V_Black and Brainerd - Importing Equality - The Impact of Globalization on Gender Discrimination

7/29/2019 V_Black and Brainerd - Importing Equality - The Impact of Globalization on Gender Discrimination

1/22

Industrial & Labor Relations Review

Volume 57, Issue 4 2004 Article 3

Importing Equality? The Impact of

Globalization on Gender Discrimination

Sandra E. Black

Elizabeth Brainerd

UCLA,Williams College,

Copyright c2004 by the authors. All rights reserved. No part of this publication may bereproduced, stored in a retrieval system, or transmitted, in any form or by any means, elec-

tronic, mechanical, photocopying, recording, or otherwise, without the prior written permis-

sion of the publisher, bepress, which has been given certain exclusive rights by the author. In-

dustrial & Labor Relations Review is produced by The Berkeley Electronic Press (bepress).

http://digitalcommons.ilr.cornell.edu/ilrreview


2/22

Importing Equality? The Impact of

Globalization on Gender Discrimination

Sandra E. Black and Elizabeth Brainerd

Abstract

A key dynamic implication of the Becker model of discrimination (1957) is that increased

product market competition will drive out costly discrimination in the long run. This paper tests

that hypothesis by examining the impact of globalization on gender discrimination in manufac-turing industries. Because concentrated industries face little competitive pressure, an increase in

competition from trade should reduce the residual gender wage gap more in these industries than in

competitive industries. The authors compare the change in the gender wage gap between 1976 and

1993 in concentrated versus competitive manufacturing industries, using the latter as a control for

changes in the gender wage gap that are unrelated to competitive pressures. They find that while

trade increases wage inequality by modestly reducing the relative wages of less-skilled workers,

at the same time it appears to benefit women by reducing the ability of firms to discriminate.

The authors thank Francine Blau, Ralph Bradburd, Janet Currie, Rebecca Demsetz, Judith Heller-

stein, Chinhui Juhn, William Pizer, Marc Saidenberg, and Joseph Tracy, as well as seminar partici-

pants at the Federal Reserve Bank of New York, Hunter College, and Williams College, for helpful

comments and discussions. Special thanks to David Jaeger for providing the code to match census

MSAs over time. Colleen Sellers and Jennifer Poole provided outstanding research assistance.


3/22

540

Industrial and Labor Relations Review, Vol. 57, No. 4 (July 2004). by Cornel l University .0019-7939/00/5704 $01.00

I

IMPORTING EQUALITY? THE IMPACT OF

GLOBALIZATION ON GENDER DISCRIMINATION

SANDRA E. BLACK and ELIZABETH BRAINERD*

A key dynamic implication of the Becker model of discrimin ation (1957) isthat increased product market competition will drive out costly discriminationin the long run. This paper tests that hypothesis by examining the impact ofglobalization on gender discrimination in manufacturing industries. Becauseconcentrated industries face little competitive pressure, an increase in compe-tition from trade should reduce the residual gender wage gap more in theseindustries than in competitive industries. The authors compare the change inthe gender wage gap between 1976 and 1993 in concentrated versus competitivemanufacturing industries, using the latter as a control for changes in the genderwage gap that are unrelated to competit ive pressures. They f ind that whi le tradeincreases wage inequality by modestly reducing the relat ive wages of less-skilledworkers, at the same t ime it appears to benefit women by reducing the abil ity offirms to discriminate.

*Sandra E. Black is Assistant Professor of Econom-ics, UCLA, and Faculty Research Fellow of NBER, andElizabeth Brainerd is Associate Professor of Econom-ics, Williams College, and Research Affiliate of CEPR,IZA, and the William Davidson Institute. The authorsthank Francine Blau, Ralph Bradburd, Janet Currie,Rebecca Demsetz, Judith Hellerstein, Chinhui Juhn,

Will iam Pizer, Marc Saidenberg, and Joseph Tracy, aswell as seminar participants at the Federal ReserveBank of New York, Hunter College, and WilliamsCollege, for helpful comments and discussions. Spe-cial thanks to David Jaeger for providing the code tomatch census MSAs over time. Colleen Sellers andJennifer Poole prov ided outstanding research assis-tance.

Copies of the computer programs used to gener-ate the results presented in the paper are availablefrom Sandra Black at [email protected].

n his seminal work on the economics ofdiscrimination, Gary Beckers theory

(1957) has the startling implication thatincreased competition in the product mar-ket will reduce discrimination againstwomen and minorities in the long run.This implies a positive relationship between

market power and employment discrimina-tion: because a firm must forego profits inorder to indulge in a taste for discrimina-tion, employers with considerable marketpower will be better able to practice dis-crimination than those with little marketpower. The theory also has the dynamicimplication that changes in market powerwill produce changes in the relative em-ployment and earnings of groups initiallysubject to discrimination. Of specific inter-est to us, in that regard, is the predictionthat increased product market competi-tion in an industry (or region) over timewill reduce earnings and employment dis-parities between men and women, all elseequal.

Black and Brainerd: Importing Equality? The Impact of Globalization on Gender Discri

Produced by The Berkeley Electronic Press, 2003


4/22

GLOBALIZATION AND GENDER DISCRIMINATION 541

The recent narrowing of the gender earn-ings gap in an era of increased competitionthrough international trade and deregula-tion might seem to offer support for thistheory. Since 1960, the time trend for thefemale:male wage ratio has closely trackedthat for imports as a share of GDP, withboth series remaining fairly constant be-tween 1960 and 1980, then increasing dra-matically through the early 1990s (Figure1). The gains for women occurred, more-over, at a time when federal anti-discrimi-

nation efforts were waning. Despite thissuggestive evidence, however, the possibil-ity that women have benefited from in-creased product market competition re-sulting from increased trade has receivedlittle formal research attention. Blau andKahn (1997), for example, concluded thatfemale wage gains in the 1980s were largelyattributable to womens gains in work expe-rience and occupational status, with im-provements in unobserved characteristics

also playing a role.1 The study also citedreduced labor market discriminationagainst women as a likely factor, but it didnot investigate whether reduced discrimi-nation was in turn related to trade-relatedchanges in competition. That possible linkis the subject of the present paper.

Did employers indeed face increasedcompetition in the 1980s? At least in somesectors, it appears that they did: a numberof industries (such as banking, trucking,telecommunications, and airlines) faced

deregulation in the mid- to late 1970s andearly 1980s, and many industries confrontedintensified competition in the form of in-

Sources. Gender wage ratio: U.S. Bureau of the Census, Current Population Reports. Imports/GDP: U.S.

Dept. of Commerce, National Income and Product Accounts.

0.75

0.70

0.65

0.60

0.55

0.50

1960 1965 1970 1975 1980 1985 1990 1995

Figure 1. Trends in Female/Male Median Wages (Full-Time Workers)

and Imports as a Share of GDP (1992 dollars).

Imports/GDP

F/M Median Wage

0.16

0.14

0.12

0.10

0.08

0.06

0.04

0.02

0.00

F/MMedianWage

Imports/GDP

1Other important contributions to understandingchanges in the gender wage gap include Goldin (1990)and ONeill and Polachek (1993); Blau (1998) pro-vides a broad overview of changes in the economicstatus of women from 1970 to 1995.

Industrial & Labor Relations Review, Vol. 57 [2003], Iss. 4, Art. 3

http://digitalcommons.ilr.cornell.edu/ilrreview/vol57/iss4/3


5/22

542 INDUSTRIAL AND LABOR RELATIONS REVIEW

creased imports from foreign competitors.This paper focuses on the latter form ofincreased competition, and attempts toanswer the question: has increased tradeled to a decline in discrimination and, as aresult, contributed to the improvement inrelative female wages? Putting it anotherway, did the market step in where the fed-eral government left off, and force at leastsome employers to reduce discriminationin order to remain viable in an increasinglycompetitive world?

Using both the Current Population Sur-

vey and the 1980 and 1990 Censuses, we testthis idea by examining the relationshipbetween changes in trade and changes inthe gender wage gap across industries aswell as across metropolitan areas. The wagedata are broken down by concentrated andcompetitive industries. Since concentratedindustries face little competitive pressureto reduce discrimination, an increase incompetition from increased trade shouldlead to a greater reduction in the genderwage gap in these industries than in com-petitive industries also hit by trade. Wecompare the change in the gender wage

gap in trade-affected concentrated versusunconcentrated sectors, using the latter asa control for changes in the gender wagegap that are unrelated to competitive pres-sures.

The positive perspective on trade weimplicitly adopt here contradicts the spiritof recent research on the links betweentrade and the structure of wages, which haslargely focused on the contribution of tradeto rising wage inequality in the UnitedStates, and particularly on the link betweentrade and the deteriorating fortunes of lessskilled workers. While analysts disagree on

the size of the impact of trade on wageinequality and relative employment, thereis little disagreement over the sign: for lessskilled workers, trade hurts.2 Our study

investigates whether, in contrast to thateffect, trade may actually benefit somegroups of workersat least in a relativesenseby reducing the ability of firms todiscriminate.3

Conceptual Framework

The Becker Model ofEmployer Discrimination

Beckers 1957 treatise on discriminationbegan by focusing on employers personal

preferences as a source of discrimination,arguing that some employers had a tastefor discrimination and would be willing topay to indulge this taste.4 As Gary Beckerhimself put it some 45 years ago:

If an individual has a taste for discrimination,he must actas ifhe were willing to pay some-thing, either directly or in the form of a reducedincome, to be associated with some personsinstead of others. When actual discriminationoccurs, he must, in fact, either pay or forfeitincome for this privilege. This simple way oflooking at the matter gets at the essence ofprejudice and discrimination. (p. 14)

Employers with a taste for discriminationagainst women will hire fewer than theprofit-maximizing number of women, em-ploying more men who are equally skilledyet more highly paid. As a result, non-discriminating employers can drive dis-criminating employers out of the marketbecause discrimination is costly: employ-ers who discriminate against women sacri-fice profits in order to indulge their tastefor discrimination. In an increasingly com-petitive market, the wage gap between menand women with equal skills will narrowand mayunder certain conditionseven-

2For an overview of the literature on trade andwage inequal ity, see Freeman (1995) and the refer-ences therein.

3Note that increased competition can make work-ers who are discriminated against absolutely worse offbut relatively better off if it induces the firm toeliminate both the rents it previously shared withworkers and the gender (or racia l) wage gap.

4Becker also analyzed the effects of discriminationby co-workers and by customers; the focus here is onhis model of employer discrimination.




6/22


tually disappear, as discriminators areforced by market pressure to change theirdiscriminatory practices or are bought outby non-discriminating firms.5

Increased import competition is onemechanism through which this narrowingof the gender wage gap could occur. Amodel consistent with this prediction isdescribed in Borjas and Ramey (1995). Inthis model, firms in noncompetitive sectorsbehave as Cournot oligopolists, choosingquantities of output produced taking thequantities produced by other firms as given.

Sectors can be noncompetitive for a num-ber of reasons, including high startup costsand barriers to entry. Rents are shared withworkers through bargaining; in the case ofinterest to us, rents can be shared dispro-portionately with men, with both womenand men willing to work in the sector be-cause wages for both groups are at or abovethe competitive wage.6 An exogenous in-crease in trade reduces rents in the indus-try and hence reduces wages; if male work-ers were enjoying more rents than femaleworkers, the gap between the two will shrinkwith the increased competition. In the

competitive sector, since wages were al-ready at the competitive level, the genderwage gap will be less affected by the in-crease in trade.7

It is clear that any form of product mar-ket competitionwhether through in-

creased imports or increased domestic com-petitionplays an important role in thismodel, suggesting a link between marketstructure and the ability of an employer topractice discrimination: discriminatingemployers with market power, presumablyearning positive economic profits, will beable to survive longer in the market thanthose operating in a competitive marketwith zero economic profits. Therefore, thegender wage gap should be smaller in com-petitive markets than in concentrated mar-kets, all else equal. This prediction appears

to provide a relatively simple test of theneoclassical theory of labor market discrimi-nation.

Some of the literature on labor marketdiscrimination has focused on testing thisimplication of Beckers theory regardingthe relationship between market power anddiscriminatory practices. One of the mostcompelling studies in this vein examinedemployment practices in the banking in-dustry and found a negative and statisti-cally significant relationship between mar-ket power in local banks and the share offemale employment in each bankthus

confirming the predictions of Beckerstheory (Ashenfelter and Hannan 1986).8

More recently, Black and Strahan (2001)studied how the deregulation in the bank-ing industry after the mid-1970s affectedbanks ability to share rents with favoredworkers. They found that banks did sharerents disproportionately with men and thatderegulation reduced this practice and sig-nificantly improved the relative wages ofwomen. Hellerstein, Neumark, and Troske(2002) tested the relationship between prof-its and female employment across firmswith market power and found that firms

employing higher proportions of womenhad higher profits, as the theory predicts.In contrast, they found little support forthe hypothesis that discriminatory firmsgrow more slowly than non-discriminatory

5See Becker (1957), Goldberg (1982), andHeckman (1998) for more detailed discussions re-garding the conditions required for this relationshipto hold. While other empirical tests of Beckershypothesis (discussed below) have focused on thenarrowing of the employment gap implied by thetheory, this paper focuses primarily on the narrowingof the wage gap between groups that are discrimi-nated against and those that are not. Becker arguedthat the wage gap will narrow because firms employ-ing the group that is subject to discrimination (whoearn lower wages) will expand relative to firms em-ploying workers not subject to discrimination, andthis will increase the wages of the former grouprelative to those of the latter (Becker 1971:44).

6See Black and Strahan (2001) for evidence of thisin the banking industry.

7See Borjas and Ramey (1995) for a more detaileddiscussion of the model.

8This study also summarizes the early evidencefrom other studies on the relationship between em-ployment discrimination and product market power.




7/22


firms. Their five-year time frame may havebeen too short to adequately test the latterhypothesis, however.

Unlike most previous researchers, wechoose to focus our analysis on one of thekey dynamic implications of the Beckermodelthat changes in the competitiveenvironment will lead to changes in thegender wage differentialrather than ex-amine the static correlation between prod-uct market competition and the genderwage (or employment) gap at any one pointin time. We take this approach because our

primary concern is to understand the ap-parent change in labor market discrimina-tion against women in the 1980s and 1990s.

Methodology

Testing the simple prediction that in-creased competition from trade leads todeclining discrimination against womenand thus a declining gender wage gap is lessstraightforward than it appears, however.It is evident that the gender wage gap hasnarrowed since the late 1970s for a varietyof reasons, many of which are unrelated to

increased competitiveness in product mar-kets. Consider the well-documented in-creases in womens educational attainmentover that period. If, for some reason,womens educational attainment increasedby more in trade-affected industries than innon-trade-affected industries, simple em-pirical tests may indicate that trade contrib-uted to the narrowing of the differential,rather than point to the underlying truecause. Therefore it will be important tocontrol for differing changes in observablecharacteristics across industries and regionsthat may confound the results. As a first

step toward this goal, we test the links be-tween trade and the residual gender wagegap, that is, the gender wage gap that re-mains after one controls for differences ineducation and potential labor market ex-perience between men and women.

It is equally important to control, if pos-sible, for differing changes in womensunobserved characteristics that may havecontributed to differing improvements inrelative female pay across industries; such

changes are speculated to have contrib-uted to the narrowing of the unexplainedportion of the gender wage gap in the1980s (Blau and Kahn 1997). These unob-served characteristics might include, forexample, a stronger commitment to thelabor force or to ones career, or improvedability or underlying productivity of womenrelative to men.

To purge our estimates of bias due tothese omitted variables, we use a methodol-ogy that (conceptually) sorts our observa-tions by industry according to whether the

industry (1) was or was not affected by atrade shock in the period under study and(2) was concentrated or was competitive.This estimator will eliminate bias due toomitted variables that (1) have a commonvalue for all trade-affected or non-trade-affected industries, such as shocks to eco-nomic conditions in manufacturing indus-tries, and (2) have a common value for allconcentrated or competitive industries,such as worker ability or labor force attach-ment. In other words, the results indicatethe impact of trade on the gender wage gapin concentrated industries relative to com-

petitive industries, netting out any factorsthat have affected the gender wage gap inmanufacturing industries, trade-affectedindustries as a whole, or concentrated in-dustries as a whole.9 Conceptually we calcu-late the following differences in the genderwage gap (note that, in practice, we allowthe impact of trade to be continuous andnot discrete):

(1) [ trade-affected non-trade-affected]

[ trade-affected non-trade-affected ]

concentrated concentrated

competitivecompetitive

9Although trade may have simi lar effec ts in thenon-manufacturing sector, the empirical analysis fo-cuses on the manufacturing sector because trade dataare unavailable for the non-manufacturing sector. Inaddition, several industries in the non-manufactur-ing sector were affected by deregulation during thesame time period (for example, trucking, airlines,banking, and telecommunications), and it would bedifficult to isolate these effects from the effects oftrade.




8/22


This is equivalent to estimating

(2) t(ln[WAGE]

im ln[WAGE]

if) =

+ tTRADE

i+ CONCEN

i

+ (tTRADE

i CONCEN

i),

where tTRADE

iis the change in the import

share in industryiand CONCENiis an indica-

tor variable equal to one if the industry wasconcentrated in 1977.10 The inclusion ofthe dummy variable for concentrated in-dustries allows for a differential change inthe gender wage gap for concentrated in-

dustries relative to competitive industries.The marginal effect of trade on concen-trated industries relative to competitive in-dustries is represented by the coefficient;this is the primary parameter of interest.

This approach implicitly makes two as-sumptions. First, it assumes that discrimi-nation against women did indeed exist, atleast at the beginning of the period understudy, and that this discrimination was re-flected in lower wages for women relative toequally skilled men. While these assump-tions are clearly controversial, several re-cent careful studies provide evidence in

their support. Two of these studies com-pared men and women with very similarhuman capital investments and labor mar-ket skills, and found that a wage gap of 1015% still exists even when one includesdetailed controls for work and skill charac-

teristics (Wood, Corcoran, and Courant1993; Weinberger 1998). Similarly, an au-dit study of hiring in Philadelphia restau-rants found that high-priced restaurantswere substantially more likely to interviewand make job offers to men than womenwith comparable work experience(Neumark 1996). These studies suggestthat gender discrimination did persist, atleast in the 1980s and early 1990s.11

The second assumption implicit in thismethodology is that increased imports areequivalent to an increase in competition

within an industry and that this increase isexogenous to the residual gender wage gap.Several studies document that increasedimports are equivalent to an increase incompetition; for example, Katics andPetersen (1994), using industry-level datafor the United States, found that increasedimport competition reduced price-costmargins during the 197686 period, andHarrison (1994) showed a strong and statis-tically significant negative relationship be-tween import penetration and price-costmargins in Cote dIvoire. This assumptionhas also been used in a number of articles,

including the work by Borjas and Ramey(1995), which inspired the method we usehere. Additionally, recent work by Kletzer(2003) uses import shares as a measure ofcompetition from trade and discusses theassumptions necessary for this procedureto be valid.12

10This approach is similar in spirit to that of Borjasand Ramey (1995), which examined the relationshipbetween wage inequality and foreign competition bycomparing the effect of imports in concentrated ver-sus competitive industries. Like Borjas and Ramey,we use 1977 concentration ratios to determine i f anindustry is concentrated and we hold this definitionconstant over the sample period. Our interest is inhow the pattern of gender discrimination changed inindustries that were initially concentrated in oursample period versus the pattern in industries thatwere init iall y competitive. If we allowed the concen-tration variable to change to reflect the changedstatus of an industryand concentration may indeedhave changed in some industries over the period, asincreased trade generated greater competitionouranalysis would fail to capture the full impact of tradeon gender discrimination across the sample period inindustries that were concentrated at the beginning.

11The persistence of wage differentials across in-dustries raises the question as to why women have notsimply moved from low- to high-paying industries,thus eliminating gender wage differentials across in-dustries over time. Recent evidence in the bankingindustry suggests that industries may be sharing rentswith workers and disproportionately wi th male work-ers. As a result, womenalthough earning less thantheir male counterpartsmay still be earning a wageabove the competitive wage and hence have no incen-tive to leave despite the discriminatory behavior. SeeBlack and Strahan (2001).

12We also considered using industry-level exchangerates or tariff rates as measures of trade; however, wewere unable to obtain exchange rates for many of theindustries and found changes in tariff rates to be avery weak indicator of differences in trade acrossindustries.




9/22


In the trade literature, some evidencesupports the idea that the increase in com-petition from trade is exogenous to thegender wage gap in an industry; for ex-ample, Sachs (1988) suggested that thetrade experience in the 1980s was in largepart due to the effects of monetary andfiscal policy on the exchange rate. Alterna-tively, one can directly test the validity ofthe exogeneity assumption by examiningthe data itself: if the assumption fails tohold, one would expect industries with ahigher gender wage gap at the beginning of

the period to be more vulnerable to trade,all else equal. However, when we test therelationship between the residual genderwage gap at the beginning of our period(1976) and the change in import sharefrom 1976 to 1993, we find a correlation ofonly .07, suggesting little correlation be-tween the two. Although these issues arefar from settled, based on current researchwe are reasonably confident that the twoassumptions regarding discrimination andthe impact of imports on competition holdfor the period under study.

Data

The primary data source for the empiri-cal work is the March Demographic Supple-ment to the Current Population Survey(CPS) from 1977 through 1994. Althoughthis data set is not ideal for the test outlinedabovein particular, it lacks a measure ofactual labor market experienceit is pref-erable to other large data sets due to therelatively long time period over which con-sistent measures of income and other vari-ables are available, and due to the largesample sizes, which enable analysis across

industries and metropolitan areas. The197794 period is chosen because 1977 wasthe first year in which a relatively largenumber of metropolitan areas are identi-fied in the CPS, and trade data are availableonly through 1994.

The sample is defined similarly to that inBorjas and Ramey (1995), which in turnmatched many of the data refinements de-scribed in Katz and Murphys (1992) studyof the wage structure. The sample includes

individuals aged 1864 who worked full-time in the civilian sector in the year priorto the survey; a full-time worker is definedas one who worked at least 30 hours in hisor her usual work week and worked morethan 48 weeks in the previous year. Self-employed individuals and individuals work-ing without pay are excluded from the analy-sis. The wage data refer to real weekly orhourly earnings in the previous year in1982 dollars; wages were deflated by theConsumer Price Index. As in the workscited above, workers earning less than $67

in weekly wages in 1982 dollars are ex-cluded from the analysis, and the wages ofworkers whose earnings are topcoded aremultiplied by 1.45. Industries in whichmale or female employment comprises lessthan 10% of total employment are alsoexcluded from the sample. Because tradedata are available only for manufacturingindustries, the analysis uses only workersemployed in that sector.

Two additional sources of informationon earnings, work, and demographic char-acteristics are used to test the sensitivity ofthe results to the choice of data set: the

1980 and 1990 Censuses, and the OutgoingRotation Groups of the CPS.13 The codesfor the Metropolitan Statistical Area (MSA)in the Censuses were matched over time ina manner consistent with Jaeger et al.(1998).

The trade data are from the NationalBureau of Economic Research (NBER)Trade Database compiled by RobertFeenstra (1996). The impact of trade on anindustry is measured using import shares,which are calculated as the ratio of im-portsmeasured as the cost in freight (CIF)value of importsto domestic shipments;

the latter data are from the NBER Manufac-turing Productivity database and are de-scribed in Bartelsman and Gray (1996). 14

13Census data were obtained from the IPUMSproject at the University of Minnesota. For moreinformation, see Ruggles and Sobek (1997) or theirwebpage at http ://www. ipums.umn.edu.

14Although one could also use (imports + exports)/(domestic shipments) as a measure of the impact of




10/22


The industry-level import shares are aggre-gated at the three-digit level based on the1980 Census definition. Across MSAs, theimpact of trade is measured as the importshare for the MSA, calculated as the aver-age of the import shares of the industries inthe MSA weighted by the number of work-ers in that industry in that MSA.

An industry is classified as concentratedif the four-firm concentration ratio was .40or greater in 1977, based on the Census ofManufactures conducted in that year.15 Thisdetermination was made at the beginning

of the sample period in order to excludethe possibility that changes in concentra-tion were due to increased trade. Appen-dix Table A1 lists the concentrated andnon-concentrated industries in the samplebased on this definition.16

Finally, the dependent variable used formost of the analysis is the change in theresidual gender wage gap over the period.To calculate this variable, we first regressthe log wage on four categorical educationvariables, age, age squared, and a non-whitedummy variable; this regression is estimatedfor the pooled sample of men and women

in each year of interest.17 The residualgender wage gap is then generated as thedifference in the average residual wage formen and women, calculated at the industryor MSA level. Although one could alsoinclude controls for occupation in the logwage equation, they are excluded here be-cause one form of discrimination againstwomen may have occurred through thetypes of jobs available to them. By exclud-ing any controls for occupation, the resultswill measure the effect of increased compe-tition through trade on employers behav-

ior regarding wages directly as well as indi-rectly through occupational changes.

Given that this study uses the change inthe gender wage gap as the dependentvariable, which is in itself a difference in log(residual) wages, it is clear that measure-ment error in this variable may affect theprecision of the estimates. As discussed inAngrist and Krueger (1998) and Bound etal. (1994), the reliability of earnings datadeclines when earnings are expressed asyear-to-year changes rather than as levels.Although this measurement error does notbias the coefficient estimates, it does in-

crease the standard errors of the coeffi-cient estimates and thus reduces the statis-tical significance of the results. On theother hand, the above studies also indicatethat the reliability of earnings estimatesincreases when one analyzes changes inearnings over longer periods; because thisstudy examines changes over a 17-year pe-riod, it is less likely that measurement errorwill affect the results in a significant way.

Results

Table 1 reports the results of estimatingequation (1) using data from the MarchCPS across manufacturing industries overthe 197693 period.18 In this equation, the

trade, many recent studies examining the relation-ship between trade and labor market outcomes haveused the import share measure, so the same practiceis followed here. Examples of studies that follow thisapproach are Borjas and Ramey (1995), Horn andEastman (1997), and Kletzer (1996a, 1996b). Notethat import shares are a conservative measure of theimpact of trade on an industry: the threat of importsalone may force employers to act more competitivelyand reduce discrimination. As a result, the importshare measure likely underestimates the impact oftrade on employer behavior. We tested the sensitivityof our results by using (imports + exports)/(domesticshipments) as a measure of the impact of trade; theresults were consistent with those presented here.

15We tested the sensiti vity of the results to thischoice of a cutoff and found the results to be insensi-tive to concentration ratios ranging from .30 to .50.Following Borjas and Ramey (1995), we considered aCIC manufacturing industry concentrated if the ma-jori ty of workers in the industry were in concentratedfour-digit (SIC) industries.

16Note that a few industries were dropped due tomissing trade data. The industries included in oursample are listed in Appendix Table A1.

17The four education categories are less than highschool, high school, some college, and college ormore. These education classifications are reasonablyconsistent with those suggested in Jaeger (1997).

18The observations are weighted by the inverse ofthe sampling variance of the dependent variable.Whi le nonzero covariances across observat ions could




11/22


change in the industry-level residual gen-der wage gap is regressed on the change inthe import share in the industry over theperiod, a dummy variable that equals one ifthe industry was concentrated in the begin-ning of the period, and the interaction ofthese two terms. The dependent variable isthe change in the residual gender wage gapfrom 1976 to 1993, so that declines in thisvariable indicate improving female relativewages over the period.

The positive and statistically significantcoefficient on the concentrated industry

dummy variable in the first column of Table1 indicates that the residual gender wagegap increased in concentrated industriesrelative to competitive industries, or in otherwords that the gender wage gap declinedmore in competitive industries than in con-centrated industries in the absence of im-port penetration. The positive coefficienton the change in import share variableindicates that, in unconcentrated indus-tries, the gender wage gap grew more inindustries that experienced greater in-creases in imports than in those that expe-rienced little or no competition from in-

creased trade. While this result may appearto contradict the theory discussed abovethat is, that if trade is a form of competi-tion, increased trade in an industry shouldreduce the gender wage gap in that indus-try relative to industries with no increase intradea second effect of trade on relativewages would work in the opposite direc-tion. This effect is the impact of trade onthe wages of less-skilled workers relative tomore-skilled workers: if trade dispropor-tionately hurts less-skilled workers, as re-cent research has suggested,19 and womencomprise a disproportionate share of less-

skilled workers, then trade will also affectthe relative wages of men and womenthrough this route. If this is the case, onewould expect trade to reduce womenswages relative to mens wages, and an in-crease in the gender wage gap should beobserved in trade-affected industries (or,the gender wage gap should narrow moreslowly in trade-affected industries). Womenmay be less skilled than men in the sensethat they have less actual labor market ex-perience than men; this has been shown tobe an important factor in explaining

changes in the gender wage gap (Blau andKahn 1997).

The key variable of interest for this pa-per, however, is the interaction betweenthe concentrated industry dummy variableand the change in import share variable.Given that we estimate a positive coeffi-cient on the trade variables, suggesting thatincreased trade increasesthe gap betweenmale and female residual wages in competi-tive industries, a negative coefficient on theinteraction term would indicate that theimpact of trade increased the gender wagegap by less in concentrated industries than

in competitive industries that were alsoaffected by trade. The coefficient on thisterm is indeed negative and statisticallysignificant, indicating that trade-affected,concentrated industries did experiencereductions in their residual gender wagegap relative to competitive industries alsohit by trade. The estimated coefficientssuggest that rising trade increased the gen-der wage gap in competitive industries (co-efficient of .268, with a standard error of.131), but actually reduced the gap in con-centrated industries (coefficient of .394,with a standard error of .25).20

To understand the economic importanceof these estimates, we calculate that the

still affect the results, we test for this possibility byestimating the basic regressions in Table 1 using theindividual as the level of analysis while correcting thestandard errors for clustering by industry. The resultsare nearly identical to those presented in Table 1 andare omitted here for brevity.

19See, for example, Murphy and Welch (1991),Wood (1994), and Borjas and Ramey (1995).

20Standardized coefficients are given in bracketsunderneath the estimates and are presented to en-able comparison across different regressions. Thestandardized coefficient is the estimated coefficientmultiplied by (standard deviation of the independentvari able / standard deviation of the dependent vari-able).




12/22


average increase in import share in concen-trated industries accounts for a decline inthe residual gender wage gap in manufac-turing of approximately .034 log points.(The overall decline in the residual genderwage gap in manufacturing was approxi-mately .14 log points during this period.)However, this positive effect is offset by the

rising residual gender wage gap in concen-trated industries relative to competitiveindustries and by the rising residual genderwage gap due to increasing imports as awhole.

The second column of Table 1 shows theresults of estimating equation (1) using theresidual gender gap in hourly earnings asthe dependent variable; the coefficient es-timates are similar to those in column (1),indicating that the results are insensitive to

the choice of weekly versus hourly wages.Column (3) uses an alternative measure ofindustry concentration, the price-cost mar-gin,21 to examine whether the results aresensitive to the measure of market struc-ture chosen. As shown, the key coefficientof interest remains negative and statisti-cally significant. The last two columns of

Table 1 increase the number of observa-tions by dividing the time period into twoperiods (the nine-year differences) andthree periods (the six-year differences),respectively, and include time dummies as

Table 1. Industry-Level Regression Results, CPS.(Dependent Variable: Change in Residual Gender Wage Gap)

9-Year 6-Year 197693 197693 197693 Differences Differences Weekly Hourly Weekly Weekly Weekly

Desc ription Earnings Earning s Earnings Earnings Earn ings

Concentrated Industrya .66** .65** .64** .46Change in Import Shareb (.28) (.28) (.29) (.32)

[.249] [.248] [.197] [.095]

Price-Cost Margin 4.55**Change in Import Share (2.29)

Concentrated Industry .19** .19** .09** .06**(.06) (.06) (.03) (.03)

Price-Cost Margin .12(.29)

Change in Import Share .27** .27** 1.31* .22 .17(.13) (.13) (.57) (.16) (.17)

1988 Dummy .001(.028)

1993 Dummy .02 .003(.02) (.028)

N 63 63 63 123 188Adjusted R2 .1253 .1260 .1150 .0287 .0005

Notes: Standard errors in parentheses; standardized coefficient in brackets. The observations are weightedby the inverse of the sampling variance of the dependent variable. Standardized coefficients are the coefficient(standard deviation of independent variable/standard deviation of dependent variable).

aA concentrated industry is defined as an industry with a four-firm concentrat ion ratio greater than or equalto .40 in the 1977 Census of Manufacturers.

bImport share is defined as imports/domestic shipments.*Statistically significant at the .10 level; **at the .05 level.

21The price-cost margin is defined as (value added labor costs)/(total sales) and was collected from theCensus of Manufacturers.




13/22


well . In both cases, the interaction betweenconcentrated industry and change in im-port share remains negative, and the stan-dardized coefficients suggest that the mag-

nitudes are similar. The coefficient is sta-tistically significant at the 2.9% level for thenine-year differences and at the 15% levelfor the six-year differences, suggesting thatmeasurement error becomes an increasingproblem as the sample is divided intosmaller and smaller time periods, as onewould expect.

Table 2 tests the sensitivity of these re-sults to the choice of data set. While theMarch CPS is an appropriate data set in thesense that its sample size is larger than thatof any longitudinal survey and it containsconsistent measures of the variables of in-

terest over the entire period under study, itis limited in that the cell sizes used forestimating equation (1) (that is, industryby year by gender) may be small. To in-crease the cell size, equation (1) is alsoestimated using the CPS Outgoing Rota-tion Group surveys as well as the 1980 and1990 Censuses.

The CPS Outgoing Rotation Group sur-veys are approximately three times largerthan the March supplement, a strong ad-

vantage for the empirical work undertakenhere. However, the Outgoing RotationGroup begins only in 1979, and it does notinclude information on the number of

weeks worked in the previous year. Thelatter problem prevents us from condition-ing on strong labor force attachment (num-ber of weeks worked), as we did with theMarch CPS. Despite these differences, how-ever, the results presented in the first twocolumns of Table 2 using the OutgoingRotation Groups are quite similar to thoseof Table 1 that used the March CPS. Thecoefficient on the interaction of concen-trated industry and change in import shareis still negative (and still statistically signifi-cant, although marginally so); in addition,the standardized coefficient suggests a

magnitude consistent with the estimatesbased on the March CPS. In both of theseregressions the change in import share vari-able is no longer statistically significant.

The last column of Table 2 presents theresults using the 1980 and 1990 Censuses1% sample. The obvious advantage of us-ing Census data is that the sample size isextremely large and therefore the industrycell sizes are much larger than in the case ofboth CPS data sets. However, because the

Table 2. Industry-Level Regression Results.(Dependent Variable: Change in Residual Gender Wage Gap)

CPS Outgoing Rota tion, 197993 Census, 198090

Independent Variable Weekly Earnings Hourly Earnings Weekly Earnings

Concentrated Industrya Change in Import Shareb .24 .25* .13*(.15) (.15) (.07)

[.305] [.317] [.302]

Concentrated Industry .07** .07** .02*(.03) (.03) (.01)

Change in Import Share .04 .04 .06*(.09) (.09) (.03)

N 64 64 74

Adjusted R2 .0344 .0415 .0321

Notes: Standard errors in parentheses; standardized coefficient in brackets. The observations are weightedby the inverse of the sampling variance of the dependent variable.

aA concentrated industry is defined as an industry with a four-f irm concentration ratio greater than or equalto .40 in the 1977 Census of Manufacturers.

bImport share is defined as imports/domestic shipments.*Statistically significant at the .10 level; **at the .05 level.




14/22


data span only ten years, there is less varia-tion in the change in import shares over theperiod.22 Nevertheless, the estimated coef-ficient on the interaction term is still nega-tive and statistically significant and, again,of the order of magnitude suggested by theother data sets.

To this point, the analysis has focused ontesting the impact of trade across indus-tries. This is appropriate because we areinterested in how trade, as a form of in-creased competition, differentially affectswages in competitive versus concentrated

manufacturing industries.23 This approachwould be less appropriate if one believedthat the changes in the gender wage gap inmanufacturing industries due to increasedtrade had spillover effects into non-manu-facturing industries. This argument, forexample, is similar to the argument givenin Borjas and Ramey (1995) for analyzingthe impact of trade on skill differentialsacross metropolitan areas rather than acrossindustries. In that paper, the authors ar-gued that the declining relative wages andemployment of less-skilled workers in con-centrated industries due to trade had

spillover effects on the wages of less-skilledworkers in the competitive sector of theeconomy; as a result, it is appropriate toanalyze the impact of trade across locallabor markets rather than across indus-tries.

While spil lover effects are unlikely to bestrong in the case of changes in the genderwage gap, we nevertheless test the sensitiv -ity of the results to this assumption by esti-mating equation (1) at the MSA level. Theresults of these tests are presented in Table3; the tests are conducted using both theMarch CPS and the 1980 and 1990 Cen-

suses.24 In this case, the MSA residual gen-der wage gap is calculated as the employ-ment-weighted average of the residual gen-der wage gap for each industry in manufac-turing in the MSA.25 The import share iscalculated similarly, as the employment-weighted average of the import share ineach manufacturing industry in the MSA;the concentration variable is defined as theshare of workers employed in a concen-trated industry in the MSA.

As indicated in Table 3, the results at theMSA level are essentially the same as the

estimates at the industry level. The coeffi-cients appear larger, but this is because theinteraction term is now the percentage ofemployment in the MSA that is in concen-trated industries, interacted with the changein the import share at the MSA level, in-stead of a zero-one dummy variable indicat-ing whether or not an industry is concen-trated interacted with the increased tradeat the industry level. The first two columnsshow the results of estimating equation (1)across MSAs using the March CPS over theentire 197693 period, using weekly andhourly earnings, respectively. These re-

sults are consistent with the industry-levelresults, and the coefficients on the interac-tion term are negative and statistically sig-nificant. The results are also similar whenthe 1980 and 1990 Census data are used,but the interaction term is no longer statis-tically significant. Note that the adjusted R-squared is negative in all cases, suggestingthat these regressions explain little of thevariation in the changes in the residualgender wage gap across metropolitan ar-eas. This is likely due in part to the rela-tively small variation in import shares andconcentrated industries across MSAs com-

22The 1970 Census is not used because there arefewer MSA indicators in that Census than in the laterCensuses.

23Numerous other studies have used industry-leveldata to examine the effects of trade; see, for example,Kruse (1988), Revenga (1992), Gaston and Trefler(1994), Kletzer (1996a, 1996b), Horn and Eastman(1997), and Campa and Goldberg (1998).

24The CPS Outgoing Rotation Group data sets arenot used for the MSA-level estimation because theylack consistent MSA identifiers over the relevant timeperiod.

25We include only manufacturing workers in theMSA estimation. This assumes that manufacturingand non-manufacturing workers are not close substi-tutes for each another.




15/22


pared with the variation in these variablesacross industries (see Appendix Table A2for means and standard deviations of theseand other variables).

Robustness Checks

One factor that may affect the resultsand that has thus far been omitted from thediscussion is the change in unionizationrates over the period. If, as seems likely,concentrated industries tended to be moreunionized than competitive industries, andmen are more highly unionized on average

than women, then the decline in unioniza-tion rates over this period would likely havereduced the gender wage gap more in con-centrated industries than in competitiveindustries. Moreover, if import shares rosemore in concentrated industries than incompetitive industries during this time, thechange in the import share in these regres-sions may simply be acting as a proxy for thechange in unionization rates, and the re-sults may simply reflect the impact of the

erosion of union power rather than theimpact of trade on the wage structure. Totest this possibility, column (1) of Table 4includes the change in the percentage ofworkers unionized in each industry in theregression. The results are virtually identi-cal to those without unionization, suggest-ing that the results do not reflect changesin unionization rates within industries.26

Another factor that may affect the resultsis technological change. The prevailingwisdom suggests that technological changehas primarily been skill-biased, which, giventhat women are disproportionately low-

skilled, may increase the gender wage gap;

Table 3. MSA Level Regression Results.(Dependent Variable: Residual Change in the Gender Wage Gap)

CPS, 197693 Census, 198090

Independent Variable Weekly Earnings Hourly Earnings Weekly Earnings

Percent in Concentrated Industrya 6.25* 6.42* 2.97Change in Import Shareb (3.37) (3.31) (2.51)

[1.10] [1.13] [.594]

Percent in Concentrated Industry .80* .80* .21(.47) (.46) (.19)

Change in Import Share 2.59* 2.69** 1.51(1.35) (1.34) (1.02)

Change in Unemployment Rate .36 .30 .25

(.93) (.92) (.33)N 43 43 132Adjusted R2 .0027 .0056 .0037

Notes: Standard errors in parentheses; standardized coefficient in brackets. The observations are weightedby the inverse of the sampling variance of the dependent variable.

aA concentrated industry is defined as an industry with a four-f irm concentration ratio greater than or equalto .40 in the 1977 Census of Manufacturers. Percent in concentrated industry is the employment-weightedaverage of the share of workers employed in a concentrated industry in each MSA.

bImport share is defined as the employment-weighted average of the import share in each industry in theMSA.

*Statistically significant at the .10 level; **at the .05 level.

26Note that changes in unionization and trade arelikely related to each another; see Horn and Eastman(1997) for an analysis of the impact of increased tradeon union density. A variable for the interactionbetween unionization and concentration in this re-gression is negative and statistically significant; how-ever, the sign and significance of the primary interac-tion term remain unchanged.




16/22


however, if technological change was bi-ased against manual labor, then womenmay have been relatively better off as aresult.27 For such a pattern to explain ourresults, it would have to be the case thattrade-affected competitive industries facedmore skill-biased technological change overthis time period than did trade-affectedconcentrated industries. The technologi-cal change would increase demand forskilled workers, driving up the wage pre-mium for both observed and unobservedskills. Because women are disproportion-

ately low-skilled, one would observe a risein the gender wage gap in trade-affectedcompetitive industries relative to trade-af-fected concentrated industries.28

In order to test this theory, we regressedthe percentage of workers who are college-educated (as a proxy for skilled workers) ineach industry on the same independentvariables as above. If the extent to whichobservable skills change is affected bywhether the industry is concentrated andwhether it is trade-affected, this might sug-gest that unobservable skills are changingin a similar manner and therefore that skill-

biased technological change is driving theresults. However, when we do estimate therelationship between skill composition andconcentrated trade-affected industries rela-tive to trade-affected competitive industries,we find no evidence that demand for skilledworkers increased in the latter relative tothe former. In other words, the coefficienton the interaction term is statistically insig-nificant.29

As a second test of the technologicalchange explanation, we also regressed thepercentage of workers who use a computerat work on the same set of independentvariables; this measure is intended to moredirectly capture the pace of technologicalchange across industries in the workplace.The data on computer use are from theOctober CPS, which asked questions aboutcomputer use in the workplace in 1984,1989, and 1993. The results indicate thatal though industr ies experiencing agreater trade shock were slower to imple-

ment technological change in the formof computer use, there was no interac-tion effect between this measure of tech-nological change and changing importshares (Table 4, column 2). This pro-vides supportive evidence that a trade-technology interaction does not explainthe results, that is, that concentration isnot simply acting as a proxy for technol-ogy. Column (3) of Table 4 indicatesthat, over the period examined in theseregressions (198493), the main resultsof interest continue to hold.

Finally, as a third test of the technologi-

cal change explanation, we estimate ouroriginal equation with the difference in theresidual gender wage gap as the dependentvariable and now include controls for bothchanging unionization rates and changingcomputer use in the industry. Note that,because we are using the computer usevariable, we estimate our equation on the198493 period. As column (3) of Table 4showed, our basic results hold for this sub-period. Column (4) presents the resultswith these extra controls; neither the chang-ing unionization rate nor the changing useof computers over this time explains away

our results.Another factor that may affect the results

is the choice of the time period over whichthe regressions are estimated. Several testsof the sensitivity of the results to this choicehave already been conducted, through theuse of the CPS Outgoing Rotation Groupand Census data sets that restricted theanalysis to years other than those tested inthe original specification. Specifically, wehave shown that the results hold whether

27See Welch (2000) for further discussion.28Note that because we are controlling for educa-

tion and age, these changes would have to be basedon unobservable skills and not just observable skills,assuming that unobserved skill is correlated withobserved skill.

29Like education, actual labor market experienceis a possible proxy for skill that could explain theresults. In order for experience to explain our re-sults, it would have to change differentially in concen-trated relative to competitive industries affected bytrade. Unfortunately, we are unable to test this pos-sibility, because the data sets used in our study lackinformation on actual labor market experience.




17/22


Table4.Regression

Resu

lts,CPS:

Specification

Checks.

Unions

TechnologicalChange

E

mployment

Minority

Changein

Changein

Changein

Changein

Percentage

Changein

Changein

Residual

Residual

PercentageofManagers

Residual

ResidualGender

Chan

gein

Gender

Gender

ofW

omen

WhoAre

White/Nonwhite

WageGap,

%Com

puter

Wage

Wage

Emp

loyees,

Women,

WageGap,

197693

Useat

Work,

Gap,

Gap,

197

693

197693

CPS-ORG,

DependentVariable:

WeeklyWages

1984

93

198493

198493

EmploymentEmployment

197993

IndependentVariable

(1)

(2

)

(3)

(4)

(5)

(6)

(7)

ConcentratedIndustry

a

.95*

.1

9

.92**

1.27**

.17

.40**

.35

ChangeinImportShareb

(.51)

(.1

5)

(.30)

(.66)

(.12)

(.14)

(.26)

[.315]

[.261]

[.318]

ConcentratedIndustry

.22**

.04*

.15**

.17**

.001

.021

.01

(.08)

(.0

2)

(.04)

(.06)

(.02)

(.03)

(.04)

ChangeinImportShare

.29**

.2

2**

.26

.27

.15

.12

.06

(.14)

(.0

9)

(.18)

(.19)

(.06)

(.09)

(.17)

ChangeinUnionization

.07

.42

(.30)

(.38)

Changein%ComputerUseatWork

.12(.27)

N

58

63

62

60

66

67

65

AdjustedR

2

.1141

.18

50

.1455

.1155

.0702

.1210

.0098

Notes:Standarderrorsinparentheses;standardizedcoefficientinbrackets.Theobservationsareweightedbythe

inverseofthesamplingvarianceofthe

dependentvariable.

aAconcentratedindustryisdefinedasanindustrywithafour-firmconcentrationratiogreaterthanorequalto.4

0inthe1977CensusofManufacturers.

bImportshareisdefinedasimports/domesticshipments.

*Statisticallysignificantatthe

.10level;**atthe.05level.




18/22


the initial period is 1976 (March CPS),1979 (CPS-ORG), 1980 (Census), or 1984,and whether the end period is 1990 (Cen-sus) or 1993 (March CPS and CPS-ORG).30

Two additional specification tests areimplemented in order to verify that in-creased competition through trade doesindeed reduce the ability of employers todiscriminate. These tests are based onfurther predictions of the Becker employerdiscrimination model, one regarding therelative employment of women and theother regarding the relative wage of mi-

norities. Regarding the former, Beckerstheory predicts that as discrimination isdriven away, not only will womens relativewages increase, but their relative employ-ment will increase as well. We have alreadyshown that as competition increases,womens relative wages increase. We nowtest the second prediction of the theory:that womens relative employment will in-crease as well. Column (5) of Table 4reports the results of regressing the changein the percentage of women employed inan industry on the same right-hand-sideva ri ab le s: the co nc en trated in dust ry

dummy, the change in import share overthe 197693 period, and the interaction ofthe two terms. Although it is statisticallysignificant at only the 15% level, the posi-tive coefficient on the interaction suggeststhat as industries face more competitionfrom international trade, concentrated in-dustries increase their relative employment

of women more than competitive indus-tries do.31

Another possibility is that women fail toadvance out of less-skilled occupations as aresult of discrimination. To test this, weexamine whether increased trade affectedthe percentage of managers in manufactur-ing who were women. We find statisticallysignificant evidence that concentrated in-dustries facing more competition from in-ternational trade increased the percentageof managers who were women (column 6).These results are consistent with Beckers

prediction and lend further credence tothe idea that trade has induced employersto reduce costly discrimination againstwomen.32

Since Beckers theory originally at-tempted to explain the consequences ofracial discrimination, it is fitting to testwhether the same predictions regardingwage differentials and market competitionhold if one examines the racial wage gaprather than the gender wage gap. Althoughthe forces influencing the relative wages ofminorities may have differed greatly fromthose influencing the relative wages of

women in this period, one might still ex-pect competitive pressures to affect a firmsability to discriminate against minorities inthe same way that competitive pressureswould affect its ability to discr iminateagainst women. Therefore, a final test is todetermine whether concentrated industriesdiffer from competitive industries in theextent to which trade affects the minority

30Addi tional sensiti vity tests indicate that the re-sults are similar whether 1991 or 1992 is used as thefinal year; this demonstrates that the results are notinfluenced by the CPS redesign that affected the 1993earnings data but not the 1991 or 1992 earnings data.The only year and dataset for which the results fail tohold is the 1994 March CPS. However, given that theresults do hold using the 1994 CPS-ORG dataset (theinteraction term is statistically significant at the 5%level for both the weekly and hourly earnings specifi-cations), and given the robustness of the results withrespect to other endpoints, we conclude that the 1994March CPS results are anomalous and that overall themodel is robust with respect to the choice of timeperiod used in the analysis.

31These gains in female relative employment alsosuggest that the improvement in relative female wagesin trade-affected, concentrated industries was notdue to women disproportionately dropping out of thelabor force due to the impact of trade in these indus-tries.

32One concern might be that our results are pick-ing up a relative demand shift that favored female-dominated occupations. However, the fact that wefind similar results when we look within broadly de-fined occupation groups (in this case, managers)supports our interpretation of our results as reflect-ing a change in discriminatory practices and not anacross-occupation demand shift.




19/22


residual wage gap (defined as the differ-ence in the average residual wage of whitemen in an industry minus the average re-sidual wage of nonwhite men in that indus-try). Because of the limited number ofminorities working in the manufacturingindustries in the sample, we use the CPSOutgoing Rotation data set to increasethe sample size. Table 4, column (7)presents the results of estimating equa-tion (1) using the change in the minorityresidual wage gap from 1979 to 1993 asthe dependent variable. Although not

statistically significant, the coefficient onthe interaction term is negative and ofthe same magnitude as earlier estimatesfor the gender wage gap, which is consis-tent with the hypothesis that increasedcompetition through trade reduces theemployers ability to discriminate.

Conclusion

Theory predicts that product marketcompetition will drive out discriminationin the labor market. Because discrimina-tion is costly in the sense that discriminat-

ing employers must forego profits in orderto indulge their taste for discrimination,firms with market power can afford to con-tinue discriminatory practices for longerthan can firms in competitive markets earn-ing zero economic profits. Thus, the loss ofmarket power in an industry is likely toreduce discrimination and increase the rela-tive wages and employment of women in

that industry. While a number of studieshave demonstrated the apparent existenceof discrimination, few have focused on thisdynamic implication.

In testing this idea across manufactur-ing industries in the United States, wehave assumed that increased interna-tional trade in recent years has acted as aform of increased competition in someindustries. Our approach compares theimpact of trade in concentrated versuscompetitive industries , and enables us tonet out the gains in relative female wages

that occurred over the period for otherreasons. The results indicate that theresidual gender wage gap narrowed morerapidly in concentrated industries thatexperienced a trade shock than in com-petitive industries that experienced atrade shock. Moreover, the results areconsistent across a variety of specifica-tions and data sets.

Although it is unlikely that increasedtrade had a substantial impact on the over-all gender wage gap in the economythemanufacturing sector currently comprisesonly about 15% of the U.S. work forcethe

empirical work in this paper suggests thatthe impact of trade on the structure ofwages should be viewed in a more positivelight than it recently has. Although trademay increase wage inequality by modestlyreducing the relative wages of less-skilledworkers, at the same time it appears tobenefit women by reducing the ability offirms to discriminate.




20/22


Appendix Table A1Industries Categorized by Whether Concentrated and Whether Affected by Trade

Concentrated Industriesa Non-Concentrated Industriesa

CIC Code Indust ry a CIC Code Industry

Non-Trade-Affected Industriesa

110 Grain Mill Products 100 Meat Products130 Tobacco Manufacturers 101 Dairy Products140 Dyeing & Finishing Textiles, Except 102 Canned & Preserved Fruits & Vegetables

Wool & Knit 111 Bakery Products182 Soaps, Cosmetics 112 Sugar & Confectionery Products250 Glass & Glass Products 120 Beverage Industries262 Misc. Nonmetallic Mineral & Stone 121 Misc. Food Prep. & Kindred Products

Products 141 Floor Coverings, Except Hard Surfaces

280 Other Primary Metal Industries 142 Yarn, Thread & Fabric Mills291 Metal Forgings & Stampings 150 Misc. Textile Mill Products292 Ordnance 160 Pulp, Paper, Paperboard Mills310 Engines & Turbines 161 Misc. Paper & Pulp Products311 Farm & Machinery Equipment 162 Paperboard Containers & Boxes352 Aircraft & Parts 181 Drugs360 Ship & Boat Building & Repairing 190 Paints, Varnishes, Related Products361 Railroad & Locomotive Equipment 191 Agricultural Chemicals

192 Industrial & Misc. Chemicals200 Petroleum Refining201 Misc . Pet ro leum & Coal Product s241 Misc. Wood Products242 Furniture & Fixtures251 Cement, Concrete , Gypsum, Plas te r

Products271 Iron & Steel Foundries282 Fabricated Structural Metal Products

290 Screw Machine Products300 Misc . Fabr icated Meta l Products341 Radio, T.V., Communications Equipment370 Cyc le s & Misc. Transportat ion

Equipment372 Opti ca l & Health Serv ices Suppl ies

Trade-Affected Industriesa

380 Photographic Supplies & Equipment 132 Knitting Mills252 Structural Clay Products 151 Apparel & Accessories, Except Knit 261 Pottery & Related Products 152 Misc. Fabricated Textile Products312 Construction & Material Handling 211 Other Rubber Products, Plastics

Machines Footwear, Belting321 Office & Accounting Machines 221 Footwear, Except Leather & Plastic322 Electronic Computing Equipment 222 Leather Products, Except Footwear340 Household Appliances 281 Cutlery, Hand Tools, Other Hardware342 Electrical Machinery, Equipment, Supplies 320 Metalworking Machinery

351 Motor Vehicles and Motor Vehicle 331 Machinery, Except ElectricalEquipment 371 Scientific & Controlling Instruments391 Misc . Manufacturing Industrie s

aA trade-af fected industry is defined as one in wh ich the import share increased by at least .10 between 1976and 1993. A concentrated industry is defined as having a four-firm concentration ratio of greater than .40 in1977.




21/22


Appendix Table A2Summary Statistics

(Standard Deviations in Parentheses)

Industry MSA

CPS, Census Outgoing CPS, Census Variable 197693 Data Rotation 197693 Data

Change in Residual Gender Wage Gap in .138 .066 .068 .186 .075Manufacturing (Weekly Earnings) (.157) (.031) (.089) (.159) (.060)

Change in Residual Gender Wage Gap in .135 .068 .185Manufacturing (Hourly Earnings) (.156) (.089) (.159)

Percent in Concentrated Industry .052 .037 .051 .043 .029Change in Import Share (.101) (.072) (.113) (.028) (.012)

Percent in Concentrated Industry .309 .354 .301 .349 .376(.466) (.481) (.462) (.152) (.113)

Change in Import Share .097 .079 .086 .121 .076(.275) (.115) (.270) (.045) (.020)

Change in Union Membership .137 .032 .130 .175 .047(.078) (.053) (.069) (.037) (.029)

Change in Unemployment Rate .001 .002(.027) (.016)

N 63 74 64 43 132

REFERENCES

Angr ist, Joshua D. , and Alan B. Krueger. 1999. Em-pirical Strategies in Labor Economics. In Orley C.Ashenfel ter and David Card, eds. , Handbook of LaborEconomics, Vol. 3A. Amsterdam: Elsevier, pp. 12771366.

Ashenfelter , Orley, and Timothy Hannan. 1986. SexDiscrimination and Product Market Competition:The Case of the Banking Industry. Quarterly Journalof Economics, Vol. 101, No. 1 (February), pp. 14973.

Bartelsman, Eric J., and Wayne Gray. 1996. TheNBER Manufacturing Productivity Database. NBERTechnical Working Paper 205.

Becker, Gary S. 1957, 1971. The Economics of Discrimi-nation. Chicago: University of Chicago Press.

Bednarzik, Robert W. 1993. An Analysis of U.S.Industries Sensitive to Foreign Trade, 198287.Monthly Labor Review, Vol. 116, No. 2 (February), pp.

1531.Black, Sandra E., and Philip E. Strahan. 2001. The

Division of Spoils: Rent-Sharing and Discrimina-tion in a Regulated Industry. American EconomicReview, Vol. 91, No. 4 (September), pp. 81431.

Blau, Francine D. 1977. Equal Pay in the Off ice .Lexington, Mass: Lexington Books.

____. 1998. Trends in the Well-Being of AmericanWomen, 19701995. Journal of Economic Literature,Vol. 36, No. 1 (March), pp. 11265.

Blau, Francine D., and Lawrence M. Kahn. 1997.Swimming Upstream: Trends in the Gender WageDifferential in the 1980s. Journal of Labor Economics ,

Vol. 15, No. 1, Part 1 (January) , pp. 142.Borjas, George J., and Valerie A. Ramey. 1995. For-eign Competition, Market Power, and Wage In-equality. Quarterly Journal of Economics, Vol. 110,No. 4 (November), pp. 10751110.

Bound, John, et al. 1994. Evidence on the Validityof Cross-Sectional and Longitudinal Labor MarketData. Journal of Labor Economics, Vol. 12, No. 3(July), pp. 34568.

Campa, Jose Manuel, and Linda S. Goldberg. 1998.Employment versus Wage Adjustment and the U.S.Dollar. Unpublished paper, Federal Reserve Bankof New York.

Feenstra, Robert C. 1996. U.S. Imports, 19721994:Data and Concordances. NBER Working PaperNo. 5515.

Freeman, Richard B. 1995. Are Your Wages Set in

Beijing? Journal o f Economic Per spec tiv es, Vol. 9, No.2 (Summer), pp. 1532.

Gaston, Noel, and Daniel Trefler. 1994. Protection,Trade, and Wages: Evidence from U.S. Manufactur-ing. Industrial and Labor Relations Review, Vol. 47,No. 4 (July), pp. 57492.

Goldberg, Matthew S. 1982. Discrimination, Nepo-tism, and Long-Run Wage Differentials. QuarterlyJournal of Economics, Vol. 97, No. 2 (May), pp. 30719.

Goldin, Claudia. 1990. Understanding the Gender Gap:An Economic History of American Women. Oxford:Oxford University Press.




22/22


Groshen, Erica L. 1991. The Structure of theFemale/Male Wage Differential: Is It Who You Are,What You Do, or Where You Work? Journal ofHuman Resources, Vol. 26, No. 3 (Summer), pp. 45772.

Harrison, Ann E. 1994. Productivity, ImperfectCompetition, and Trade Reform: Theory and Evi-dence. Journal of International Economics, Vol. 36,pp. 5373.

Heckman, James J. 1998. Detecting Discrimina-tion. Journal of Economic P erspect ives, Vol. 12, No. 2(Spring), pp. 10116.

Hellerstein, Judith K., David Neumark, and KennethR. Troske. 2002. Market Forces and Sex Discrimi-nation. Journal of Human Resources, Vol. 37, No. 2(Spring), pp. 35380.

Horn, John T., and J. Gregory Eastman. 1997. Im-porting Unionization? The Effects of Import Com-petition on U.S. Manufacturing Unionization Den-sity, 19741992. Unpublished paper.

Jaeger, David A. 1997. Reconciling the Old and NewCensus Bureau Education Questions: Recommen-dations for Researchers. Journal of Business &Economic Statist ics , Vol. 15, No. 3 (July), pp. 300309.

Jaeger, David A., Susanna Loeb, Sarah E. Turner, andJohn Bound. 1998. Coding Geographic Areasacross Census Years: Creating Consistent Defini-tions of Metropolitan Areas. NBER Working Paper6772.

Katics, Michelle M., and Bruce C. Petersen. 1994.The Effect of Rising Import Competition on Mar-ket Power: A Panel Data Study of U.S. Manufactur-ing. Journal of Industria l Economics, Vol. 42, No. 3

(September), pp. 27786.Katz, Lawrence F., and Kevin M. Murphy. 1992.

Changes in Relative Wages, 19631987: Supplyand Demand Factors. Quarterly Journal of Economics,Vol. 107, No. 1 (February), pp. 3578 .

Kletzer, Lori G. 1996a. The Impact of ForeignCompetition on Occupational Employment andWages in U.S. Manufacturing, 197192. Unpub-lished paper, University of California, Santa Cruz.

____. 1996b. Gender Differences in the Incidenceand Consequences of Job Displacement from Im-port-Sensitive Industries. Unpublished paper,University of California, Santa Cruz.

____. 2003. Imports, Exports and Jobs: What Does TradeMean for Employment and Job Loss? Kalamazoo, Mich.:W.E. Upjohn Institute for Employment Research.

Kruse, Douglas L. 1988. International Trade andthe Labor Market Experience of Displaced Work-ers. Industrial and Labor Relations Review, Vol. 41,No. 3 (April), pp. 40217.

Murphy, Kevin M., and Finis Welch. 1991. The Roleof International Trade in Wage Differentials. InMarvin Kosters, ed., Workers and Their Wages. Wash-ington, D.C.: AEI Press, pp. 3969.

Neumark, David, with the assistance of Roy J. Bankand Kyle D. Van Nort. 1996. Sex Discriminationin Restaurant Hiring: An Audit Study. QuarterlyJournal of Economics, Vol. 111, No. 3 (August), pp.91541.

ONeill, June, and Solomon Polachek. 1993. Whythe Gender Gap in Wages Narrowed in the 1980s.Journal of Labor Economics, Vol. 11, No. 1 (Part 1), pp.20528.

Revenga, Ana. 1992. Exporting Jobs? The Impact ofForeign Competition on Employment and Wages inU.S. Manufacturing. Quarterly Journal of Economics,Vol. 107, No. 1 (February) , pp. 25584.

Ruggles, Steven, and Matthew Sobek. 1997. Inte-grated Public Use Microdata Series: Version 2.0.Minneapolis: Historical Census Projects, Universityof Minnesota.

Sachs, Jeffrey D. 1988. Global Adjustments to aShrinking U.S. Trade Deficit. Brookings Papers onEconomic Activit y, Vol. 88, No. 2, pp. 63974.

Weinberger, Catherine J . 1998. Race and GenderWage Gaps in the Market for Recent Col lege Gradu-

ates. Industrial Relations, Vol. 37, No. 1 (January),pp. 6784.

Welch, Finis. 2000. Growth in Womens RelativeWages and in Inequality among Men: One Phenom-enon or Two? American Economic Review Papers &Proceedings, Vol. 90, No. 2 (May), pp. 44449.

Wood, Adrian. 1994. North-South Trade, Employment,and Inequality. Oxford: Clarendon Press.

Wood, Robert G., Mary E. Corcoran, and Paul N.Courant. 1993. Pay Differences among the HighlyPaid: The Male-Female Earnings Gap in LawyersSalaries. Journal of Labor Economics, Vol. 11, No. 3(July), pp. 41741.


V_Black and Brainerd - Importing Equality - The Impact of Globalization on Gender Discrimination

Documents