23 If the effects of CRA91 were immediate and involvedonly a discrete shift with no effect on the trend in returns toexperience then we could use either the pre-CRA91 or thepost-CRA91 period to see how wages might have beenchanging in the absence of the law However becauseCRA91 may affect wages with lag and may induce sometrend shifts we need to concentrate on the period before thelaw to remove its effects from the placebo analysis

24 Because we use the NLSY only for comparison pur-poses we do not provide background information on thisdata source Henry S Farber and Robert Gibbons (1996)offer details on using the NLSY to measure actual experi-ence Descriptions of our experience measure and samplerestriction criteria are available from the authors upon re-quest



We estimate equations (8) and (9) using9447 individual-year wage observations forwhite men and women between the ages of 27and 31 The NLSY-estimated prepost change infemale returns to experience is 00066 logpoints which is similar in magnitude to theestimates obtained using cell-average experi-ence in Table 3 However this coef cient is notsigni cantly different from zero The annualeffects [bt from equation (8)] are displayed inFigure 4(b) While each year effect is measuredwith considerable sampling variance the over-all pattern is similar to that obtained from CPSdata in Figure 4(a) While we cannot place agreat deal of con dence in any of our NLSY-based estimates what evidence there is suggeststhat the increase in female returns to experiencearound the time of CRA91 is not the resultof a mismatch between actual and potentialexperience

Finally we ask whether the magnitudes ofour estimates of womenrsquos relative increase inreturns to experience are reasonable given themagnitudes of expected litigation costs Thisis naturally an imprecise discussion becausethe exact estimates of costs stemming fromemployment-discrimination litigation are notavailable The estimate in Table 3 column (1)suggests that all else equal a 30-year-old wom-anrsquos wage was 15 percent higher than a 25-year-old womanrsquos after CRA91 than it wouldhave been before CRA91 The median annualwage for 25-year-old white women in our sam-ple employed in full-time jobs in 1994 was$17000 Our estimate suggests therefore thatthe increase in annual expected costs of litiga-tion and litigation prevention associated withCRA91 were $200ndash$300 higher for a 25-year-old woman than for a 30-year-old woman Ap-plying James N Dertouzos and Lynn AKarolyrsquos (1992) estimate that the costs of dam-age awards themselves re ect only about 1 per-cent of the total costs of potential litigation thistranslates to $2 or $3 of actual damages

To compare this gure to amounts actuallyawarded consider that the EEOC awarded $44million in gender-discrimination claims in 1994which was nearly double the 1991 amountFederal courts awarded over $200 million indamages in all employment-discriminationcases in 1994 although most of these caseswere originally led or involved discriminationbefore CRA91 was enacted New employment-

discrimination suits led in federal court in1994 demanded over $5 billion in damages a165-percent increase from 1990 We are unableto determine the shares of these gures thatstem from gender-based cases however Whenstate fair employment practice judgements andstate court damages are added the impact ofCRA91 could be enough to explain the $200ndash$300 differential

Another way to consider the costs of discrim-ination suits is to look at prices for EmploymentPractices Liability Insurance (EPLI) This prod-uct which has grown signi cantly in recentyears but still has fairly low market penetrationindemni es companies against the legal costsand damages resulting from discrimination andother employment-practices litigation Fromconversations with two insurance agents welearned that the approximate cost of EPLI is $62per employee per year although this gure var-ies considerably with rm size rm type andthe buyerrsquos internal human resource policiesThe contribution of CRA91-protected workersto this cost is presumably signi cantly higherAlso insurers are unwilling to provide EPLI atthe quoted rates unless the employer develops aset of internal procedures to minimize the prob-ability of litigation Thus the expected costof employment-practices litigation is probablyat least a few hundred dollars per year perprotected worker Overall these back-of-the-envelope calculations lead us to think that theexperience effects measured in Table 3 are com-parable to what we might have expected

BlacksmdashWe now analyze the effects ofCRA91 on returns to experience for black menPanel B of Table 3 and Figure 5(a) display theresults from estimation of equations (8) and (9)using CPS potential and cell-average experi-ence measures We nd no evidence to indicatethat CRA91 affected returns to experience forblack men relative to white men All of ourestimates of bpost in the table are negativemdashindicating a drop in returns for experience forblacksmdashbut none is statistically distinguishablefrom zero Addition of a black trend in returnsto experience does not affect the results25

25 When looking at black employment over the 1988ndash1996 period it is important to consider the effects of the1990 and 1991 federal minimum wage increases We


Despite the smaller sample the NLSY doesoffer some support for the assertion that returnsto experience increased for blacks When weestimate equation (9) with our NLSY samplewe obtain a coef cient of 0026 for bpost Thispoint estimate of the post-CRA91 effect islarger than any of the estimates in Table 3 al-though it is quite imprecise Plotting the bt

parameters from estimation of equation (8) us-ing NLSY data [see Figure 5(b)] we see that therelative change in returns to experience forblack men was actually largest just prior to thepassage of CRA91

We conclude that there is little evidence to

suggest that returns to experience for black menincreased as a result of the passage of CRA91We interpret this as consistent with our modelgiven the age patterns of black male EEOCclaims However we cannot entirely rule outtwo alternative interpretations First it is possi-ble that CRA91 simply did not have a signi -cant effect on the legal environment faced byblack plaintiffs As we described above differ-ent federal courts applied different interpreta-tions of the CRA of 1866 in the years leading upto Patterson If before Patterson employersbehaved as though the CRA of 1866 could beapplied to termination-based cases and wereslow to change their actions after the decisionthen we would expect the 1991 Act to have hadlittle effect on blacks This is inconsistent how-ever with the fact that other studies examiningCRA91 have documented effects on employ-ment outcomes of black men (see Oyer andSchaefer 2000)

Second recall that CRA91 explicitly allowsblack men to use the CRA of 1866 in termination-based suits and that such claims do not need togo through the EEOC Because data on suits led directly in federal court are unavailable itis possible that our Figure 1(b) (which makesuse of EEOC data alone) is not an accuraterepresentation of the age distribution of claimsin the post-CRA91 period This is problematicfor our analysis if the age distribution of EEOCplaintiffs differs signi cantly from the age dis-tribution of federal court plaintiffs in the post-CRA91 period If this were the case howeverone might expect the age pattern of EEOC com-plaints in the years after the Patterson decisionbut before CRA91 (when terminated blackworkers had no recourse without the EEOC) tobe markedly different from that after CRA91We nd the pre- and post-CRA91 age distribu-tions of black EEOC plaintiffs to be quite similar

An Extension to Education as a SignalmdashFi-nally we brie y consider another source ofinformation for employersmdasheducational achieve-ment of potential employees Suppose educationalattainment signals productivity in the manner sug-gested by A Michael Spence (1973) and that thissignal becomes less important as the worker agesbecause potential employers gain alternativesources of information (such as work history) re-garding the employeersquos productivity Then wewould expect that an increase in potential costs of

experimented with Tobit speci cations and with left-censoring wages at a constant minimum wage throughoutthe period we study This had no substantive effect on ourresults Also note that while we ignored the minimum wagein the previous section a much smaller fraction of whitewomen earn minimum wage compared to black men



litigation arising from termination would increasethe returns to education for younger protectedworkers relative to older protected workers26

We offer a simple test of this assertion byexamining how the returns to education changedaround the time of CRA91 We estimate

w it 5 a 1 bpost~dpost 3 dp 3 sit 1 xi 1 laquo it

where s is years of schooling and other vari-ables are de ned as above separately for CPSrespondents between the ages of 25 and 32 andfor those between 33 and 39 The coef cientbpost estimates how the relative-to-white-menreturns to education for protected workerschanged after the passage of CRA91 Thisspeci cation allows us to control for over-all age-group-speci c and protected-group-speci c trends in returns to education Thesecontrols are particularly important here as thereis ample evidence to suggest there have beensigni cant changes in the returns to educationoverall and among certain demographic groupsduring the 1980rsquos and 1990rsquos27

If the reasoning outlined above is correctthen we would expect bpost to be higher foryounger workers than for older workers Wepresent results in Table 4 For both blacks andwomen the point estimates are consistent withthe assertion that CRA91 increased returns toeducation for young protected workers The es-timates in columns (1) and (2) suggest that for

young black men the returns to a year of edu-cation increased by 15 percent relative to olderblacks Similarly columns (3) and (4) indicatethat returns to education increased by 07 per-cent for young white women relative to olderwhite women Not surprisingly given that weare using four sources of variation simulta-neously these difference are not statisticallysigni cant at conventional levels The p-valuestesting the hypotheses that the youngold dif-ference in bpost is zero are 016 and 019 forwomen and blacks respectively While our nding here is not strong it does providesome evidence consistent with employersrsquo useof education as a signal of terminationprobability

V Conclusion

This paper studies how the Civil Rights Actof 1991 affected employment outcomes formembers of protected groups While mostprior work on the effects of employment-discrimination litigation examines average ef-fects on protected workers we focus on how thelaw affected the distribution of wages and em-ployment across members of protected groupsWe develop a simple model to study the effectson labor markets when the costs of potentiallitigation on the part of displaced employeesincreases The novel features of our model arethat workersrsquo characteristics are assumed to be-come more easily observable as workers be-come more experienced and that rms engage inboth for-cause and discriminatory rings We nd the effect of increases in litigation costs onreturns to experience depends crucially on (i)

26 Our argument here is similar to Farber and Gibbons(1996) except that we consider possible costs of bad esti-mates of employeesrsquo productivity

27 See for example Katz and Murphy (1992)

TABLE 4mdashCHANGES TO PROTECTED WORKERSrsquo RETURNS TO EDUCATION POST-CRA91

Protected group Blacks Blacks Women WomenAge range 25ndash32 33ndash39 25ndash32 33ndash39

Experience measure (1) (2) (3) (4)

Protected 3 education 3 post 00069 200077 200003 200070(00082) (00077) (00034) (00034)

Observations 51861 48717 81812 74740R2 01735 01961 02075 02658

Notes Dependent variable is the log of average hourly wage for the year Data from the1988ndash1996 March CPS limited to black men non-Hispanic white men and women aged25ndash39 Sample limited to individuals working at least 1500 hours in the preceding yearThese OLS regressions include controls for potential experience experience squared educa-tion and indicators for state year and half-percentage-point state unemployment rate cate-gories and yearprotected status interactions Standard errors are in parentheses


how employeesrsquo propensity to le employment-discrimination litigation conditional on employ-ment varies with experience and (ii) how theincrease in propensity to sue (stemming fromthe increase in litigation costs) conditional onemployment varies with experience

We test this assertion using a data set ofcomplaints led with the EEOC and using theAnnual Demographics File from the CPS We ndthat women become less likely to le wrongfultermination claims as they age Consistent withour model we also nd an increase in womenrsquosreturns to experience when CRA91 increased ex-pected costs of employment-discriminationlitiga-tion Black men on the other hand become morelikely to le a wrongful termination complaint asthey age so our model does not provide a de n-itive prediction about their returns to experienceWe detect no change in black returns to experi-ence around the time of the passage of CRA91Our analysis suggests that a law that has fairlynegligibleaverage effects on protected groups canhave important redistributive effects within thesegroups

APPENDIX CONSTRUCTION OF ldquoCELL AVERAGErdquoPROXIES FOR ACTUAL EXPERIENCE

This Appendix provides additional descrip-tion of the construction of our ldquoCell Average Irdquoand ldquoCell Average IIrdquo proxies for actual expe-rience used in Table 3 We rst partitioned ourwage regression sample (from Table 2) intocells by birth year gender race and educationWe use two categories for race (black and non-Hispanic white) and four for education (did not nish high school high-school graduate somecollege and college graduate) We then gatherinformation from the 1964ndash1995 March CPSon how individuals in each cell accumulatedwork experience over time For each CPS weexamine all respondents who are in one of ourcells and for whom age is greater than theminimum of 18 and that respondentrsquos educationplus six For each such respondent we computework experience in that year as the minimum ofone and hours worked divided by 2080

To calculate our Cell Average I measure we rst compute the average of this work experi-ence measure across individuals within a givencell in each CPS year We then sum these av-erages across CPS years to compute the cumu-lative average experience for members of each

cell As an example consider white femalehigh-school graduates born in 1960 Beginningin 1979 (because this is when individuals in thiscell turned 19) we compute average work ex-perience across all such individuals who re-sponded to the CPS For all members of this cellwho appear in our data for the year 1990 weuse the sum of average work experience ofindividuals in the cell from 1979 through 1989 asour Cell Average I measure of work experience

To calculate our Cell Average II measure webegin with the wage regression sample fromTable 2 For each year (1987ndash1995) and foreach cell we compute the share of all CPSrespondents who worked at least 1500 hoursand thus quali ed for our wage regression sam-ple To compute the actual experience proxy forthat year and cell we then go to the historicalCPS data The procedure is best illustrated byan example Suppose that of the white femalehigh-school graduates born in 1960 who re-sponded to the March 1991 CPS 60 percentworked 1500 or more hours in 1990 We againbegin in 1979 and compute the average experi-ence in that year of the 60 percent of whitefemale high-school graduates born in 1960 whoworked the most hours We repeat this calcula-tion for the years 1980 through 1989 We sumthese average experience gures across years1979ndash1989 to obtain our Cell Average II mea-sure of work experience for white female high-school graduates whom we observe to beemployed in 1990

REFERENCES

Abram Thomas G ldquoThe Law Its InterpretationLevels of Enforcement Activity and Effecton Employer Behaviorrdquo American EconomicReview May 1993 (Papers and Proceed-ings) 83(2) pp 62ndash66

Acemoglu Daron and Angrist Joshua D ldquoCon-sequences of Employment Protection TheCase of the Americans with Disabilities ActrdquoJournal of Political Economy October 2001109(5) pp 915ndash57

Antecol Heather and Kuhn Peter ldquoGender as anImpediment to Labor Market Success WhyDo Young Women Report Greater HarmrdquoJournal of Labor Economics October 200018(4) pp 702ndash28

Barbezat Debra A and Hughes James W ldquoSexDiscrimination in Labor Markets The Role


of Statistical Evidence Commentrdquo AmericanEconomic Review March 1990 80(1) pp277ndash86

Blau Francine D and Kahn Lawrence M ldquoSwim-ming Upstream Trends in the Gender WageDifferential in 1980rsquosrdquo Journal of Labor Eco-nomics January 1997 Pt 1 15(1) pp 1ndash42

Bound John and Johnson George ldquoChanges inthe Structure of Wages in the 1980rsquos AnEvaluation of Alternative ExplanationsrdquoAmerican Economic Review June 199282(3) pp 371ndash92

Bound John and Freeman Richard B ldquoWhatWent Wrong The Erosion of Relative Earn-ings and Employment among Young BlackMen in the 1980rsquosrdquo Quarterly Journal of Eco-nomics February 1992 107(1) pp 201ndash32

DeLeire Thomas ldquoThe Wage and EmploymentEffects of the Americans with DisabilitiesActrdquo Journal of Human Resources Fall2000 35(4) pp 693ndash715

Dertouzos James N and Karoly Lynn A ldquoLaborMarket Responses to Employer LiabilityrdquoRAND Report R-3989-ICJ 1992

Donohue John J and Siegelman Peter ldquoTheChanging Nature of Employment Discrimi-nation Litigationrdquo Stanford Law ReviewMay 1991 43(5) pp 983ndash1033

Farber Henry S and Gibbons Robert ldquoLearningandWage Dynamicsrdquo Quarterly Journalof Econom-ics November 1996 111(4) pp 1007ndash47

Gladden Tricia and Taber Christopher ldquoWageProgression Among Low Skilled Workersrdquoin David E Card and Rebecca M Blankeds Finding jobs Work and welfare reformNew York Russell Sage Foundation 2000pp 160ndash92

Goldin Claudia Understanding the gender gapOxford Oxford University Press 1990

Heckman James J and Willis Robert J ldquoABeta-logistic Model for the Analysis of Se-quential Labor Force Participation by Mar-ried Womenrdquo Journal of Political EconomyFebruary 1977 85(1) pp 27ndash58

Hoyman Michele and Stallworth Lamont ldquoSuitFiling by Women An Empirical AnalysisrdquoNotre Dame Law Review October 198662(1) pp 61ndash82

Katz Lawrence F and Murphy Kevin MldquoChanges in Relative Wages 1963ndash1987Supply and Demand Factorsrdquo QuarterlyJournal of Economics February 1992107(1) pp 35ndash78

Morgan Phoebe A ldquoRisking Relationships Un-derstanding the Litigation Choices of Sexu-ally Harassed Womenrdquo Law and SocietyReview April 1999 33(1) pp 67ndash91

Moulton Brent R ldquoRandom Group Effects andthe Precision of Regression Estimatesrdquo Jour-nal of Econometrics August 1986 32(3) pp385ndash97

Nager Glen D and Broas Julia M ldquoEnforce-ment Issues A Practical Overviewrdquo Loui-siana Law Review July 1994 54(6) pp1473ndash86

Neumark David and Stock Wendy A ldquoAge Dis-crimination Laws and Labor Market Ef -ciencyrdquo Journal of PoliticalEconomy October1999 107(5) pp 1081ndash125

OrsquoNeill June and Polachek Solomon ldquoWhy theGender Gap in Wages Narrowed in the1980srdquo Journal of Labor Economics Janu-ary 1993 Pt 1 11(1) pp 205ndash28

Oyer Paul and Schaefer Scott ldquoLayoffs andLitigationrdquo RAND Journal of EconomicsSummer 2000 31(2) pp 345ndash58

Posner Richard A ldquoThe Ef ciency and the Ef- cacy of Title VIIrdquo University of Pennsylva-nia Law Review December 1987 136(2) pp513ndash21

Robinson Robert K Allen Billie Morgan Terp-stra David E and Nasif Ercan G ldquoEqual Em-ployment Requirements for Employers ACloser Review of the Effects of the CivilRights Act of 1991rdquo Labor Law JournalNovember 1992 43(11) pp 725ndash34

Selmi Michael ldquoFamily Leave and the GenderWage Gaprdquo North Carolina Law ReviewMarch 2000 78(3) pp 707ndash82

Shaw Kathryn ldquoThe Persistence of Female La-bor Supply Empirical Evidence and Implica-tionsrdquo Journal of Human Resources Spring1994 29(2) pp 348ndash78

Spence A Michael ldquoJob Market SignalingrdquoQuarterly Journal of Economics August1973 87(3) pp 355ndash74


of higher damages may make experiencedworkers more likely to sue conditional on be-ing red which may imply that increases inlitigation costs will reduce rmsrsquo demand forthese workers Developing these effects furtherwe demonstrate that the effect of increases inlitigation costs on returns to experience de-pends crucially on (i) how employeesrsquo propen-sity to le employment-discrimination litigationconditional on employment varies with experi-ence and (ii) how the increase in propensity tosue (stemming from the increase in litigationcosts) conditional on employment varies withexperience

The passage of the Civil Rights Act of 1991(CRA91) provides an opportunity to study thisrelationship empirically This Act contains anumber of provisions that increased the ex-pected costs to rms of displacing protectedemployees The group protected by CRA91 isbroad and includes racial minorities femalesand those with disabilities While previous fed-eral employment discrimination legislation typ-ically limited plaintiff recovery to lost wagesCRA91 allows employees to sue for up to$300000 in punitive damages By extendingthe Civil Rights Act of 1866 CRA91 allowsemployees claiming unlawful termination onthe basis of race to sue for unlimited punitivedamages CRA91 also gives either side in asuit the right to a jury trial this presumablyfavors plaintiffs as juries are thought to bepartial to claims of individuals over those of rms

We proceed by identifying relationships be-tween the propensity to sue as a function of ageand changes in the returns to experience amongprotected workers around the time of the pas-sage of CRA91 Using data on complaints ledwith the Equal Employment Opportunity Com-mission (EEOC) we nd that wrongful termi-nation complaints drop sharply with age forwomen but rise steadily with age for blacksWe tie this nding into an analysis of the returnsto experience for these protected groups Usingdata from the 1988ndash1996 annual demographics le of the Current Population Survey (CPS) we nd that CRA91 had relatively minor aggregateemployment and wage effects However thelaw does appear to have affected returns toexperience in the way suggested by our modelWe show that returns to experience increasedfor women but not for blacks shortly after the

passage of the Act This nding is consistentwith the experiencepropensity-to-sue relation-ships found in the EEOC data Together these ndings offer a pattern that ts with our modelof litigation costs and returns to experience Wetake these results as evidence that antidiscrimi-nation protections (and employment protectionsin general) have redistributive effects Theseeffects appear to operate not merely acrossgroups of differing protected status but alsowithin groups of identical protected status

Although our empirical analysis focusessolely on CRA91 our model applies equallywell to any erosion of employment-at-willWhile CRA91 did aid the growth in employ-ment discrimination litigation the number ofsuch suits had been increasing steadily for atleast two decades before the passage of CRA91Our analysis therefore suggests that this increas-ing tide in litigation (or more generally con-tinued erosions in employment-at-will) mayhave been a contributing factor in the observedincrease in returns to experience over thatperiod1

I The Civil Rights Act of 1991and the Legal Environment

The Civil Rights Act of 1991 which tookeffect on November 21 1991 strengthened sev-eral prior pieces of employment-discriminationlegislation including the Civil Rights Act of1866 the Civil Rights Act of 1964 (Title VII)the Age Discrimination in Employment Act(ADEA) and the Americans with Disabilities

1 John J Donohue and Peter Siegelman (1991) docu-ment the growth in employment-discrimination litigationthroughout the 1970rsquos and 1980rsquos Lawrence F Katz andKevin M Murphy (1992) and John Bound and GeorgeJohnson (1992) show that the returns to experience in-creased for all workers over this period Bound and RichardB Freeman (1992) show this effect was stronger for blacksthan for whites while June OrsquoNeill and Solomon Polachek(1993) and Francine D Blau and Lawrence M Kahn (1997)show the effect was stronger for women than for menDavid Neumark and Wendy A Stock (1999) consider theeffects of employment protections on returns to experienceover the 1980rsquos but from a very different perspective Theystudy age-discrimination laws and suggest that by provid-ing an enforcement mechanism for implicit contractsthese laws increase the steepness of wage pro les for allworkers Unlike ours their analysis does not explicitlylink wage pro les to litigation costs imposed by protectedworkers















A Workers and Firms







5


















(1) f1 R9

5 1 2 d1 2 a~1 2 f1 ~1 2 d1 w1t

1 d1G1 qd ~w1t 1 Dqd ~w1t 1 D 1 k

1 a~1 2 f1 ~1 2 d1 G1 qc ~w1t 1 D

3 qc ~w1t 1 D 1 k

(2) gf2R9

5 1 2 d2 2 a~1 2 f2 ~1 2 d2 w2t

1 d2G2 qd ~w2t 1 Dqd ~w2t 1 D 1 k

1 a~1 2 f2 ~1 2 d2 G2 qc ~w2t 1 D

3 qc ~w2t 1 D 1 k











1 a~1 2 fi ~1 2 di G i qc ~wit 1 D

3 ~qc ~wit 1 D 1 k









3 qd ~w it 1 D 1 k)



3 qc ~w it 1 D 1 k)








1 a~1 2 fi ~1 2 di G i qc ~w it 1 D




D Extensions














































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES












































A Workers and Firms







5


















(1) f1 R9

5 1 2 d1 2 a~1 2 f1 ~1 2 d1 w1t

1 d1G1 qd ~w1t 1 Dqd ~w1t 1 D 1 k

1 a~1 2 f1 ~1 2 d1 G1 qc ~w1t 1 D

3 qc ~w1t 1 D 1 k

(2) gf2R9

5 1 2 d2 2 a~1 2 f2 ~1 2 d2 w2t

1 d2G2 qd ~w2t 1 Dqd ~w2t 1 D 1 k

1 a~1 2 f2 ~1 2 d2 G2 qc ~w2t 1 D

3 qc ~w2t 1 D 1 k











1 a~1 2 fi ~1 2 di G i qc ~wit 1 D

3 ~qc ~wit 1 D 1 k









3 qd ~w it 1 D 1 k)



3 qc ~w it 1 D 1 k)








1 a~1 2 fi ~1 2 di G i qc ~w it 1 D




D Extensions














































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES




































A Workers and Firms







5


















(1) f1 R9

5 1 2 d1 2 a~1 2 f1 ~1 2 d1 w1t

1 d1G1 qd ~w1t 1 Dqd ~w1t 1 D 1 k

1 a~1 2 f1 ~1 2 d1 G1 qc ~w1t 1 D

3 qc ~w1t 1 D 1 k

(2) gf2R9

5 1 2 d2 2 a~1 2 f2 ~1 2 d2 w2t

1 d2G2 qd ~w2t 1 Dqd ~w2t 1 D 1 k

1 a~1 2 f2 ~1 2 d2 G2 qc ~w2t 1 D

3 qc ~w2t 1 D 1 k











1 a~1 2 fi ~1 2 di G i qc ~wit 1 D

3 ~qc ~wit 1 D 1 k









3 qd ~w it 1 D 1 k)



3 qc ~w it 1 D 1 k)








1 a~1 2 fi ~1 2 di G i qc ~w it 1 D




D Extensions














































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES
































5


















(1) f1 R9

5 1 2 d1 2 a~1 2 f1 ~1 2 d1 w1t

1 d1G1 qd ~w1t 1 Dqd ~w1t 1 D 1 k

1 a~1 2 f1 ~1 2 d1 G1 qc ~w1t 1 D

3 qc ~w1t 1 D 1 k

(2) gf2R9

5 1 2 d2 2 a~1 2 f2 ~1 2 d2 w2t

1 d2G2 qd ~w2t 1 Dqd ~w2t 1 D 1 k

1 a~1 2 f2 ~1 2 d2 G2 qc ~w2t 1 D

3 qc ~w2t 1 D 1 k











1 a~1 2 fi ~1 2 di G i qc ~wit 1 D

3 ~qc ~wit 1 D 1 k









3 qd ~w it 1 D 1 k)



3 qc ~w it 1 D 1 k)








1 a~1 2 fi ~1 2 di G i qc ~w it 1 D




D Extensions














































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES


































(1) f1 R9

5 1 2 d1 2 a~1 2 f1 ~1 2 d1 w1t

1 d1G1 qd ~w1t 1 Dqd ~w1t 1 D 1 k

1 a~1 2 f1 ~1 2 d1 G1 qc ~w1t 1 D

3 qc ~w1t 1 D 1 k

(2) gf2R9

5 1 2 d2 2 a~1 2 f2 ~1 2 d2 w2t

1 d2G2 qd ~w2t 1 Dqd ~w2t 1 D 1 k

1 a~1 2 f2 ~1 2 d2 G2 qc ~w2t 1 D

3 qc ~w2t 1 D 1 k











1 a~1 2 fi ~1 2 di G i qc ~wit 1 D

3 ~qc ~wit 1 D 1 k









3 qd ~w it 1 D 1 k)



3 qc ~w it 1 D 1 k)








1 a~1 2 fi ~1 2 di G i qc ~w it 1 D




D Extensions














































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES


































1 a~1 2 fi ~1 2 di G i qc ~wit 1 D

3 ~qc ~wit 1 D 1 k









3 qd ~w it 1 D 1 k)



3 qc ~w it 1 D 1 k)








1 a~1 2 fi ~1 2 di G i qc ~w it 1 D




D Extensions














































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES


































1 a~1 2 fi ~1 2 di G i qc ~w it 1 D




D Extensions














































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES








































































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES





























































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES




















































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES









































Entire sample(1)

Pre-CRA91(2)

Post-CRA91(3)

White men(4)

White women(5)

Black men(6)


(1156) (1103) (1168) (1037) (1158) (947)Hourly wage 1165 1103 1254 1315 1018 1016

(72) (68) (77) (748) (663) (947)Age 3216 3203 3235 3219 3216 3192

(423) (424) (423) (423) (423) (432)Education 135 134 136 136 135 128


(16) (17) (15) (16) (16) (15)



(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES































(6) y it 5 a 1 apdp

1 Ot 5 87

95



















Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES





































Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Female 202253 202132 224059 251967 202704 211530(00216) (04096) (1243) (24621) (00087) (01875)

[200845] [200800]Female 3 post-CRA91 00538 00544 520 2859 00416 200021

(00115) (00238) (711) (1468) (00054) (00107)[00201] [00203]




Dependent variable

Full-timeemployment

(1)

Full-timeemployment

(2)

Hoursworked

(3)

Hoursworked

(4)

Logwages

(5)

Logwages

(6)

Black 203517 218710 237114 2175856 202254 200217(00421) (09185) (2198) (51885) (00182) (04186)

[201199] [206075]Black 3 post-CRA91 00052 200645 2239 26514 200026 00073

(00259) (00537) (1496) (2933) (00120) (00237)[00016] [200206]














~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES








































~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES


































~8 w it 5 a 1 ap dp

1 Ot 5 87

95




1 xi 1 laquoit















(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES






































(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 156552 156552 156292 156292 156161 156161R2 02540 02540 02527 02528 02498 02499



(1)Potential

(2)Cell average I

(3)Cell average I

(4)Cell average II

(5)Cell average II

(6)



Observations 100578 100578 100183 100183 99918 99918R2 02155 02155 02143 02143 02136 02136


































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES






























































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES






















































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES














































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES





































V Conclusion








Observations 51861 48717 81812 74740R2 01735 01961 02075 02658










REFERENCES






































REFERENCES
























































Litigation Costs and Returns to Experience...Litigation Costs and Returns to Experience ByPAULOYERANDSCOTTSCHAEFER* We develop a model linking maximum damage awards available to plaintiffs

Documents