bertrand, m. and k. f. hallock (2000). -the gender gap in top corporate jobs.-

Cornell University ILR SchoolDigitalCommons@ILR

Faculty Publications - Human Resource Studies Human Resource Studies

10-1-2001

The Gender Gap in Top Corporate JobsMarianne BertrandUniversity of Chicago

Kevin F. HallockCornell University, [email protected]

Follow this and additional works at: http://digitalcommons.ilr.cornell.edu/hrpubsPart of the Human Resources Management Commons

This Article is brought to you for free and open access by the Human Resource Studies at DigitalCommons@ILR. It has been accepted for inclusion inFaculty Publications - Human Resource Studies by an authorized administrator of DigitalCommons@ILR. For more information, please [email protected].

The Gender Gap in Top Corporate Jobs

AbstractUsing the ExecuComp data set, which contains information on the five highest-paid executives in each of alarge number of U.S. firms for the years 199297, the authors examine the gender compensation gap amonghigh-level executives. Women, who represented about 2.5% of the sample, earned about 45% less than men.As much as 75% of this gap can be explained by the fact that women managed smaller companies and wereless likely to be CEO, Chair, or company President. The unexplained gap falls to less than 5% with anallowance for the younger average age and lower average seniority of the female executives. These results donot rule out the possibility of discrimination via gender segregation or unequal promotion. Between 1992 and1997, however, women nearly tripled their participation in the top executive ranks and also strongly improvedtheir relative compensation, mostly by gaining representation in larger corporations.

Keywordsgender, gap, compensation, corporate, pay, women, data, age, discrimination, executive, ceo

DisciplinesHuman Resources Management

CommentsSuggested CitationBertrand, M., & Hallock, K. (2001). The gender gap in top corporate jobs [Electronic version]. Industrial andLabor Relations Review, 55, 3-21.http://digitalcommons.ilr.cornell.edu/hrpubs/14/

Required Publisher StatementPosted with permission from the Industrial and Labor Relations Review.

This article is available at DigitalCommons@ILR: http://digitalcommons.ilr.cornell.edu/hrpubs/14

3Industrial and Labor Relations Review, Vol. 55, No. 1 (October 2001). by Cornell University.0019-7939/00/5501 $01.00

T

THE GENDER GAP IN TOP CORPORATE JOBS

MARIANNE BERTRAND and KEVIN F. HALLOCK*

Using the ExecuComp data set, which contains information on the fivehighest-paid executives in each of a large number of U.S. firms for the years199297, the authors examine the gender compensation gap among high-levelexecutives. Women, who represented about 2.5% of the sample, earned about45% less than men. As much as 75% of this gap can be explained by the fact thatwomen managed smaller companies and were less likely to be CEO, Chair, orcompany President. The unexplained gap falls to less than 5% with an allowancefor the younger average age and lower average seniority of the female execu-tives. These results do not rule out the possibility of discrimination via gendersegregation or unequal promotion. Between 1992 and 1997, however, womennearly tripled their participation in the top executive ranks and also stronglyimproved their relative compensation, mostly by gaining representation inlarger corporations.

*Marianne Bertrand is Assistant Professor of Eco-nomics at the Graduate School of Business at theUniversity of Chicago, Faculty Research Fellow at theNational Bureau of Economic Research, and an Affili-ate of CEPR. Kevin Hallock is Associate Professor ofEconomics and of Labor and Industrial Relations atthe University of Illinois at Urbana-Champaign.

For helpful comments, the authors thank seminarparticipants at the University of Illinois, StanfordUniversity, and the joint meeting of the Society ofLabor Economists and European Society of LabourEconomists in Milan, as well as Marianne Ferber,Peter Feuille, Todd Fister, Wallace Hendricks, JohnJohnson, and Craig Olson. Sherrilyn Billger pro-vided outstanding research assistance.

1See, for example, Catalyst (1999), Morris (1998),Jones (1999), and Meyer (1999).

Some of the data used in this study are fromStandard and Poors ExecuComp data base and mustbe purchased from Standard and Poors. Computerprograms used in the analysis are available from theauthors upon request. The first drafts of this paperwere written while Bertrand was on the faculty atPrinceton and Hallock was visiting the IndustrialRelations Section at Princeton. E-mail: [email protected]; [email protected].

his paper analyzes gender differencesamong top executives in a large set of

U.S. public corporations. Our motivationfor undertaking this study is twofold. Firstand foremost is the fact that, notwithstand-ing the curiosity this topic raises both in themedia and in policy circles,1 we know of no

systematic study to date of how well womenare doing in top corporate jobs. We pro-vide the first detailed description of therelative position of female top executives inthe 1990s.

Our second motivation is more academic.Problems plaguing many past studies of thegender pay gapin particular, unobservedcharacteristics of both workers and jobsare likely to be less present in this specificoccupational group. Most of the previouswork has indeed identified an unexplainedgender gap that cannot be attributed to

4 INDUSTRIAL AND LABOR RELATIONS REVIEW

observable differences between men andwomen.2 While this unexplained gap couldbe due to labor market discrimination, itcould also be attributable to differencesbetween men and women that are unob-servable (at least to the econometrician),such as a relative lack of long-term careercommitment among women.3 It is reason-able to assume that such unobservable dif-ferences are minimized in the group of topexecutives we propose to study. Men andwomen in this sample are likely to be simi-lar in that both share a high level of jobmotivation and high career ambitions.

Several authors have previously exam-ined gender pay differences among thehighly paid. Examples include investiga-tions of lawyers by Wood, Corcoran, andCourant (1993), and Biddle andHamermesh (1998); of university faculty byBarbezat (1987), Barbezat and Hughes(1990), Ferber and Greene (1982), Gander(1997), Hoffman (1976), Johnson andStafford (1974), Katz (1973), and Ransomand Megdal (1993); of engineers by Mor-gan (1998); of physicians by Baker (1995);and of firm managers in the United King-dom by Gregg and Machin (1993).4 No onebefore, however, has focused on gendercompensation differentials among top ex-ecutives. There have been two substantialbarriers to conducting an investigation such

as ours. First, the required data simply didnot exist before. Second, it has been widelybelieved that too few women were in thesetop positions to carry out a formal analysisof their relative pay.

We use Standard and Poors ExecuCompdata, which contain information on com-pensation for the top five executives for allfirms in the S&P 500, S&P Midcap 400, andS&P SmallCap 600 for the years 199297.Included is information on base salary,bonus, and the value of granted stock op-tions in the current year.5 The ExecuCompdata set has three main advantages for ourpurpose. First, it is very large. The samplewe use in most of our analysis includesmore than 42,000 executive-year observa-tions. All publicly traded firms are re-quired to disclose the names and compen-sation of the top five highest-paid em-ployees annually. This large sample size isespecially important for us because we wantto estimate gender differences with suffi-cient statistical precision in an economicsector where female representation is small.A second advantage of the data set is that itcovers a variety of occupational categoriesamong the top managerial jobs and notonly Chief Executive Officers (CEOs). Weare thus able to investigate the importanceof occupational differences at the top. Fi-nally, because the data set covers a widecross-section of firms, the role of firm sizeand industrial specialization in the gendercompensation gap can be assessed.

The Gender Gap

The ExecuComp data set is unique formany reasons, including its wide variety ofmeasures of compensation, details concern-ing firm characteristics, and large samplesize. We can also arrange the data as a

2An exception is Groshen (1991). Groshen showedthat most of the gender gap can be attributed to sexsegregation rather than wage differences by sex withinoccupation, industries, and establishments. Using alarger sample but a similar empirical methodology,Bayard, Hellerstein, Neumark, and Troske (1999),however, found that a large part of the sex gap re-mains unexplained after accounting for sex segrega-tion.

3Of course, it is also possible that lower pay leadsto lower career commitment. In addition, it could bethat some compensating differential exists wherebywomen on average select lower-paying jobs than menand, at the same time, enjoy amenities on these jobsthat the higher-paying jobs lack. We cannot explorethis issue empirically in this paper.

4Gregg and Machin (1993) explored pay gaps fora much more general class of managers than the classwe focus on. Another study that focuses on suchlower-level managers for the United States is Jacobs(1992).

5Most studies of CEO pay do not include the valueof stock options granted in a given year. Hall andLiebman (1998), however, documented the growingimportance of granted options in the compensationof CEOs since the early 1980s. ExecuComp reportsthis information, and we have included it in our totalcompensation measure.

GENDER GAP IN TOP CORPORATE JOBS 5

panel, since we have multiple observationson a set of firms over time. Most crucial forour work, however, is the identification ofthe gender of each manager. Given thesubstantial discussion of a dearth of womenin managerial positions in the United States(see, for example, Catalyst 1999), we wereconcerned that examining the question ofa compensation gap would be difficult.However, due to ExecuComps substantialsize, we were able to identify more than1,134 female executive-year observations(449 unique individuals) on the basis of thegender variable included in the data. Thisis roughly 2.4% of all observations in thesample.

Panel A of Table 1 summarizes meancompensation by gender for the basicExecuComp sample. The table displaystotal compensation but also decomposestotal compensation into its major elements:salary, bonus, other annual compensation,and the value of options granted in thecurrent year. Pooling all the ExecuCompyears together, total compensation was, onaverage, 33% lower for women than formen.6 On average, women earned a little

Table 1. Summary Statistics on Compensation, 199297.(Standard Errors in Parentheses)

(1) (2) (3) (4)Characteristic All Managers Men Women p-Valuea

Panel A: High-Level Managersb

Total Current Payc 1,323.0 1,333.7 894.1 0.000(13.8) (14.1) (58.1)

Salary 336.3 338.6 246.9 0.000(1.0) (1.0) (6.1)

Bonus 257.6 260.7 136.1 0.000(3.8) (3.8) (8.4)

Other Annual Payd 20.6 20.8 12.7 0.006(0.7) (0.7) (2.8)

Value Granted Optionse 489.7 492.6 370.8 0.003(10.5) (10.7) (39.5)

N 46,708 45,574 1,134

Panel B: Managers from CPSfAnnual Labor Earnings 45.6 52.4 36.0 0.000

(0.09) (0.13) (0.11)

N 73,411 43,011 30,400

Sources: The data on high-level managers in panel A are from Standard and Poors ExecuComp database for19921997. The data on managers from the CPS are from the 1992 to 1997 Merged Outgoing Rotation Groupsof the Current Population Survey.

Notes: All data are reported in real 1997 thousands of dollars adjusted using the consumer price index.aThis is the p-value for the difference in sample means between men and women within each row.bHigh-level managers include the top five highest-paid executives in each firm in the ExecuComp database.cTotal current pay is the sum of salary, bonus, other annual pay, and the value of stock options granted in the

current year.dOther annual pay includes the dollar value of annual compensation not categorized as salary or bonus.eValue of granted options is the value of stock options granted in the current period. This is not the value

of options cashed in in a given year.fThe managers from the CPS are the set of full-time workers who report an occupation category between 3

and 22 in the 1980 Census of Population Occupation Classification. Annual income is constructed from averageweekly earnings.

6When we later control for year effects, the gendergap in total compensation reaches about 45%, thenumber reported in the introductory paragraphs.


less than $900,000 (1997 dollars) in totalcompensation, compared to more than $1.3million for the average male executive.

How does the gender gap among top-level managers compare to that amonglower-level managers? Panel B addressesthis question. We use the Merged Outgo-ing Rotation Groups of the Current Popu-lation Survey (CPS) over the same period,199297. Given that the ExecuComp dataare so specialized, we define a manager asanyone reporting that he or she worked inan executive, administrative and manage-rial occupation, excluding managementrelated occupations.7 We focus on full-time workers only, that is, individuals whoworked at least 35 hours per week. Annualsalaries are constructed based on averageweekly earnings. One can see that thegender earnings gap among middle-levelmanagers is very similar to that among topmanagers: about 46%.

We now consider gender differences inthe composition of the compensation pack-age. Several features are worth noticing.First, women seem to have received a largershare of their compensation in the form ofstock options than did men. This patternmay reflect the fact that the sample ofwomen was larger in the later years, whenthe use of stock options was more com-mon.8 Also, compared to men, womenreceived less compensation in the form ofbonuses and more in the form of salary.

Decomposing the Gender Gap

In this section, we investigate how vari-ous characteristics of female top executive

employment might account for the gendergap. We explore issues such as firm size,industrial segregation, occupational segre-gation, and individual demographic char-acteristics.

The Role of Firm Size

Women in top managerial positionstended to work for much smaller corpora-tions than did men. Panel A of Table 2clearly illustrates this fact. Female execu-tives firms were 3545% smaller, whethersize is measured as the value of shareholderwealth, sales, total assets, or number ofemployees.9 In an analysis not reportedhere, we found that companies of extremesize were chiefly responsible for the rela-tionship between firm size and gender. Wecomputed the fraction of women by decilesof firm market value. Women constitutedabout 3.5% of top management employ-ment in the bottom two deciles and only1% in the top decile. In all the otherdeciles, the fraction of women fluctuatedbetween roughly 2% and 3%, and the de-cline was not monotonic in size.

It is a well-known fact in the executivecompensation literature that CEOs tend tobe paid more the larger the firms size(Murphy 1985; Kostiuk 1990; Rosen 1992).If this pay-size correlation also holds forother top executives, it is reasonable to askhow much of the gender gap can be attrib-uted to the under-representation of womenin large firms. The first columns of Table 3answer this question.

The dependent variable for all regres-sions in Table 3 is the logarithm of real totalcompensation. All regressions in the paper(Tables 3, 6, 8, and 9) include yearly timeindicators, and standard errors are White-corrected standard errors. One can see incolumn (1) that the gender gap is larger,44%, when one controls for year effects

Controlling for year effects is crucial, because, as wewill show below (in the section Trends in Participa-tion and Earnings), the number of women was largerin the later years of the sample, when compensationlevels were on average higher.

7This corresponds to categories 322 in the 1980Census of Population Occupation Classification.Obviously, this definition formally also includeshigher-level managers. But because this group issmall, most of the individuals in the CPS sample arelower-level managers.

8See the section Trends in Participation and Earn-ings.

9Shareholder wealth is the total number of sharestimes the year-end share price. The correlation be-tween ln(shareholder wealth) and ln(assets) is 0.8;between ln(shareholder wealth) and ln(number ofemployees), 0.7; and between ln(assets) andln(employees), 0.7.


than the gap implied by Table 1, whichdoes not control for any covariates. Thiscan be explained by the fact that there aremore observations for women in the latersample years. Column (2) shows that thegender gap is, as we had expected, substan-tially reduced when we control for the valueof shareholder wealth. The elasticity ofmanagerial compensation to the value ofshareholder wealth is about 0.4.10 About athird of the gender compensation differen-tial, or 15 percentage points, can be ac-counted for by the lower participation ofwomen in large firms.

The Role of Industrial Segregation

The female executives in our sample werenot uniformly represented in all industrial

sectors. This can be seen in Table 4. Womenwere more likely to be managing compa-nies that specialized in health and socialservices and in trade. These were alsosectors in which a disproportionate shareof lower-level managers were women, as wecan see from the CPS results in column 6.In contrast, very few women held top-levelpositions in agriculture, construction, min-ing, and heavy manufacturing industries.

The banking sector is an interesting case.While it had the largest share of women inlower-level management among all sectors,the share of women at the top was lowerthan in most other sectors.11 Does the

Table 2. Firm and Manager Characteristics.(Standard Errors in Parentheses)

(1) (2) (3) (4)Characteristic All Managers Men Women p-Valuea

Panel A: Firms

Market Value (millions) 3,768.7 3,799.3 2,538.3 0.000(45.7) (46.4) (245.4)

Salesb (millions) 3,423.6 3,470.9 1,893.5 0.000(42.1) (43.0) (117.9)

Assetsc (millions) 7,473.0 7,525 5,379.6 0.000(117.2) (119.1) (644.6)

Employeesd (thousands) 16.9 17.1 9.2 0.001(0.2) (0.2) (0.5)

Ne 46,708 45,574 1,134

Panel B: Managers

Age 52.6 52.6 47.5 0.000(0.5) (0.06) (0.5)

N 17,236 16,960 276

Seniority 13.2 13.3 7.7 0.000(0.1) (0.1) (0.5)

N 14,189 13,845 344

Source: The data are from Standard and Poors ExecuComp database for 199297.aP-value for difference in sample means by gender for each variable.bSample sizes for sales are 46,665, 45,533, and 1,132.cSample sizes for assets are 46,703, 45,569, and 1,134.dSample sizes for employees are 45,581, 44,482, and 1,099.eSample size for Market Value. This size variable is used in most of the analysis below.

10This elasticity is slightly higher than that docu-mented in Rosen (1992) for CEOs only.

11Bird (1990) previously documented substantialgrowth of female employment in bank managementstarting in the 1970s, partly as a result of pressures bythe Equal Employment Opportunities Commission(EEOC). She further noticed that these women


apparent industrial segregation of femaleexecutives account for some of the gendergap in compensation? The data reportedin Table 4 show no obvious pattern of aconcentration of women in low-wage indus-tries. While managers in health and socialservices as well as in trade were paid slightlyless than the average manager in the sample,managers in the industries where womenwere very scarce were paid below average,too.

Columns (3)(5) in Table 3 confirm thisobservation in a more rigorous statisticalway. Columns (3) and (4) show that thefemale dummy variable stays unchangedwhether we add 8 broad industry dummiesor 115 finer dummies. Moreover, none ofthe gender gap remaining after controllingfor firm size (column 2) can be accountedfor by the differential representation ofwomen in different industries (column 5).In summary, there is no evidence of a sys-tematic allocation of women in low-payingindustries.

The Role of Occupational Segregation

Table 5 presents the share of women invarious occupations. We constructed occu-pational categories based on the title vari-able in ExecuComp. There are more than5,100 unique occupation tit les inExecuComp. Some of these titles clearlyrepresent similar occupations. For example,ex. vp and exec. vp are just differentways of representing executive vice presi-dent. But many are more complicated andcannot naturally be merged together. Webroke the occupation categories into 31unique groupings, including Chair andCEO, Vice-Chair, President, Chief Finan-cial Officer (CFO), Chief Operating Of-ficer (COO), and so on. Because some ofthe executives in the sample reported morethan one occupation in their job title, weconstructed two different occupational cat-egories for the first and second occupationreported for each manager in ExecuComp.

The occupational breakdown reportedin Table 5 is a further consolidation of our31 categories into only 11 based on the firstoccupation reported, except for the Chairand CEO category. Indeed, as most of theCEOs in the sample are also Chairs of their

Table 3. Gender Pay Gap for High-Level Executives.(Dependent Variable is the Log of Total Compensation;

White-Corrected Standard Errors in Parentheses)

Variable (1) (2) (3) (4) (5)

Female 0.44*** 0.28*** 0.44** 0.43*** 0.27***(0.05) (0.04) (0.05) (0.04) (0.03)

Market Value 0.37*** 0.39***(0.004) (0.005)

Stock Return/1,000 0.04 0.02(0.03) (0.03)

8 Industries no no yes no no

115 Industries no no no yes yes

Constant 6.48*** 3.86*** 6.48*** 6.48*** 3.89***(0.01) (0.03) (0.01) (0.01) (0.02)

R2 0.030 0.345 0.056 0.177 0.410

N 46,670 46,670 46,670 46,670 46,670

Source: The data are from Standard and Poors ExecuComp database for 199297.Notes: All regressions control for time indicator variables.*Statistically significant at the .10 level; **at the .05 level; ***at the .01 level.

tended to be mostly employed in retail banking andespecially branch management, where chances ofadvancement were very low.


companies and sometimes reported theirtitle as CEO and Chairman and some-times as Chairman and CEO, all respon-dents who reported at least one of theseoccupations in their title were put in theCEO/Chair category. Finally, we rankedthese occupations based on our intuitiveassessment of their relative prestige. Col-umn (3) of Table 5 reports the mean com-pensation for each occupation relative tothe overall mean compensation in thesample. This confirms that our intuitionwas roughly correct except with respect toCFOs, whose relatively low compensationon average in our sample came as a surpriseto us.

The most important fact in Table 5 is theunder-representation of women in the topthree occupational categories and top fouroccupations (Chair, CEO, Vice-Chair, andPresident).12 Women who had made it intothe top managerial level (that is, they were

in the ExecuComp sample) were less likelyto be at the very top than were men.13 Thefraction of women among CEOs, Chairs,and Vice-Chairs was much less than 1%.There were also fewer female presidentsthan there would have been if female topexecutives were randomly distributed acrossoccupations.14 Once we look beyond these

Table 4. Relative Pay, Percent Female, andFemale/Male Wage Gaps by Broad Industry Categories.

High-Level Managers Managers from the CPS

(1) (2) (3) (4) (5) (6) (7) (8)% Industry Female/ % Industry Female/

Number Female Wage/ Male Number Female Wage/ Malein in Market Wage Gap in in Market Wage Gap

Industry Industry Industry Wage in Industry Industry Industry Wage in Industry

Agriculture 222 0.00 0.60 312 42.63 0.77 .75***Mining, Oil, Construction 3,197 1.38 0.93 0.47*** 4,338 18.21 1.01 .72***Food, Tobacco, Textile 6,361 2.63 0.99 0.81 3,953 31.77 1.08 .76***Chemical, Concrete, Autos 10,252 1.11 0.93 0.49*** 10,486 22.20 1.23 .76***Transport, Communication 6,112 2.45 0.77 0.49*** 6,495 31.25 1.10 .82***Wholesale / Retail Goods 5,484 3.61 0.81 0.65*** 10,671 40.72 0.76 .71***Banking 5,593 2.34 1.49 0.61*** 8,975 50.14 1.05 .69***Personal & Business Serv. 5,378 3.18 1.20 0.89 9,296 39.46 0.88 .77***Health and Social Services 4,109 3.87 0.95 0.62*** 18,958 60.07 0.99 .74***

Source: The data on high-level managers are from Standard and Poors ExecuComp database for 199297.The data on managers from the CPS are from the 1992 to 1997 Merged Outgoing Rotation Groups of the CurrentPopulation Survey.

***Significant at the 0.001 level or better.

12This is an example of what is known as verticalsegregation (see Blau, Ferber, and Winkler 1998).Also see Ferber and Loeb (1997) for a related ex-ample in higher education.

13Note, however, that in column 4 we report theratio of the average pay of women to the average payof men within occupations. For the CEO/Chaircategory, this ratio is positive and marginally signifi-cant (p-value 0.08). It appears that although very fewwomen made it to this top spot, once they got there,their average compensation (without consideringcontrol variables) was quite high.

14This finding is consistent with a vast prior litera-ture that has shown that a substantial part of thedifference in pay between men and women is attribut-able to the fact that women are less likely to hold thehigher-paying jobs. See, among others, Goldin (1990)and Blau and Ferber (1987). While sex segregationby occupation can be reconciled with some form oftaste discrimination by employers, employees, or cus-tomers (Becker 1957; Arrow 1973), many authorshave preferred to rely on human capital models toexplain this fact. See Lazear and Rosen (1990) forone such model. Another interpretation is offered byReskin and Ross (1990) and Strober (1984).


four top occupations, there is also an ap-parent negative correlation between thefraction of female executives in an occupa-tion and relative compensation in that oc-cupation, but the correlation is far fromstrong. For example, women were over-represented among CFOs whose compen-sation was relatively lower than we initiallyexpected, but CFOs were still paid morethan most of the categories of Vice Presi-dents.

In Table 6, we turn to a regression analy-sis in order to more precisely quantify theimpact of sex segregation by occupation onthe gender earnings gap. The sample ofexecutives for which we can construct occu-pation is about 9% smaller than the origi-nal sample. The unconditional gender gapin this sample is 47% (column 1 in Table 6)and is not statistically different from the44% gap found in Table 3 (which coversthe entire sample). The scarcity of femaleCEOs and Chairs only explains as much as13% of the compensation differentials (col-umn 2). Nearly half of the 47% gap can beexplained by the scarcity of women in thetop four occupations of Chair, CEO, ViceChair, and President (column 4). If onefurther controls for firm size (column 5),the gender compensation differential falls

to 12%. Interestingly, adding further occu-pational controls only very weakly reducesthe remaining gender compensation gap(columns 68 relative to column 4). Add-ing more than 60 detailed controls for bothfirst and second occupations in the job title(column 9) reduces the gender gap by an-other 7 percentage points compared tocolumn (4).

Finally, columns (10) and (11) of Table6 examine the combined effect of occupa-tional segregation, industrial segregation,and firm size. As noted above, industryindicators do not reduce the coefficient onthe female indicator at all (compare col-umn 8 to column 10). Controlling for firmsize after controlling for occupational cat-egories (column 11) still has a large effecton the female dummy. The magnitude ofthe effect, however, is smaller than in Table3. This very likely indicates that womenwere even less likely to hold the top jobswhen they worked for larger corporations.

By constructing occupational categoriesbased on the job title variable, we were alsoable to extract information on broad fieldof activity for a subsample of the observa-tions (see Appendix table). Womens rep-resentation was highest in fields such ashuman resources, utility services, and retail

Table 5. Relative Pay, Percent Female, andFemale/Male Compensation Gaps by Broad Occupation Groups.

(1) (2) (3) (4)Number % Occupation Female/Male

in Female in Wage/ Wage GapPosition Occupation Occupation Market Wage in Occupation

CEO / Chair 8,987 0.52 1.93 1.75*Vice Chair 2,000 0.85 1.53 0.50***President 5,840 1.71 1.30 0.58***CFO 326 6.44 0.61 0.67COO 164 1.83 1.16 0.61Other Chief Officer 2,155 1.58 1.48 0.47***Executive VP 8,581 2.66 0.83 1.10Senior VP 8,006 3.45 0.56 0.88**Group VP 493 0.81 0.44 0.91VP 7,468 4.27 0.37 0.79***Other Occupations 695 2.88 0.55 0.40***

Source: The data are from Standard and Poors ExecuComp database for 199297.***Means for men and women are significantly different at the 0.01 level; **at the 0.05 level; *at the 0.10

level.


Tab

le 6

. G

end

er P

ay G

ap f

or H

igh

-Lev

el E

xecu

tive

s w

hen

Det

aile

d M

anag

er O

ccu

pat

ion

Is

Con

sid

ered

.(D

epen

den

t V

aria

ble

is t

he

Log

of

Tot

al C

omp

ensa

tion

; Wh

ite-

Cor

rect

ed S

tan

dar

d E

rror

s in

Par

enth

eses

)

Inde

pend

ent V

aria

ble

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Fem

ale

0.4

7***

0.3

4***

0.3

1***

0.2

5***

0.1

2***

0.2

6***

0.2

2***

0.2

2***

0.1

8***

0.1

9***

0.1

1***

0.1

3***

(0.0

5)(0

.05)

(0.0

4)(0

.04)

(0.0

4)(0

.04)

(0.0

4)(0

.05)

(0.0

4)(0

.04)

(0.0

3)(0

.02)

CE

O /

Ch

air

0.82

***

0.87

***

0.99

***

0.89

***

0.99

***

1.24

***

(0.0

2)(0

.02)

(0.0

2)(0

.02)

(0.0

2)(0

.06)

Vic

e C

hai

r0.

77**

*0.

89**

*0.

51**

*0.

50**

*0.

50**

*(0

.04)

(0.0

4)(0

.03)

(0.1

2)(0

.12)

Pre

sid

ent

0.64

***

0.66

***

0.65

***

0.89

***

(0.0

2)(0

.02)

(0.0

2)(0

.06)

CFO

0.0

70.

17(0

.09)

(0.1

1)

CO

O0.

48**

*0.

73**

*(0

.12)

(0.1

3)

CO

0.39

***

0.63

***

(0.1

1)(0

.13)

EV

P0.

57**

*(0

.06)

SVP

0.26

***

( 0.0

6)

GV

P0.

21**

*(0

.07)

VP

0.1

2**

(0.0

6)

31 O

ccs.

no

no

no

no

no

no

no

yes

no

yes

yes

yes

31 +

35

Occ

s.n

on

on

on

on

on

on

on

oye

sn

on

on

o11

5 In

ds.

no

no

no

no

no

no

no

no

no

yes

yes

no

Firm

Eff

ects

no

no

no

no

no

no

no

no

no

no

no

yes

Lo

g V

alu

e0.

36**

*0.

35**

*0.

27**

*(0

.004

)( 0

.004

)( 0

.009

)

Ret

urn

/1k

0.06

**0.

010

.04

(0.0

2)( 0

.03)

( 0.0

3)

Co

nst

ant

6.50

***

6.29

***

6.22

***

6.09

***

3.63

***

6.09

***

5.83

***

6.64

***

6.90

***

6.63

***

4.33

***

4.95

***

(0.0

1)( 0

.01)

( 0.0

1)( 0

.01)

( 0.0

3)( 0

.01)

( 0.0

6)( 0

.09)

( 0.1

0)( 0

.09)

( 0.0

8)( 0

.07)

R2

0.02

90.

143

0.17

00.

217

0.49

40.

219

0.26

60.

273

0.28

10.

382

0.57

50.

710

N42

,677

42,6

7742

,677

42,6

7742

,677

42,6

7742

,677

42,6

7742

,677

42,6

7742

,677

42,6

77

Sou

rce :

Th

e d

ata

are

fro

m S

tan

dar

d a

nd

Po

or

s E

xecu

Co

mp

dat

abas

e fo

r 19

929

7.N

otes

: A

ll r

egre

ssio

ns

con

tro

l fo

r ti

me

ind

icat

or

vari

able

s. T

he

om

itte

d o

ccu

pat

ion

cat

ego

ry i

n t

he

fin

al c

olu

mn

s is

all

oth

er o

ccu

pat

ion

s.*S

tati

stic

ally

sig

nif

ican

t at

th

e .1

0 le

vel;

**a

t th

e .0

5 le

vel;

***

at t

he

.01

leve

l.


banking. Controlling for field in additionto occupation did not affect the coefficienton the female indicator variable. We donot report these results in the tables.

In column (12) of Table 6, we allow forfirm-specific effects in pay and add indi-vidual firm fixed effects to the regression.15The chi-squared value of the Hausman testof fixed-effects versus random effects ishighly significant (p-value less than 0.001)and indicates that inferences based on thefirm fixed effects specification in column(12) of Table 6 are most appropriate. Inany event, the coefficient estimate on fe-male when controlling for individual firmfixed effects (0.13) is nearly identical tothat in the previous specification with in-dustry fixed effects (0.11 in column 11).16

Oaxaca Decomposition

Another way to consider wage gaps be-tween groups is described in Oaxaca (1973).This method decomposes the overall gapinto a portion that is due to differences inobservable skills between groups and a partthat is still unexplained. This is easily doneby running separate regressions for menand women and then rewriting the overallwage gap in various ways as described be-low. First, define f and f (a vector) ascoefficient estimates from a regression oflog compensation on a constant and a set ofcovariates for women only and X

f (a vector)

as the mean characteristics of women. m,m, and X

m are similarly defined for men.The overall gap between men and women is

(1) w = m + mX

m f fX

f

There are two popular ways to re-write thisequation. The first is based on adding andsubtracting mX

f, which yields

(2) w = (m f) +(m f)X

f + m (X

m X

f).

In this case we are assuming that the re-turns to male characteristics, m, are thebaseline. The second common decomposi-tion is found by adding and subtractingfX

m to equation (1), which yields

(3) w = (m f) +(m f)X

m + f(X

m X

f).

In this case we are assuming that the re-turns to female characteristics, f , are thebaseline.17 In both equations (2) and (3),the first two terms are the part of the totalgap left unexplained and the third term isthe part of the gap due to explained differ-ences in skills.

We present results for a simple Oaxaca(1973) decomposition in Table 7. In thiscase, we use the covariates used in column(5) of Table 6: year indicators, indicatorsfor the top three occupations, log stockmarket value, and stock return in the previ-ous year. We chose this parsimonious speci-fication because, as indicated by Table 6,these covariates alone account for nearlyall of the explained variation in compensa-tion. As stated above, we decompose thetotal gap assuming that the male wage struc-ture is the true wage structure (as in equa-tion 2) and then assuming that the femalewage structure is the true wage structure(as in equation 3). The results in Table 7confirm our previous findings. Most of thetotal gap in compensation by gender forthese top managers (between 71% = 0.30/0.42 and 88% = 0.37/0.42) was due to ob-servable differences between men andwomen.18

The Role of Age and Tenure

A major drawback of the ExecuCompdata set is that it does not report age andtenure consistently for all observations.

15In this case, we cannot also control for industry,since industry does not vary within firms (for the mostpart).

16We also re-computed this specification with anindividual person random effect model. In this case,the coefficient on female is nearly identical, 0.12.

17Of course, these are just extreme cases, and anycombination of m and f could also be a possibility(see Ransom and Oaxaca 1994).

18Further details of the decomposition, includingthe separate regressions by gender, are available fromthe authors on request.


Table 7. Basic Oaxaca Decomposition.

Unexplained Gap Due toDecomposition Total Gap Gap Skill Differences

Oaxaca Decomposition #1 (returns to male are baseline) 0.42 0.12 0.30

Oaxaca Decomposition #2 (returns to female are baseline) 0.42 0.05 0.37

Source: The data are from Standard and Poors ExecuComp database for 199297.Note: Separate regressions are run for men and women (see text). All regressions control for time indicator

variables, status as CEO, Chair, Vice Chair, President, log(market value), and shareholder return.

These two variables are available for only asubset of the observations in the sample.19Focusing on that subsample of the data,Panel B of Table 2 shows that women inthese top managerial jobs were very similarto men with respect to their labor forceattachment and career commitment, butdiffered considerably from their male coun-terparts with respect to age and seniority intheir corporations. Women in thesubsample for which age and tenure areavailable were about 5 years younger thanthe men, on average (47.5 versus 52.6 yearsold), and had 5.6 fewer years of seniority intheir company (7.7 versus 13.3 years). (It isinteresting to note that the gaps in age andtenure are about the same.) Because re-turns to age and experience are large in themarket for executives, we expect that therelative youth and low seniority of the fe-male executives is another important de-terminant of the gender gap. This is for-mally shown in Table 8.

Because the sample used in this sectionis much smaller, we re-estimate the uncon-ditional gender compensation gap for thisgroup. As seen in column (1) of Table 8,the point estimate on the female indicator,0.61, is substantially larger than in the

previous larger samples. Yet, standard er-rors are large. Once we control for firmsize (as in column 2), the remaining gen-der gap is again much smaller. If we furtheradd the three top occupation dummies (col-umn 3), the gap falls to 8%. However,standard errors are again large, and wecannot reject the possibility that the coeffi-cient on the female indicator variable iseither 0 or the same as the female dummyin the larger sample for the same set ofcontrols (column 5 of Table 6). If wecontrol for occupation effects (column 4),the point estimate for the female dummydrops to 0.05, and if we control for bothoccupation and industry effects (column5), it drops to 0.09. Again, because stan-dard errors are large, we cannot reject thepossibility that the coefficients on the fe-male dummies are the same as the corre-sponding ones in Table 6.

The womens relative youth cannot initself fully explain the gender gap. Column(6) of Table 8 shows that a 33% differencein compensation still exists between menand women after we account for age andseniority. However, this is not preciselyestimated. There is also a clear but imper-fect correlation between executives ageand the size of the companies they man-aged. When age and tenure are included(last five columns of Table 8), the additionof a firm size control does not improve theR2 as much, nor does it decrease (in abso-lute value) the coefficient on the femaledummy as much (column 7 versus column2). Whether or not we control for age andseniority, adding the three top occupationdummies (column 8) leads to about thesame improvement in R2 as in column (3),

19We are not aware of any reason why age andtenure are only reported for a subset of the data. Weinvestigated how individual and firm characteristicsdiffer between our basic sample and the sample inwhich age and tenure are available. We found thatindividuals in the subsample were slightly more likelyto be female (2.4% of the overall sample was female,compared to 2.6% of the subsample) and worked forsmaller firms.


Tab

le 8

. G

end

er P

ay G

ap f

or H

igh

-Lev

el M

anag

eria

l P

ay w

hen

Age

an

d T

enu

re o

f M

anag

er A

re C

onsi

der

ed.

(Dep

end

ent

Var

iabl

e is

th

e L

og o

f T

otal

Com

pen

sati

on; W

hit

e-C

orre

cted

Sta

nd

ard

Err

ors

in P

aren

thes

es)

Not

Con

trol

lin

g fo

r A

ge a

nd T

enur

eC

ontr

olli

ng f

or A

ge a

nd T

enur

e

Inde

pend

ent

Var

iabl

e(1

)(2

)(3

)(4

)(5

)(6

)(7

)(8

)(9

)(1

0)

Fem

ale

0.6

1***

0.2

5***

0.0

80

.05

0.0

90

.33

0.1

40

.04

0.0

060

.05

(0.2

0)(0

.17)

(0.1

5)(0

.14)

(0.1

3)(0

.20)

(0.1

7)(0

.14)

(0.1

4)(0

.13)

Age

0.20

***

0.08

***

0.06

***

0.06

***

0.08

***

(0.0

3)(0

.02)

(0.0

2)(0

.02)

(0.0

2)

Age

20

.002

***

0.0

06**

*0

.001

***

0.0

01**

*0

.000

7***

(0.0

003)

(0.0

002)

(0.0

002)

(0.0

002)

(0.0

002)

Ten

ure

0.01

7**

0.0

050

.02*

**0

.019

***

0.0

22**

*(0

.007

)(0

.005

)(0

.005

)(0

.005

)(0

.004

)

Ten

ure

20

.000

10.

0000

0.00

02*

0.00

02*

0.00

03**

*(0

.000

2)(0

.000

1)(0

.000

1)(0

.000

1)(0

.000

1)

Lo

g V

alu

e0.

45**

*0.

39**

*0.

38**

*0.

38**

*0.

424*

**0.

40**

*0.

389*

**0.

390*

**(0

.01)

(0.0

1)(0

.01)

(0.0

1)(0

.01)

(0.0

1)(0

.011

)(0

.013

)

Ret

urn

/10

000

.02

0.03

0.04

0.0

50

.01

0.02

0.02

0.0

6(0

.04)

(0.0

3)(0

.03)

(0.0

3)(0

.03)

(0.0

3)(0

.03)

(0.0

3)

CE

O /

Ch

air

0.90

***

0.92

***

(0.0

4)( 0

.04)

Vic

e C

hai

r0.

43**

*0.

44**

*(0

.08)

( 0.0

8)

Pre

sid

ent

0.66

***

0.66

***

(0.0

4)( 0

.04)

31 O

ccs.

no

no

no

yes

yes

no

no

no

yes

yes

115

Ind

s.n

on

on

on

oye

sn

on

on

on

oye

s

Co

nst

ant

6.92

***

3.69

***

3.49

***

4.26

***

4.25

***

0.75

1.26

***

1.86

***

2.70

***

2.18

***

(0.0

4)( 0

.08)

( 0.0

7)( 0

.12)

( 0.1

3)( 0

.68)

( 0.4

8)( 0

.44)

( 0.4

5)( 0

.42)

R2

0.01

60.

427

0.53

30.

544

0.60

90.

136

0.44

90.

544

0.55

60.

612

N7,

379

7,37

97,

379

7,37

97,

379

7,37

97,

379

7,37

97,

379

7,37

9

Sou

rce:

Th

e d

ata

are

fro

m S

tan

dar

d a

nd

Po

or

s E

xecu

Co

mp

dat

abas

e fo

r 19

929

7.N

otes

: A

ll r

egre

ssio

ns

con

tro

l fo

r ti

me

ind

icat

or

vari

able

s.*S

tati

stic

ally

sig

nif

ican

t at

th

e .1

0 le

vel;

**a

t th

e .0

5 le

vel;

***

at t

he

.01

leve

l.


where we do not control for age and tenureat all. The female dummy decreases (inabsolute value), from 0.14 to 0.04. Thesefindings, while imprecise, indicate that thegender compensation gap could be lessthan 5% after all observables are controlledfor.20

Trends in Participation and Earnings:Is the Glass Ceiling Cracking?

One of the major labor market trends inthe United States in the past two decadeshas been the convergence in outcomes be-tween men and women. Focusing on differ-ences in earnings, Blau and Kahn (1997)showed that womens relative position con-siderably improved, especially during the1980s, when men experienced a real de-cline in earnings while female real wagesgrew very rapidly. They showed that part ofthis shrinking gender pay gap could beexplained by an improvement in femalehuman capital, especially in the form oflabor market experience, and by a smallerunexplained gender gap, that could re-flect either a reduction in labor marketdiscrimination or an improvement inwomens unmeasured characteristics. Yet,another important factor in explaining thedecline in the gender gap has been animportant shift in occupational categoriesfor women. More specifically, the repre-sentation of women in managerial and pro-fessional jobs has been growing while theshare of women in low-paying clerical andrelated jobs has not.

In this section, we address the questionof whether these trends also exist amongtop executives. In other words, we ask

whether there is any evidence that the glassceiling is cracking little by little in U.S.corporations. We ask whether the relativeparticipation of women in these top mana-gerial jobs has increased over time, andalso study trends in relative compensation.It is important to note that because ourdata set only covers the period 199297, weare unable to investigate relative gains inthe 1980s, the period during which most ofthe catch-up by women occurred, at least inthe other segments of the labor market.

Table 9 reports trends over time in thefraction of women in top-level management.While the fraction of women in lower-levelmanagement only went from 40% to 43%over the sample period (column 9), thefraction of women in top-level managementnearly tripled, going from 1.29% in 1992 to3.39% (column 1) in 1997.21 The fractionof firms with at least one woman in the topexecutive ranks (one of the top 5 mosthighly paid) grew from 5.4% in 1992 to15.03% in 1997. Although the fraction offirms with strictly more than one woman inthese top positions was much smaller, italso grew a great deal over the period, from0.17% in 1992 to 1.95% in 1997.

Also note that the fraction of firms withno women in one year and at least onewoman in the next year grew steadily overtime, from 2.19% in 1993 to 3.85% in 1997(column 4). Female top executives alsoseem to have improved their relative earn-ings quite substantially. While the ratio ofaverage female to average male compensa-tion in 1992 is a puzzle to us, it appears thatfemale relative compensation rose ex-tremely quickly between 1993 (52%) and1997 (73%).22 During these same years, wefind that the ratio of average female toaverage male annual earnings in our CPS

20If we examine relatively new entrants to the topmanagerial jobsthat is, men and women with onlyfive or fewer years of labor market experiencewefind that the conditional wage gap is also statisticallyinsignificant and the point estimate is not muchdifferent from our estimate on the complete set ofdata. If we estimate a specification like that in Table6, column (11), for the 5,113 executives with 5 yearsof experience or less, the coefficient on the femaleindicator is 0.093 with a standard error of 0.134.

21See Catalyst (1999) for descriptive evidence thatis consistent with this finding.

22Because we were concerned by possible changesin the set of companies covered by ExecuComp overtime, we investigated the robustness of these findingsby limiting the sample to the companies that werepresent in 1992. The findings were unaffected.


Tab

le 9

. I

nfo

rmat

ion

ove

r T

ime

(199

219

97)

for

Tw

o Se

ts o

f M

anag

ers.

(Wh

ite-

Cor

rect

ed S

tan

dar

d E

rror

s in

Par

enth

eses

)

Hig

h-L

evel

Man

ager

s fr

om E

xecu

Com

p (c

olum

ns 1

8)

CPS

Sam

ple

(col

umns

91

1)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

Frac

. Fir

ms

with

No

Wom

enFr

ac.

Frac

.in

One

F/M

F/M

F/M

Firm

sFi

rms

Yea

r an

dC

ompe

n-R

egre

ssio

nF/

MSh

are

Jobs

Com

pen-

Reg

ress

ion

Frac

tion

with

1

with

> 1

The

n

1 in

satio

nA

djus

ted

Firm

in

Top

Fr

actio

nsa

tion

Adj

uste

dYe

arFe

mal

eW

oman

Wom

anN

ext

Gap

aD

iffer

entia

lbSi

ze G

apG

roup

cFe

mal

eG

apd

Diff

eren

tiale

1992

1.29

5.44

0.17

0.

67*

0.2

21**

0.42

***

0.64

***

40.4

10.

67**

*0

.271

***

(0.0

91)

(0.0

09)

1993

1.52

7.25

0.39

2.19

0.52

***

0.1

37**

*0.

52**

*0.

48**

*40

.25

0.69

***

0.2

54**

*(0

.053

)(0

.009

)

1994

2.22

10.2

51.

133.

440.

56**

*0

.163

***

0.52

***

0.36

***

41.7

90.

68**

*0

.247

***

(0.0

53)

(0.0

09)

1995

2.60

11.7

61.

583.

660.

57**

*0

.144

***

0.56

***

0.32

***

41.2

00.

68**

*0

.260

***

(0.0

43)

(0.0

09)

1996

3.01

13.4

21.

723.

920.

61**

*0

.085

**0.

61**

*0.

40**

*42

.07

0.70

***

0.2

49**

*( 0

.043

)( 0

.009

)

1997

3.39

15.0

31.

953.

840.

73**

*0

.013

0.77

***

0.37

***

42.8

60.

72**

*0

.225

***

(0.0

49)

( 0.0

09)

Sou

rce:

Th

e d

ata

are

fro

m S

tan

dar

d a

nd

Po

or

s E

xecu

Co

mp

dat

abas

e fo

r 19

929

7.a *

**D

iffe

ren

ce i

n a

vera

ge f

emal

e/av

erag

e m

ale

com

pen

sati

on

by

year

is

sign

ific

ant

at t

he

0.01

lev

el; *

*at

the

0.05

lev

el; *

at t

he

0.10

lev

el.

bT

his

is th

e co

effi

cien

t on

the

fem

ale

ind

icat

or

fro

m a

n a

nn

ual

reg

ress

ion

of l

n (

pay

) o

n a

n in

dic

ato

r fo

r fe

mal

e, p

lus

the

oth

er c

ova

riat

es in

clu

ded

in T

able

6, c

olu

mn

(11

) . W

hit

e-co

rrec

ted

sta

nd

ard

err

ors

are

rep

ort

ed.

c By

to

p g

rou

p

we

mea

n t

ho

se jo

bs

clas

sifi

ed a

s C

EO

/ch

air,

vic

e ch

air,

or

pre

sid

ent.

dA

dju

stin

g fo

r h

ou

rs.

***D

iffe

ren

ce in

ave

rage

fem

ale/

aver

age

mal

e co

mp

ensa

tio

n b

y ye

ar is

sig

nif

ican

t at

th

e 0.

01 le

vel;

**a

t th

e 0.

05 le

vel;

*at

th

e 0.

10le

vel.

e Th

is is

the

coef

fici

ent o

n th

e fe

mal

e in

dic

ato

r fr

om

an

an

nu

al r

egre

ssio

n o

n ln

(p

ay)

on

an

ind

icat

or

for

fem

ale,

exp

erie

nce

, exp

erie

nce

-sq

uar

ed, a

lin

ear

term

fo

r ed

uca

tio

n, a

n i

nd

icat

or

for

bla

ck, a

nd

on

e d

igit

in

du

stry

in

dic

ato

rs.


sample went from 67% to 72%. In addi-tion, in column (6) we have reported coef-ficients on the female indicator from ourspecification in column (11) of Table 6 byrunning separate regressions by year. Thesubstantial and statistically significant esti-mate of 0.221 from 1992 declined throughthe sample period to a statistically insignifi-cant 0.013 in 1997.

Note, however, that the regressions-ad-justed gap for managers in the CPS changedmuch less over time (column 11). Whatcaused such a rapid decline in the gendergap for female managers at the top? In theprevious section, we isolated two main fac-tors that strongly hampered womens rela-tive earnings: under-representation in largefirms, and under-representation in the topfour occupations. Were female top execu-tives in 1997 doing better on either of thesetwo fronts? Column (7) of Table 9, whichdisplays the ratio of average female manag-ers company size to average male manag-ers company size, clearly indicates thatfemale executives were steadily gaining ac-cess to larger U.S. corporations.23 On theother hand, column (8) shows no evidencethat womens representation in the topoccupational group (CEOs, Chairs, ViceChairs, and Presidents) was improving overtime. If anything, womens representationin this group was declining, though notsteadily.

Summary and Concluding Remarks

We have shown that, contrary to whatsome commentators have claimed, the glassceiling is somewhat porous and somewomen, even if only a limited number, areinvolved in the top-level management ofU.S. corporations. Over all years in thesample we examined, about 2.4% of theexecutives were women.24 Although this

number is small, it increased substantiallyin the later years of the sample period.

Our results further indicate that the gen-der gap in compensation among top execu-tives was at least 45%. An important fact isthat female managers were under-repre-sented in large corporations. Because thereturns to firm size are very high among topexecutives, this explains up to 15 percent-age points of the gender gap. Interestingly,while female managers do not seem to havebeen distributed randomly across indus-tries, there is no evidence that sex segrega-tion by industry explains any of the ob-served gender gap. On the other hand, wefound that sex segregation by occupationwas important. The scarcity of female CEOs,Chairs, Vice-Chairs, and Presidents accountsfor as much as half of the unconditionalgender compensation gap. Once we lookbeyond these four top occupations, how-ever, there is no significant evidence of aconcentration of women in the lower-com-pensation occupations.

A last crucial factor is that women in thesample were much newer than their malecounterparts in this top stratum of manage-rial jobs. Women in the sample were muchyounger, and had much less seniority intheir company, than men. Part of the effectof age and seniority on the gender gapseems to be reflected in the size of compa-nies women managed. All in all, we findthat the unexplained gender compensa-tion gap for top executives was less than 5%after one accounts for all observable differ-ences between men and women.

Finally, we asked whether there is anyevidence of a growing crack in the glassceiling over the period under study. Wefound that the participation of women inthe top corporate jobs was growing dra-matically in the years we looked at, from1.3% in 1992 to 3.4% in 1997. Also, thegender compensation gap declined, verymuch like in the other segments of thelabor market. Most of the decline appearsto have been correlated with a decline insex segregation by firm size. Female topexecutives were heading larger and largercorporations.

Because top executives probably consti-

23This could be due to a lessening of employers orcustomers tastes for discrimination or to changingtastes on the part of the female executives themselves.

24This is still a small fraction relative to otherforms of organization. For example, Hallock (1998)found that about 20% of public charities are headedby women.


tute a fairly homogeneous group with re-spect to job motivation, career commit-ments, and human capital, a finding of anunexplained gender compensation gap inthis sample could have reasonably beeninterpreted as evidence for taste discrimi-nation against women. In fact, we find thatthe conditional gender gap in this sample isvery small. This obviously does not implythe absence of discrimination. Low gen-eral participation, sex segregation by firmsize, and sex segregation by occupationcould reflect some form of taste discrimina-tion. The absence of a significant condi-tional gender gap simply means that womenand men who held similar functions infirms of similar size received fairly equaltreatment in terms of compensation.

Additional caveats should be stated here.First, the latter results should not be gener-alized to a claim that all female executivesin the United States are paid like their malecounterparts in the same occupation cat-egory and in firms of the same size. Inves-tigating that issue would require a differentdata set that, to our knowledge, does notexist. Second, one might argue that thevery few women who made it into our sampleare truly exceptional and should not in fact

be compared to the average man in thesample but rather to the highest-ability menin the sample.25 Under that view, one mighthave expected that these women shouldhave been paid more than the average maleexecutives. The data clearly reject a posi-tive female-male gender gap in earnings.

Future work might involve a more for-mal analysis of why some companies decideto promote women to top jobs while othersdo not.26 For example, one might inquirewhether various characteristics of the boardof directors, such as the sex and age distri-bution of their members, are correlatedwith the selection of women into top execu-tive positions. More fundamentally, onemight try to understand what factors makesmall companies more likely to attract fe-male top executives and why women arevirtually absent from the very top of theU.S. corporate world.

25Our analysis has assumed that men and womenin our sample form a homogeneous group and thatthe distribution of unobservables is similar acrossgenders.

26One might also investigate gender differences inmobility.


Appendix

Relative Pay, Percent Female, and Female/Male Wage Gaps by Broad Field Groups

High-Level Managers Only

(1) (2) (3) (4)% Field Wage/ Female/Male

Number in Female Market Wage GapField Field in Field Wage in Field

Human Resources 343 14.86 0.89 0.60***Finance/Accounting 1,682 2.02 1.01 1.41Legal/Regulatory Affairs 258 10.85 1.28 0.51***Sales 683 2.05 0.99 0.77Marketing/Merchandising, Advertising 721 6.52 1.00 0.89Product Devel.: R&D, Engin., Design 702 3.99 1.02 1.52U.S. Operations 1,132 0.88 0.90 1.48International Operations 332 1.20 1.25 1.81Corporate Affairs 907 4.96 1.05 1.63Customer Service 120 12.5 0.89 1.36Product Management/ Manufacturing 268 4.10 0.96 0.78Real Estate/Construction 66 4.55 0.81 0.62Utility Services 26 30.77 0.34 1.33Retail Banking, Credit 84 16.67 0.74 0.98Sourcing, Procurement 21 9.52 1.03 1.23Administration 299 5.69 0.88 0.46***Communication, Information 89 6.74 0.70 0.78Healthcare 59 1.69 1.36 0.53a

Mergers & Acquisitions 50 0.00 0.91 NAa

Computers 38 2.63 1.40 0.93a

Developed Markets 7 42.86 2.84 1.78Investment 35 2.86 1.99 0.13a

Distribution 30 3.33 0.70 0.48a

Missing 38,756 2.04 2.27 0.67***

aOnly one woman (or zero in the case of mergers) in these casesno t test for differences in means.


REFERENCES

Arrow, Kenneth J. 1973. The Theory of Discrimina-tion. In Orley Ashenfelter and Albert Rees, eds.,Discrimination in Labor Markets. Princeton: PrincetonUniversity Press, pp. 333.

Baker, Lawrence. 1996. Differences in Earningsbetween Male and Female Physicians. New EnglandJournal of Medicine, Vol. 334, pp. 96064.

Barbezat, Debra. 1987. Salary Differentials by Sex inthe Academic Labor Market. Journal of HumanResources, Vol. 22, No. 3 (Summer), pp. 42228.

Barbezat, Debra A., and James W. Hughes. 1990.Sex Discrimination in Labor Markets: The Role ofStatistical Evidence. Comment. American EconomicReview, Vol. 80, No. 1 (March), pp. 27786.

Bayard, Kimberly, Judith Hellerstein, David Neumark,and Kenneth Troske. 1999. New Evidence on SexSegregation and Sex Differences in Wages fromMatched Employee-Employer Data. Working pa-per, March.

Becker, Gary S. 1957. The Economics of Discrimination.Chicago: University of Chicago Press.

Biddle, Jeff, and Daniel S. Hamermesh. 1998. Beauty,Productivity, and Discrimination: Lawyers Looksand Lucre. Journal of Labor Economics, Vol. 16, No.1 (January), pp. 172201.

Bird, Chloe. 1990. High Finance, Small Change:Womens Increased Representation in Bank Man-agement. In Barbara Reskin and Patricia Roos,eds., Job Queues, Gender Queues. Philadelphia: TempleUniversity Press.

Blau, Francine, and Marianne Ferber. 1987. Dis-crimination: Empirical Evidence from the UnitedStates. American Economic Review, Vol. 77, No. 2, pp.31620.

Blau, Francine, Marianne Ferber, and Anne Winkler.1998. The Economics of Women, Men, and Work, 3rdedition. Englewood Cliffs, N.J.: Prentice Hall.

Blau, Francine, and Lawrence Kahn. 1997. Swim-ming Upstream: Trends in the Gender Wage Differ-ential in the 1980s. Journal of Labor Economics, Vol.15, No. 1 (January), pp. 142.

Catalyst. 1999. Catalyst Census of Women CorporateOfficers and Top Earners. New York.

Ferber, Marianne A, and Carole A. Greene. 1982.Traditional or Reverse Sex Discrimination? TheCase of a Large Public University. Industrial andLabor Relations Review, Vol. 35, No. 4 (July), pp. 55064.

Ferber, Marianne, and Lane Loeb. 1997, AcademicCouples: Problems and Promise. Urbana: University ofIllinois Press.

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

Gander, James P. 1997. Gender-Based Faculty-PayDifferences in Academe: A Reduced-Form Ap-proach. Journal of Labor Research, Vol. 18, No. 3(Summer), pp. 45161.

Gregg, Paul, and Stephen Machin. 1993. Is the GlassCeiling Cracking? Gender Compensation Differen-tials and Access to Promotion among UK Execu-

tives. University College London Discussion Paper9405, July.

Groshen, Erica. 1991. The Structure of the Fe-male/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.

Hall, Brian, and Jeffrey Liebman. 1998. Are CEOsReally Paid Like Bureaucrats? Quarterly Journal ofEconomics, Vol. 113, No. 3 (August), pp. 65391.

Hallock, Kevin F. 1998. Gender CompensationDifferences among Executives in Nonprofits. Work-ing paper, University of Illinois, August.

Hoffman, Emily. 1976. Faculty Salaries: Is ThereDiscrimination by Sex, Race, and Discipline? Ameri-can Economic Review, Vol. 66, No. 1 (March), pp.19698.

Jacobs, Jerry. 1992. Womens Entry into Manage-ment: Trends in Earnings, Authority, and Valuesamong Salaried Managers. Administrative ScienceQuarterly, Vol. 37, No. 2 (June), pp. 282301.

Johnson, George, and Frank Stafford. 1974. TheEarnings and Promotion of Women Faculty. Ameri-can Economic Review, Vol. 64, No. 6 (December), pp.888903.

Jones, Terril Yue. 1999. Gender Motors. Forbes,May 17, p. 50.

Katz, David. 1973. Faculty Salaries, Promotions, andProductivity at a Large University. American Eco-nomic Review, Vol. 63, No. 3 (June), pp. 46977.

Kostiuk, Peter. 1990. Firm Size and ExecutiveCompensation. Journal of Human Resources, Vol. 25,No. 1 (Winter), pp. 90105.

Lazear, Edward, and Sherwin Rosen. 1990. Male-Female Wage Differentials in Job Ladders. Journalof Labor Economics, Vol. 8, No. 1, pp. 10623.

Meyer, Michael. 1999. In a League of Her Own.Newsweek, August 2, p. 56.

Morgan, Laurie. 1998. Glass-Ceiling Effect or Co-hort Effect? A Longitudinal Study of the GenderEarnings Gap for Engineers, 1982 to 1989. Ameri-can Sociological Review, Vol. 63, No. 4 (August), 47993.

Morris, Kathleen. 1998. The Rise of Jill Barad.Business Week, May 25, pp. 11218.

Murphy, Kevin J. 1985. Corporate Performance andManagerial Remuneration. Journal of Accountingand Economics, Vol. 7, Nos. 13, pp. 1142.

Oaxaca, Ronald. 1973. Male-Female Wage Differen-tials in Urban Labor Markets. International Eco-nomic Review, Vol. 14, No. 3 (October), pp. 693709.

Ransom, Michael R., and Sharon Bernstein Megdal.1993. Sex Differences in the Academic LaborMarket in the Affirmative Action Era. Economics ofEducation Review, Vol. 12, No. 1 (March), pp. 2143.

Ransom, Michael, and Ronald Oaxaca. 1994. OnDiscrimination and the Decomposition of WageDifferentials. Journal of Econometrics, Vol. 61(March), pp. 551.

Reskin, Barbara F., and Particia A. Roos, eds. 1990.Job Queues, Gender Queues: Explaining Womens In-


roads into Male Occupations. Philadelphia: TempleUniversity Press.

Rosen, Sherwin. 1992. Contracts and the Market forExecutives. In Lars Wein and Hans Wijkander,eds., Contract Economics. London: Blackwell.

Strober, Myra H. 1984. Toward a General Theory ofOccupational Segregation: The Case of PublicSchool Teaching. In Barbara F. Reskin, ed., Segre-

gation in the Workplace: Trends, Explanations, Rem-edies. Washington, D.C.: National Academy Press,pp. 14456.

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.

Cornell University ILR SchoolDigitalCommons@ILR10-1-2001

The Gender Gap in Top Corporate JobsMarianne BertrandKevin F. HallockThe Gender Gap in Top Corporate JobsAbstractKeywordsDisciplinesComments

bertrand, m. and k. f. hallock (2000). -the gender gap in top corporate jobs.-

Documents

unexplained gap

tthe gender gap

gender segregation

younger average age

corporate jobsabstractusing

female executives

lower average seniority

possibility of discrimination