Separate and Unequal: Occupation-Establishment Sex Segregation and the Gender Wage Gap

by Trond Petersen, Laurie A. Morgan
Separate and Unequal: Occupation-Establishment Sex Segregation and the Gender Wage Gap
Trond Petersen, Laurie A. Morgan
The American Journal of Sociology
Start Page: 
End Page: 
Select license: 
Select License

Separate and Unequal: Occupation-Establishment Sex Segregation and the Gender Wage ~ a ~ '

Trond Petersen

University of California, Berkeley

Laurie A. Morgan University of Illinois at Urbana-Champaign

The authors report the first large-scale empirical investigation of within-job wage differences between men and women in the same occupation and establishment, using data first on blue-collar and clerical employees from 16 U.S. industries in 1974-83 and second on employees in 10 professional and administrative occupa- tions. The authors report three findings. First, wage differences at the occupation-establishment level were small even without con- trols for individual-level characteristics. Hence, within-job wage discrimination was much less important than occupation-establishment segregation for observed wage differences. Second, establishment segregation was an important cause, although not as important as occupational segregation, of wage differences. Third, establishment segregation was extensive, as was occupational segre- gation.


Wage differences between men and women caused by discrimination can result from several processes. The first is where women are differentially allocated to occupations and establishments that pay lower wages. This

' We are grateful to Marta Elvira, Kenneth Koput, Katrine Teigen, and Vemund Snartland for research assistance. We also thank Erling Barth, Robert Erikson, Jan Hoem, Geir HZgsnes, Carl LeGrand, David Levine, Karin Martin, Arne Mastekaasa, Eva Meyerson, Charles O'Reilly, Seymour Spilerman, Donald Tomaskovic-Devey, and the AJS reviewers for useful comments and discussions. We also thank Erica Groshen who provided data from several of the 16 Industry Wage Surveys analyzed here. The research was supported by the Institute of Industrial Relations at the University of California, Berkeley, and by National Science Foundation grant SES-8912502. Direct correspondence to Trond Petersen, Walter A. Haas School of Business, 350 Barrows Hall, University of California, Berkeley, California 94720.

O 1995 by The University of Chicago. All rights reserved. 0002-9602196110102-0003$01.50

AJS Volume 101 Number 2 (September 1995): 329-65 329 process may involve discrimination partly through differential access to occupations and establishments, that is, the matching process at the point of hire, and partly through subsequent promotion^.^ We call this process "allocative discrimination." The second is where occupations held pri- marily by women are paid lower wages than those held primarily by men, although skill requirements and other wage-relevant factors are the same. This is the issue addressed by comparable worth. We call this process "valuative discrimination." The third is where women receive lower wages than men within a given occupation within a given establish- ment. We call this process "within-job wage di~crimination."~

Allocative and valuative discrimination involve the segregation of men and women into different occupations, establishments, or both and may occur with- out within-job wage discrimination. Thus, it may be the case that where men and women share the same jobs they receive the same pay but that in most cases they simply do not share the same jobs.

One conjecture currently accepted by many researchers is that wage differences are less an issue of within-job wage discrimination and more a matter of allocative and valuative processes. That is, the segregation of women into lower-paying occupations, establishments, or both and lower pay in occupations held primarily by women are more important than pay differences within the same job in explaining the gender wage gap. Treiman and Hartmann (1981, pp. 92-93) write, "Although the committee recognizes that instances of unequal pay for the same work have not been entirely eliminated, we believe that they are probably not now the major source of differences in earnings."

This conjecture is drawn primarily from a large literature that focuses on pay differences across and within occupations. One pattern of findings is that the wage gap between men and women becomes smaller as occupa- tional controls become finer (Treiman and Hartmann 1981; Marini 1989),

We follow the convention that discrimination occurs when wage differences between men and women are not accounted for by average differences in productive attributes, which is not to say that the attainment of productive attributes, e.g. education, is itself not related to discrimination (e.g., American Association of University Women 1992).

Treiman and Hartmann (1981, pp. 8-9) refer to allocative discrimination as employ- ment discrimination, and valuative and within-job wage discrimination as wage dis- crimination. Both allocative and within-job wage discrimination are illegal. The for- mer is covered by Title VII of the 1964 Civil Rights Act, while the latter is covered by the Equal Pay Act of 1963 (see England 1992, chap. 5). Valuative discrimination is discrimination against classes of jobs occupied primarily by women but not discrimi- nation against any specific individual. Its legal status is unclear, but the current legal situation can be summarized as one in which the courts do not interpret Title VII to require comparable worth unless the plaintiff can show that a job was intentionally paid less because the incumbents are women (see England 1992, chap. 5).

suggesting that a large proportion of the wage gap is explained by occupa- tional di~tribution.~

For example, Treiman and Hartmann (1981, pp. 33-

39) explained 10%-20% of the raw wage gap using 222 occupational cate-

gories and 35%-40% using 479 categories. These studies usually draw on

data from the census or national probability samples that allow no analysis

of practices in specific establishments. Additional evidence suggests that,

within occupations, the distribution of women across firms or establish-

ments also accounts for some portion of the wage gap. For example, Blau

(1977) found that in 11 clerical occupations, differences in men's and

women's wages were larger between than within establishments.

Yet the prevailing conjecture remains a conjecture. It has not been shown that men and women receive equal pay within given occupations in given establishments. What has been shown is that sex segregation is extensive and pervasive (Bielby and Baron 1984), but not the extent to which sex segregation accounts for the wage gap or that, when sex segre- gation is absent, the sexes receive equal treatment. To confirm such a claim, one needs data on wages of men and women in the same detailed occupational group or position within the same establi~hment.~

Such data

are not widely available except on isolated establishments.

This article reports the first relatively large-scale empirical investiga- tion of wage differences between men and women within the same de- tailed occupational position within the same establishment. We use estab- lishment-level data from a wide variety of industries. In each establishment, individual-level wage data for a large array of detailed occupational groups were collected, providing more accurate wage as well as occupational data than probably any other surveys available (ex- cept in some case studies of single establishments, e.g., Hartmann [1987]). We focus first on production and clerical employees in 16 U.S. industries in the 1974-83 period, primarily 1974-78, analyzing data on about 870,000 employees, 700 industry-specific occupations, 6,000 estab- lishments, and 71,000 occupation-establishment pairs, where each occu- pation within an establishment is an occupation-establishment pair. Sec- ond, we focus on seven professional and three administrative occupations across a broad range of industries in 1981, analyzing data on about

A recent and extensive set of occupational case studies can be found in Reskin and Roos (1990), mostly using data from the 1980 U.S. census (see, e.g., U.S. Bureau of Census 1984). When three-digit occupational groups are studied and there are few or no controls for other characteristics, women's annual earnings are about 30%-40% below men's in most of the 11 occupations studied.

Treiman and Hartmann (1981, p. 33) write on the standard analyses of wage differ- ences: "This exercise illustrates that further analysis of occupational segregation re- quires much more detailed data than are currently available from the census or from national sample surveys."

740,000 employees distributed across 2,162 establishments and 16,433

occupation-establishment pairs.

Apart from Blau's (1977) study of 11 clerical occupations, the studies that most resemble our design are Groshen (1991) and Tomaskovic-Devey (1993). Unlike the present study, neither computes the amount of within- job wage differences between men and women (i.e., in the same occupa- tion and establishment). Tomaskovic-Devey uses a random sample of employees in North Carolina, and so is unable to make this calculation, because he has no data on men and women working in the same jobs. However, he does have information on the sex composition of each re- spondent's job, which he includes as an independent variable in a regression analysis. Groshen (1991) uses data on six of the 16 industries ana- lyzed here, but she does not compute within-job wage differentials, performing instead the same type of analysis as Tomaskovic-Devey. Both authors report that the sex composition of jobs (i.e., occupation- establishment) accounts for a large portion of the wage gap.

We make no attempt to settle the important conceptual issues that go along with the empirical patterns we address, namely the sources of observed patterns, neither from the demand side, that is, discriminatory behavior by employers, or the supply side, that is, behaviors by employ- ees and prospective employees (see England 1992, chap. 2). Nevertheless, our results have implications for the kinds of theoretical issues that are most in need of being addressed and for the type of data that need to be collected and analyzed.


We use two large-scale data sets. The first data set comes from 16 Industry Wage Surveys (IWSs) conducted by the U.S. Bureau of Labor Statistics (BLS) in the period 1974-83 (see, e.g., U.S. Department of Labor 1976a), corresponding to industry codes at three and more digits as defined in the Standard Industrial Classification Manual (see U.S. Executive Office of the President 1987). Eleven industries were surveyed in 1974-78, while five were surveyed in 1980-83. The populations for the surveys and the sampling from the populations are described in the U.S. Department of Labor publications listed in the note to table 1 (e.g., 1976a, p. 48). Of the 16, 11 are manufacturing industries, while 5 are service industries. The selection of industries was to a large extent determined by availability from the BLS.6 Table 1 lists the industries analyzed.

These surveys ceased to be collected in September 1990, after about 100 years in operation. Information on gender of employees was no longer collected by the mid- 1980s. The data analyzed below may be the last Industry Wage Survey to be made available to researchers, due to changes in data-processing procedures at the BLS.

In each industry, the BLS drew a sample of several hundred establish- ments, often covering a large proportion of the establishments in the industry. Separate offices or plants of a single company are considered separate establishments. For each establishment, information was ob- tained from establishment records both on establishment characteristics and on a large number of the blue-collar and clerical workers in the establishment. Within each industry, only a selection of occupations were surveyed, on average 42 occupations per industry.' The occupations were selected by the BLS in order to provide a wide representation of blue- collar or clerical occupations in an industry. The individual-level data, on tapes purchased from the BLS, provide information on each individual in the relevant occupation and establishment, as well as on characteristics of the establishment in which the individual worked. Professional and managerial employees were excluded from the data collection. Since these occupations may have exhibited wider variations in wages, even at the occupation-establishment level, there may have been less occupation- establishment level variation in wages here than in samples including professionals and managers.

Information on the establishment characteristics includes the follow- ing: size (number of employees); region and area within region; whether it is located in a standard metropolitan statistical area; union status of establishment and, if unionized, the name of the union organizing the majority of the workers; production technology or major products; the number of employees remunerated by each of 10 different payment schemes; and provision of fringe benefits.

For each employee surveyed, information was obtained on sex, occupa- tion (an industry-specific code), method of wage payment (incentive- or time-rated), and hourly earnings. No information was collected on race, age, experience, or education. The occupational classification is unusu- ally detailed, corresponding in many cases to nine digits in the Dictionary of Occupational Titles (see U.S. Department of Labor 1977~). In other cases, the titles are specific to the BLS data, based on industry-specific codes, but are usually as detailed as the nine-digit titles in the Dictionary of Occupational title^.^

'In the 11 manufacturing industries, most of the data are on production occupations. In the five service industries, the data are mostly on service, including clerical and technical, occupations.

The occupational codes in the IWS data are reported on "job lists" and are intended to reflect jobs in the establishments surveyed. We are therefore able to report within- job wage differences. A job is commonly defined as a specific position, with particular duties and responsibilities, in a specific setting, such as grinder in a given establish- ment (e.g., Treiman and Hartmann 1981, p. 24).

Wage data are straight-time hourly wages in 13 industries and full-time weekly earnings in three, excluding premium pay for overtime and work on weekends, holidays, and late shifts. Thus, we do not conflate pay earned on regular hours with pay earned on overtime and irregular hours, making the wage data less prone to bias than virtually any other study used for assessing wage discrimination. Men work more overtime hours than women (see, e.g., U.S. Department of Labor 1982, table C-33), either due to a preference for more overtime or due to better access to overtime hours, and overtime hours are usually paid at a higher rate.g Nonproduction bonuses, such as year-end bonuses, are also excluded, whereas incentive pay is included (e.g., U.S. Department of Labor 1976a,

p. 48).

Although these data have some serious deficiencies, notably the few individual-level characteristics measured, they also have some unique features that are not likely to be available or replicated in other surveys. First, in each of the industries studied, the data give information about several hundred establishments and a substantial proportion of their em- ployees, namely all employees within each of the occupations included in the survey. This allows us to study intra- versus interestablishment processes, in particular to compare men and women in the same occupa- tion and establishment. No other data set available or likely to be col- lected in the near future is so extensive in this regard.

Second, the wage data are exceptionally good. Most survey data record only monthly or annual earnings. In those cases one needs to impute wages from weeks worked and usual hours worked per week in the period earnings cover (see Stolzenberg 1975, pp. 65 1-52). This procedure is likely to lead to some error, partly, as discussed above, in connection with the overtime versus regular hours issue. The wage data in the IWSs, in contrast, come from establishment records, are not subject to recall error, and are extraordinarily reliable.

Third, the level of occupational detail in these data is unusual (corre-

To see the importance of the issue, consider an example. Suppose that women and men earn equal pay for each hour worked, be it regular or overtime, so there is no within-job wage discrimination. Suppose further that the pay for overtime hours is 50% above that of regular hours. Suppose, finally, that in a sample of full-time employees, the women all work 40 regular hours a week but no overtime, while the men work 40 hours at regular time and an additional 10 hours of overtime. A survey that reports only the total hours worked in a week and the total earnings for the hours worked, not distinguishing regular and overtime hours, will in this example report that men on the average earn 10% more per hour than women. This wage disparity may be taken as evidence of discrimination with respect to the wage rate paid, when what it reflects is that men work more overtime hours. Allocative discrimination may of course operate in the assignment of overtime work.

sponding to nine digits in the Dictionary of Occupational Titles).'' The

intraoccupational variation in personal characteristics, such as education,

is therefore likely to be limited. Hence, the resulting omitted variable

bias should be small. For example, in the case of salespersons in depart-

ment stores, Petersen (1992, table 1) shows that the variation in educa-

tional attainment is small within the 11 sales occupations available for

that industry. This variation is also likely to be small in other industries,

but the information available on occupation in the census is unfortunately

too crude to match with the occupational information in the IWSs.

Table 1 gives descriptive statistics for each of the 16 industries: the percentage female in the industry, the number of workers, the number of occupations, the number of establishments, the number of occupation- establishment pairs, and the average wage by sex. The table shows that the sample sizes are large within each industry in terms of workers, occupations, and establishments.

The second data set we use is the National Survey of Professional, Administrative, Technical, and Clerical (PATC) employees in 198 1, also conducted by the BLS (U.S. Department of Labor 1981~). The sampling and data collection design for this survey is similar to that in the IWS data. Data were collected on weekly pay for full-time employees in 23 PATC occupations. We use the data for the seven professional and three administrative occupations. The remaining 13 technical and clerical oc- cupations add little new to the analysis of the IWS data, which cover a wide array of clerical and some technical occupations in the five service industries. Each occupation is further divided into a set of ranks, corre- sponding to a hierarchy in terms of authority, responsibility, and qualifi- cations required, yielding for the professional and administrative occupa- tions altogether 51 occupation-by-rank groups, with some managerial positions included among the higher ranks. For example, for chemists, rank I is an entry-level job requiring a bachelor's degree in chemistry and no job experience, while rank VIII is a job where the incumbent "Makes decisions and recommendations that are authoritative and have far-reaching impact on extensive chemical and related activities of the company" (U.S. Department of Labor 1981a, p. 54). The data were collected from broad industries: mining, construction, manufacturing, transportation, communication, electric, gas, sanitary services, retail

'O A sense of this detail can be obtained from Petersen (1991, table I), where wage data on 44 detailed, mostly production, occupations in the nonferrous foundries indus- try are analyzed. Nine of the occupations analyzed are chippers, grinders, core assem- blers, die-casting machine workers (operate only), die-casting machine workers (set-up only), shipping clerks, receiving clerks, truckers (forklift), and truckers (not forklift). A list of the occupations or jobs and their relationships to titles in the Dictionary of Occupational Titles is available from the authors upon request.

trade, finance, insurance, and selected services (U.S. Department of La- bor 1981a, p. 31).

We have been given a somewhat limited access to the PATC data. We have information on all of the 10 professional and administrative occupations and the 51 occupation-by-rank groups, from 2,162 establish- ments, covering probably all of the approximately 740,000 employees for which data were collected (see U.S. Department of Labor 1981a, table 11, p. 11). However, unlike the IWS data, we do not have access to the individual-level wage data. We only know the average wages by sex within each of 16,433 occupation-by-rank-establishment pairs. Thus, where a pair is sex integrated, which is the case for 4,036 of the 16,433 pairs, we can compute the wage gap at the occupation-by-rankestablishment level, the level of greatest interest here. l1 Unfortunately, we cannot compute the overall wage gap in an occupation or numbers comparable to those reported in table 1 for the IWS data, because we do not know the number of men and women employed in the occupation-by- rank-establishment pairs. Therefore, no further presentation of these data is made until Section VI below.


Descriptive Statistics for Wage Differences

The discussion of methods focuses on the procedure for analyzing the IWS data. Only one of the four central quantities reported can also be computed in the PATC data, as indicated below.

We report all statistics separately for each of the 16 industries in the IWS data.'' The raw (average) wages, either hourly or full-time weekly earnings, for women and men in an industry are given by Zf and Z,, reported in columns 7-8 in table 1. The average wages for women and men in occupation o are Go,f and E,,, and the relative wages are Z,,,



-wo,~E0,,. The average wages for women and men in establishment

e are we, and Z,,, and the relative wages are Z,,


,.57, f/Ze,


average wages for women and men in occupation-establishment pair oe

,,are Go, and Go,,

and the relative wages are Eoe,,= Z,,flGoe,m.

Using these averages and ratios, we report several statistics separately for each industry. The raw relative wages between men and women are

"In the entire PATC data set, we have access to data on 23 occupations, 96 occupation-by-rank groups, 2,725 establishments, 39,159 occupation-by-rank-establishment pairs, of which 8,104 are integrated, covering about 1.4 million employees.

l2 Unfortunately, we cannot compare wages across industries, because the industry data come from different years and hence reflect inflation as well as general wage increases.

given as the ratio of average women's to average men's wages (multiplied by 100):

We make three decompositions of the raw relative wages. Each decom- position answers the following question: Suppose sex segregation-by occupation, establishment, or occupation-establishment-were abolished; what then would the remaining gender relative wages be? The wage gap or female wage penalty at each of the three levels obtains then as 100 minus the relative wages at the corresponding level. At the occupation-establishment level, the remaining wage gap can reasonably be interpreted as an estimate of the upper bound on the amount of within- job wage discrimination, the gap one would observe in the absence of occupation-establishment sex segregation. It is an upper bound because the remaining wage gap at the occupation-establishment level could have been caused by other factors, such as differences in experience and hu- man capital between men and women in a given category. The difference between the raw wage gap and the wage gap at the occupation-establishment level can thus be interpreted as being attributable to occu- pation-establishment sex segregation. This part, of course, need not be caused by discrimination alone.

The relative wages controlling for occupation obtain as

It is computed over every integrated occupation, those where both men and women are present. Here, No(,,is the number of integrated occupa- tions.

Similarly, the establishment-level relative wages are given as

which is computed for every integrated establishment, those where both men and women are present. Here, N,(,, is the number of integrated establishments.

The occupation-establishment level relative wages are given as

which is computed for every integrated occupation-establishment pair,

those where both men and women are present. Here, No,(,,is the number

of integrated occupation-establishment pairs. We also compute this mea-

sure in the PATC data.

For each of the three categories-occupation, establishment, and occu- pation-establishment-the numbers in equations (2)-(4) give the unweighted average of the within-category relative wages. The numbers do not take into account the distribution of employees or the sex distribu- tion on the various observations within a category, where an observa- tion in a category refers to the relative wages in a given occupation, establishment, or occupation-establishment pair.13 We have also com- puted the decompositions using three alternative weighting procedures: by the proportion of workers, of men, or of women employed in the category.l4

Naturally, other measures could be used, but none occupies a theoreti- cally privileged status (see Blau 1977, pp. 65-67). The pattern of our results and conclusions are unaffected by the particular weights used. We report the unweighted averages for two reasons.'' The first is that we want to report the average of what goes on at the occupation, estab- lishment, or occupation-establishment level. At the occupation-establish- ment level, we thus report the average behavior of employers at that level, rather than the relative wages faced by, say, the average woman at the occupation-establishment level.

The second reason we use the unweighted average is that we would like to assess what would happen to the wage gap were the categories to become desegregated. Using weights based on the current size distribu- tion of categories, or the current sex segregation, does not then make

l3 The regression estimates, discussed below, do take these distributions into account. l4 The results using the three alternative weights are available in Petersen and Morgan (1995, app. B). For example, at the occupation-establishment level, we computed


where Po, is respectively the proportion of workers, of male workers, or of female workers that is employed in integrated occupation-establishment pair oe. The basis for the proportion is the number of workers, of males, or of females employed in integrated occupation-establishment pairs in the given industry.

l5 We report averages of the within-cell relative wages, not medians. Studies that use annual earnings for full-time workers often report medians after having corrected for hours worked (e.g., Treiman and Hartmann 1981, table 8). The measure of hours worked is often imprecise, and medians minimize the impact of over- or under-reporting of hours worked. Our data do not suffer from this problem of erroneous outliers.

sense, because it assigns most weight to categories with a large number of employees, which are more likely to be integrated by pure chance alone, and those are not the weights that will be observed after desegrega- tion. Using the unweighted relative wages among currently integrated categories may make more sense, because those weights may better de- scribe what will happen in a desegregated state.

Using the results in equations (2)-(4) we decompose the raw wage gap obtainable from (1) into the parts that are due separately to occupation, establishment, and occupation-establishment sex segregation. The per- centage of the wage gap from (1) due to occupational segregation alone is given by

The percentage of the wage gap from (1) due to establishment segregation alone is given by

The percentage of the wage gap from (1) due to occupation-establishment segregation alone is given by

By comparing (5)-(7) we can evaluate what is more important for ex- plaining the wage gap: occupational, establishment, or occupation-establishment segregation.

Regression Analysis for Wage Differences

Most of our analysis will focus on the results that obtain from the proce- dures just discussed. However, to make our analysis comparable to stan- dard regression models for wage differences, we extend the results with some regression models.

We use linear regression andysis of the logarithm of the hourly wage on establishment-level and individual-level characteristics. We chose the semilogarithmic form of the wage equation for its ease of interpreta- tion: a coefficient is interpretable (roughly), after having been multiplied by 100, as the percentage change in the dependent variable resulting from a unit increase in the independent variable (see Petersen 1989, sec. 3).

Below, the subscripts used are as follows: i for individuals, o for occu- pations, and e for establishments. So, the subscript ioe denotes individual i in occupation o in establishment e.

Two generic equations are in focus:


Here, xi,, includes the covariates for individual i in occupation o and establishment e and Sioeis the employee's sex (1 = women; 0 = men). The only individual-level characteristic included in x,, is method of pay (incentive vs. hourly), while occupation is captured in qoeHowever, xioe contains a large number of establishment-level variables, each of which is entered as a dummy or set of dummy variables. In many industries the regression equations include as many as 70 parameters and in one case as many as 109.16 Finally, Noand Neare the numbers of occupations and establishments.

In the analysis of covariance terminology (e.g., Judge et al. 1985, chap. 13.3), equation (8) corresponds to the total estimator, while equa- tion (9) corresponds to the within estimator. In (9), qoeis treated as a fixed effect, entered as one or two sets of dummy variables. We report estimates where (i) qoe= q,, occupation effects only, (ii) qoe= qe, establishment effects only, (iii) qoe= qo + qe,occupation and establish- ment effects, and (iv) qoe= qoe,occupation-establishment interaction effects."

The set of equations in (8)-(9) allows one to determine whether the differences within occupation-establishment pairs resemble the differ-

l6 At the establishment level the regressors included are size (number of employees); region; whether it is located in a standard metropolitan statistical area; union status of the establishment and, if unionized, the name of the union organizing the majority of the workers; and production technology and major products. As noted above, at the individual level, the regressors included are sex, method of wage payment (incen- tive- or time-rated), and occupation.

l7 For the estimators in (ii)-(iv) we used the within estimator (e.g., Judge et al. 1985, chap. 13.3). We also computed several other estimators, including, in the analysis of covariance terminology, all the between estimators, where data are aggregated to various levels (e.g., the occupation or establishment level), as well as random-effects estimators corresponding to the fixed-effects estimators (except the one with the occu- pation-establishment interaction, which we were not able to compute in some of the industries). These additional estimators provide the same pattern of results as those reported below. They are available from the authors upon request.

ences between occupations and establishments. For example, if there are no differences between men and women from the within-occupation- establishment estimator but quite large differences between men and women from the within-occupation estimator, then we can conclude that wage differences between men and women to a large extent come about through allocation to given occupation-establishment pairs, not only through allocation to given occupations.

Segregation Measures

The final statistics we present are of the degree of segregation across occupations, establishments, and occupation-establishment pairs. The first measure is the standard Duncan and Duncan (1955) segregation index. It equals zero if there is no segregation and one if the segregation is complete. In the case of occupational sex segregation, the index mea- sures the proportion of women (or men) who would have to change occu- pation in order for there to be no occupational sex segregation. When it equals 50, 50% of the women (or men) need to change occupation; when it equals one of the extremes (0 or I), none or all of the women (or men) need to change occupation. The interpretations are analogous in the case of the indices for establishment and occupation-establishment segrega- tion.18 The second measure we report is the percentage of occupations, establishments, and occupation-establishment pairs that were perfectly segregated, either all female or all male, as well as the proportions of men (and women) that were in those categories.


Descriptive Statistics for Wage Differences

Table 2, columns 2-5, gives the raw relative wages, the relative wages controlling for occupation, the relative wages controlling for establish- ment, and the relative wages controlling for occupation-establishment, computed from equations (1)-(4), all from the IWS data pertaining mostly to blue-collar and clerical occupations, as well as some technical occupa- tions. Column 1 shows the percentage female in each industry, taken

l8 The formula, for occupation-establishment segregation, is

where is the proportion of the women in the industry who are employed in occupation-establishment pair oe, while Po,,,is the same proportion for men. The formulas for establishment and occupational segregation alone are similar.


      Occupation-     Occupation
%F Raw Occupation Establishment Establishment Occupation Establishment Establishment
INDUSTRY (1) (2) (3) (4) (5) (6) (7) (8)
.................................Men's and boys' shirts                
Banking* ..................................................                
Lie insurance* ...........................................                
..............................................Wool textiles                
Cotton and synthetic fiber textiles ...................                
.......................Miscellaneous plastic products                
Hotels and motels .......................................                
Computer and data processing services* ..........                
Wood household furniture ............................                
..........................Textile dyeing and finishing                
...................................Nonferrous foundries                
Paints and varnishes ....................................                
Industrial chemicals ....................................                
Fabricated structural steel ............................                
Average across industries? ............................                

NOTE.-For description of data see Sec .I1. For description of procedures, see Sec .111.Col. 1, %F. gives the percentage female in the industry .Col. 2 gives raw relative wages from eq . (1). Cols. 3-5 give the average of the within occupation, within establishment, and within occupation-establishment relative wages, camptited from eqq . (2)-(4) . Cols. 6-8 give the percentage of the raw wage gap obtained from col . 2 that can be attributed to occupation, establishment, and occupatioa-establishment segregation, computed from eqq . (5)-(7) .

* Based on weekly wages for full-time employees .
t Each column gives the unweighted average across industries of the percentages in the respective column .

from column 1 of table 1. Column 2 shows that the average of the indus- try-specific raw relative wages of women compared to men was 81%. Thus the average of the industry-specific female wage penalties was 19%, where the wage penalty is the percentage by which women earned less than men, computed as 100 minus the relative wages.

The relative wages in columns 3-5 are computed on the basis of inte- grated categories-that is, integrated occupations, establishments, or occupation-establishment pairs. As a result, the number of employees over which these statistics are computed is much lower than the number of employees over which the raw relative wages are computed.

Column 3 shows that the female wage penalty is reduced to an average, across the 16 industries, of 8% when we control for occupation using equation (2). Column 4 shows that the female wage penalty is reduced, across the 16 industries, to an average of 15% when we control for establishment using equation (3). Column 5 shows that the female wage penalty is reduced to about 0%-4% when we control for the occupation-establishment pair using equation (4). The average of the industry-specific female wage penalties, controlling for the occupation- establishment pair, is 1.7%.

These results are striking. Controlling for occupation or establishment alone reduces the wage gap somewhat but not drastically. Controlling for the occupation-establishment pair, however, reduces the wage gap to a point of virtually no difference between men and women. The re- maining female wage penalty is, on average, 1.7% across the 16 indus- tries, even without controlling for any individual-level characteristics such as education, age, seniority, and race. Occupation-establishment segregation accounts better for wage differences between men and women than any other set of variables studied in the literature on wage differences.l9

Columns 6-8 give a decomposition of the raw wage gap obtained from column 2 into the percentage that is due to occupation, to establishment, and to occupation-establishment segregation, when each dimension is considered alone, using equations (5)-(7). Column 6 shows that occupa- tional sex segregation alone accounts for about 64% of the wage gap, somewhat more than other studies (e.g., Treiman and Hartmann 1981; Reskin and Roos 1990). Column 7 shows that establishment sex segrega- tion alone accounts for about 24% of the wage gap. This is a new result. To date, occupational segregation has been the primary focus of gender

l9 As discussed in the introduction, Groshen (1991) and Tomaskovic-Devey (1993) report similar results on wage differences. However, neither author studied the wage differential at the occupation-establishment level.

wage gap studies. Column 8 shows that occupation-establishment sex segregation alone accounts for a large percentage of the wage gap, rang- ing from a low of 70.5% to a high of 98.4%. In 14 of 16 industries, more than 80% of the wage gap is explained by occupation-establishment segregation, and among seven of those more than 90% is explained, while in two industries only 70% and 77% of the gap is explained.

Again, we have the striking result that the interaction of occupation and establishment segregation explains not only dramatically more of the wage gap than either establishment or occupational segregation alone, it also explains most of wage differences between men and women. Had men and women been equally distributed on occupation-establishment pairs, on the average 89% of the wage gap would have disappeared, assuming that other forms of discrimination would not have emerged (see Reskin 1988).

One point requires discussion here. The female wage penalty for the workers in our sample is about 20%. This amount is substantially less than the about 40% penalty for all full-time workers during the period of the surveys (see, e.g., Goldin 1990, p. 61; see also the case studies based on the 1980 census in Reskin and Roos [1990]). There are several reasons for this discrepancy. One is that these studies look at median or average annual earnings for full-time employees. It is probable that men on the average work more hours per year than women and that they also work more overtime hours (see Sec. I1 above). Therefore, differences between men and women are probably overstated. Another reason for this difference is that the occupations covered here make up a relatively narrow range of the overall spectrum. In particular, professional and managerial occupations are excluded from the sample. The excluded oc- cupations are expected to be on average higher paid than those included and to be occupied primarily by men. As a result, we have, when comput- ing the raw wage gap, some degree of de facto control for occupation.

One may also speculate about what causes the differences in the wage gap across industries, but that is not our purpose here. To do so meaning- fully would require a larger selection than our 16 industries. We note that the pattern of wage differences at the various levels is rather similar across the 16 industries. The five service industries display on the average the lowest wage gaps at the occupation-establishment level, 0.7%, while the same average across the 11 manufacturing industries is 2.2%. We cannot say whether this difference is due to the manufacturing-service distinction or whether it is because 10 of the 11 manufacturing industries were surveyed in the earlier period, 1974-77, while all five service indus- tries were surveyed in the later period, 1978-83, a period in which per- haps the wage gap was smaller.

Regression Analysis of Wage Differences

Table 3 gives estimates of the sex effect (y) from equations (8) and (9) of Section 111. See note 16 in Section I11for the set of regressors incl~ded.~' To permit easy comparison between regression estimates and the preced- ing analysis, column 1 gives the women's raw wage penalty, expressed as the percentage by which women earned less than men, computed from column 2 of table 2. Column 2 in table 3 gives the effect of sex with all control variables except occupation and establishment. We refer to this as the total estimator. It gives the net wage difference between women and men not controlling for occupation, establishment, or occupation- establishment but with several other controls, including establishment- level characteristics, such as size and region. Apart from hospitals, where women earned more than men, the sex effect ranges from -.08 to -.45, meaning that women on the average earned less than men per hour, given several relevant establishment-level controls. The average of the industry-specific sex effects is -.180, a wage difference of about 18%.

Column 3 gives the same set of estimates except with detailed occupa- tional controls. These are referred to as the within-occupation estimators. In each industry, the sex effect, although still substantial, is reduced by quite a bit. The average of the industry-specific sex effects is -.070, a wage difference of about 7%. This means that occupational sex segrega- tion accounts for some of the wage difference between men and women but certainly not all of it.

In column 4 we introduce controls for establishments, one dummy variable per establishment, but there are no controls for occupation. Controlling for the establishment effects reduces the sex effect somewhat but not by as much as controlling for occupation. The average of the industry-specific sex effects is -. 147, a wage difference of about 15%.

In column 5 we introduce controls for both establishments and occupa- tions, one set of dummy variables for each of the two dimensions. Intro- ducing these controls reduces the sex effect further. The average of the industry-specific sex effects is -,044, a wage difference of about 4%.

In column 6 we have included controls for the establishment-occupation pair, yielding the establishment-occupation interaction effect. Introducing this control further reduces the sex effect as compared to columns 4 and 5.21 Indeed, the sex effect is now quite small. In two of

20 A list of the regressors included can be obtained from the authors. '' The differences between the effects reported in cols. 5 and 6, the additive effects of occupation and establishment versus the interaction effects between the two, show that it was not merely a matter of women being in low-paying occupations and low- paying establishments. Rather, an occupation would pay less in an establishment that hired mostly women in that occupation than in an establishment that hired mostly

the 16 industries, the sex effect is not significantly different from zero. In two industries it is positive, and in 10 of the 12 remaining industries, it is 3.5% or less. The largest sex effect is 4.5%, found in the wood household furniture industry. The average of the industry-specific sex effects is -.015, a wage difference of about 1.5%. Most important, these occupation-establishment level differences obtain in the absence of con- trols for any individual-level characteristics, except for the payment sys- tem, hourly versus piece rate. This small sex effect can reasonably be interpreted as an estimate of the upper bound on the amount of within- job wage discrimination.

Similar decompositions of the wage gap as given in columns 6-8 in table 2 can be computed from the regression analyses in table 3. These replicate the qualitative pattern in table 2 ." Thus, the regression analy- sis leads to the same conclusions as the descriptive analysis based on table 2.


The main finding of Section IV is that wage differences (in the period covered) between men and women were primarily the result of sex segre- gation, particularly segregation at the occupation-establishment level. Wage differences at the occupation-establishment level were quite small. We therefore end by reporting the degree of sex segregation in the indus- tries studied. We emphasize that the segregation measures reported per- tain to the set of occupations included in the surveys, which exclude managerial and professional occupations. Thus, a perfectly segregated establishment was perfectly segregated in the occupations included in the survey, not necessarily in all occupations.

Table 4 reports the Duncan and Duncan (1955) measures of segregation at the occupation, establishment, and occupation-establishment level for each industry (see n. 18, Sec. 111). In 12 of the 16 industries, occupational sex segregation is somewhat greater than establishment segregation. Occupation-establishment segregation is by far the most extreme, with indices ranging from .62 to .97. Note that occupation-establishment seg- regation by necessity must yield a larger index than either occupation or establishment alone.

Table 5 gives further information on the amount of segregation. For each industry there are two lines, and within each line there are three groups, one each for occupations, establishments, and occupation-

men in that occupation, even though the former establishment needed not pay uni- formly lower wages in other occupations within the establishment. ZZ The decompositions are available from the authors upon request.



%F Occupational Establishment Establishment
Industry (1) (2) (3) (4)
Men's and boys' shirts ..........        
Hospitals ............................        
Banking .............................        
Life insurance ......................        
Wool textiles .......................        
Cotton and synthetic fiber tex-        
tiles ................................        
Miscellaneous plastic products .        
Hotels and motels .................        
Computer and data processing        
services ...........................        
Wood household furniture ......        
Textile dyeing and finishing ....        
Machinery ........................        
Nonferrous foundries ............        
Paints and varnishes .............        
Industrial chemicals ..............        
Fabricated structural steel ......        

Average across industries* ......

NOTE.-For description of data see Sec. 11. For description of procedures see Sec. 111. Col. 1, %F, gives the percentage female in the industry. Cols. 2-4, give the Duncan and Duncan (1955) segregation indices. For the formula in the case of occupation-establishment sex segregation, see n. 18, Sec. 111.

* Each column gives the unweighted average across industries of the percentages or segregation indices in the respective column.

establishment pairs. The first line reports the percentage of occupations, establishments, and occupation-establishment pairs that were completely female segregated, completely male segregated, and integrated. The sec- ond line (numbers given in parentheses), gives the percentage of the women in the industry who were in a totally female segregated category, the percentage of the men in the industry who were in a totally male segregated category, and the percentage of the workforce in the industry who were in an integrated category, respectively. Again, we have done this for occupations, establishments, and occupation-establishment pairs.

In each of the 16 industries, 50% or more of the occupations were integrated. In 12 of 16 industries 80%-100% of the occupations were integrated, and in three industries all the occupations were integrated. Sex segregation was higher at the establishment level. In 4 of 16 industries less than 40% of the establishments were integrated. But still, in 8 of 16 industries 80%-100% of the establishments were integrated, and in one

industry all the establishments were integrated. For both establishment

and occupational segregation, the distribution of employees on the three

categories-female segregated, male segregated, and integrated-was

similar to the distribution of the occupations and establishments them-


Once one turns to occupation-establishment segregation the picture is entirely different. In 8 of 16 industries, less than 15% of the occupation- establishment pairs were integrated. In the other 8 industries, only 20%- 35% of the pairs were integrated. In 15 of 16 industries, the percentage of the workforce employed in integrated occupation-establishment pairs is greater than the percentage of the pairs that were integrated. This indicates that integration was more likely in occupation-establishment pairs with large numbers of employees. These results demonstrate in another way that segregation by occupation-establishment is much higher than segregation by occupation or by establishment alone.

Regarding sex segregation at the occupation-establishment level, we note that some portion of this may have come about through pure chance, for the simple reason that many of the occupation-establishment pairs employ very few workers. Across the 16 industries, 23.6% of the 71,214 occupation-establishment pairs employ only one worker, and 56.0% of the pairs employ only 1-4 workers. By definition then, 23.6% of the pairs must be perfectly segregated. And by chance, a large proportion of the others will be as well. For example, with a workforce composition of 40% female, random matching of workers and jobs would lead to perfect segregation in about 5 1% of the occupation-establishment pairs employing two workers, in about 28% of the pairs employing three work- ers, and in about 16% of the pairs employing four workers. The sheer size distribution of occupation-establishment pairs would, by pure chance, lead to about 36% of the occupation-establishment pairs being perfectly segregated. Of course, even if the observed segregation could be shown to arise exclusively from random processes, there may be non- randomness in how men are allocated to the high-paying and women to the low-paying occupation-establishment pairs.


The results in the preceding sections were based on data mostly for blue- collar, clerical, and some technical employees. It might be suspected that the wage gap among professional, administrative, and managerial employees is greater. In this section we turn our attention to analysis of wage differences and sex segregation among these employees using PATC data.





% % % % % % % % %

%F Women Men Integrated Women Men Integrated Women Men Integrated INDUSTRY (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)


'.A N Men's and boys' shirts ............................... 92.5

Hospitals ................................................. 84.8

Banking .................................................. 82.8

Life insurance .......................................... 75.9

Wool textiles ............................................ 55.4

Cotton and synthetic fiber textiles ................. 54.6 .O 11.6 88.4
(.O) (6.6) (97.0)
Miscellaneous plastic products ..................... 5 1.9 .O 11.9 88.1
(.O) (8.8) (95.8)
Hotels and motels ..................................... 5 1.5 .O .O 100.0
(.O) (.O) (100.0)

Computer and data processing services .......... 44.5

Wood household furniture ........................... 35.9

Textile dyeing and finishing ........................ 19.8

Machinery ............................................ 15.5

Nonferrous foundries ................................. 9.4

Paints and varnishes .................................. 5.6

Industrial chemicals ................................... 2.5

Fabricated structural steel ........................... .6



Average across industries* .......................... 42.7

NOTE.-For description of data see Sec. 11. Col. 1, %F, gives the percentage female in the industry. Within each industry, the first line gives the percentage of

occupations (cols. 2-4), establishments (cols. 5-3, and occupation-establishment pairs (cols. 8-10), that are filled only by women, only by men, or by both women and

men (integrated), respectively. The numbers in parentheses give the percentage of the women, of the men, and of the workforce that are in occupations (cols. 2-4),

establishments (cols. 5-7), and occupation-establishment pairs (cols. 8-10) that are filled only by women, only by men, or by both women and men (integrated). For

instance, in the banking industry 10.2% of the occupations are filled exclusively by women, and .3% of the women in the banking industry are in these women-only

occupations. These distributions characterize only occupations included in the surveys, which exclude all exempt occupations. Therefore, the numbers for segregated

establishments mean that these establishments are totally segregated in the nonexempt occupations included in the survey.

* Each column gives the unweighted average across industries of the percentages in the respective column.

Table 6 gives the average of the occupation-by-rank-establishment level relative wages, first for the seven professional and next for the three administrative occupations. The entries in the table correspond to those in column 5 of table 2. The first column, titled "overall," gives the average occupation-by-rank-establishment level relative wages across all the ranks in the occupation. Columns 2-8 (marked I-VII) give the rela- tive wages by rank within the occupation, from I (low) to VII (high). The numbers of occupation-by-rank-establishment pairs used to com- pute each of the average relative wages are in parentheses.

Across the 10 occupations, the average wage penalty at the occupation- by-rank-establishment level is 3.1%. This penalty increases with the rank within the occupations, from 1% at the lowest to about 5% at the highest ranks, which include some managerial positions. Note that the highest ranks within an occupation vary between the occupations: VII for engineers and IV for auditors. Of course, the extent to which the penalty of about 5% at the highest ranks is due to within-job wage discrimination or due to other factors such as experience cannot be ad- dressed with these data.23

Results for sex segregation are given in table 7, corresponding to the percentages in the first line for each industry in columns 8-10 of table

5. For each occupation, column 1 gives the percentages of the occupation- establishment pairs, not taking into account the rank of employees within an occupation, that fall into each of three segregation statuses: employing men only, women only, and integrated, plus the number of pairs that form the basis for the percentages (N). Columns 2-10 give the same percentages for occupation-by-rank-establishment pairs, first across all ranks (col. 2) and second separately for each rank (cols. 3-10) within an occupation.

Across the 10 occupations, not taking into account the ranks, 39.2% of the occupation-establishment pairs are integrated. Across occupation- by-rank, 24.6% of the occupation-by-rank-establishment pairs are inte- grated, which is comparable to what was found among blue-collar and clerical employees in the IWS data in Section V. One occupation- directors of personnel-stands out as having a much higher level of segre-

23 For the four technical occupations in the PATC data the corresponding overall wage gap at the occupation-by-rank-establishment level is 1%. At the lowest rank, women earn on average 1% more than men, while at the highest ranks, which vary by occupation, women earn about 3.5% less than men. For the nine clerical occupa- tions in the PATC data, women on average earn 0.7% more than men at the occupa- tion-by-rank-establishment level. At the lowest rank women on average earn 1.3% more than men, while at the highest ranks they earn 0.5% more than men. This result is consistent with the results in table 2, where the lowest gaps were found in service industries, where the data cover a large number of clerical and technical employees.

gation than other occupations, where only 2.3% of the occupation-

establishment pairs are integrated and 86.8% employ only men. Except

for minor variations, as the rank within an occupation increases, both

the percentage of pairs employing only women and the percentage of

integrated pairs decline, while the percentage of pairs employing only

men increases. Thus, the higher echelons of a profession are domi

nated by men. At the highest ranks in five of the 10 occupations, more

than 90% of the occupation-by-rank-establishment pairs employ only


In conclusion, we have two findings. The first is that the average wage

gap or penalty at the occupation-by-rank-establishment level is quite

low, an average of 3.1%. Taking occupation-by-rank-establishment as

a "job" then, this result is consistent with the conclusion from section

4: even among professional, administrative, as well as some managerial jobs in the professions, within-job wage discrimination is not a driving force for the gender wage gap. The second finding is that the degree of integration within these professional and administrative occupations decreases with rank. Whether or not women's relative absence from the higher ranks is explained by differences in experience is an open question. Together with the wage gap finding though, this result suggests that, to the extent there is a wage gap within professional and administrative occupations that is not accounted for by differences in experience or training (human capital), it is driven by differences in rank within the occupation along with establishment matching.


The findings of this article are simple to summarize. First, wage differ- ences within given occupation-establishment pairs were relatively small: on the average 1.7% among blue-collar and clerical as well as some technical employees (using IWS data), while on average 3.1% in seven professional and three administrative occupations, ranging from 1% in the lower to 5% in the higher ranks in an occupation (using PATC data). Thus, occupation-establishment segregation, not within-job wage discrimination was the driving force for observed wage differences. Second, among blue-collar and clerical employees, establishment segre- gation was important for wage differences between men and women, although not as important as occupational segregation. Third, among blue-collar and clerical employees sex segregation across establishments was extensive but perhaps not as extensive as sex segregation across detailed occupational groups. Among professional and administrative employees sex segregation increased strongly with the rank within an occupation.

  OCCUPATION Overall (1) I (2) I1 (3) 111 (4) IV (5) V (6) VI (7) W (8)
  Accountants ........................................ 97.3              
  Chief accountants ................................. (1,319) 94.7              
  Auditors ............................................. (1) 97.8              
  ...............................Public accountants (294) 99.3              
  ...........................................Attorneys ............................................Chemists (142) 96.4 (234) 96.9              

The first finding is important. It shows that occupation-establishment segregation accounted for more of the gender wage gap than any other variable or set of variables currently used in studying the gender wage gap. Occupational segregation accounts for about 40% of the wage gap, while human capital and other variables account for about 40% (e.g., Treiman and Hartmann 1981, chap. 2). But no set of variables, either individual or structural, accounts for as much as 89%, as occupation- establishment segregation did here in the case of blue-collar and clerical employees in the IWS data.

This first finding establishes the conjecture already made in the litera- ture but not yet documented: wage differences are to a larger extent generated by occupation-establishment segregation than by within-job wage discrimination. Along with the first, the second finding shows the need to study establishment as well as occupational segregation. The third finding confirms what has already been established for California (Bielby and Baron 1984), but using data covering the entire United States and information on about 30 and 20 times as many employees and estab- lishments respectively, though focusing on a narrower set of occupations.

The implications of the findings are straightforward. In terms of policy, allocative and valuative processes should be given the most attention, and within-job wage discrimination, which is covered by the Equal Pay Act of 1963 and which has been the implicit or explicit focus of much discussion and research, should receive less. Future research on differen- tial wage attainment between men and women should be refocused as follows. The emphases should be less on within-job wage discrimination and more on three prior processes. The first is the entry of employees into occupations and establishments, that is, the differential access of men and women to positions during the initial hiring or matching process, an allocative mechanism, a topic that is not easy to research (see Col- linson, Knights, and Collinson 1990). The second is career advancement within establishments, that is, the differential rates of promotion for men and women, also an allocative mechanism, a line of research already well under way (e.g., Spilerman 1986). The third is how jobs occupied primarily by women tend to be paid less than those occupied by men, the comparable worth issue, or what we refer to as valuative discrimina- tion, also a line of research already well under way (see England 1992).

Two issues arise in identifying allocative processes as responsible for the gender wage gap. The first is whether segregation patterns are due primarily to discrimination or to differences in productive capacities. The meaning of our results for theory and policy depends on which mecha- nism operated here. This question cannot be settled with our data and answering it is obviously a task for future research. It requires informa- tion about productive capacities and would require an analysis of the matching between these and particular jobs.

The second issue concerns the role of supply-side sources of differential attainment between men and women. Of particular interest are the con- straints put on women's career attainment from family obligations and household choices (e.g., Hochschild 1989).'~ While these constraints may result most proximally from a traditional household division of labor, not directly from discriminatory behaviors by employers, household deci- sions are made in light of labor market opportunities, so the two are interdependent. The role of supply-side behaviors and their interrelation- ship with employer behaviors in generating the observed occupation- establishment segregation is in need of research. We cannot settle these issues here.

As to the generalizability of our findings, the results we report are based first on blue-collar and clerical occupations located in 11 manufac- turing and five service industries and second on a selection of professional and administrative occupations, including some managerial jobs in the professions. Obviously there is a need to replicate these findings for a wider set of occupations and industries. It is likely that the general pat- tern reported here will be upheld by such extensions, at least for blue- collar, clerical, technical, administrative, and professional employees, although this is not something we can know. However, the situation for managerial employees may be more complex. There may be more varia- tion in wages and possibly larger differences between men and women within given occupation-establishment pairs in these lines of work. At the same time, occupational sex segregation may be less pronounced in these occupations (see Beller 1984). Therefore, the amount of the wage gap reducible to occupation-establishment segregation may be less in these occupations, and a larger part of it due to within-job wage differ- ences.

There is also a need to replicate the findings using more recent data. However, there is no reason to believe that present conditions are differ- ent from those in the 1974-83 period these data cover. Most likely, the conclusions would be strengthened: within-job wage differences probably would account for an even smaller part of the wage gap today than they did 10-15 years ago. Likewise, sex segregation is likely to be less today than 10-15 years ago (see Jacobs 1989). These extensions should use

24 Reskin and Hartmann (1986, p. 75) conclude: "In sum, although the research evidence does not enable us to say that women's greater responsibility for child care, housework, and family care necessarily contributes to sex segregation in the work- place, it almost certainly plays an important role in limiting their opportunities in general. "

data similar to those used here, data that provide detailed occupational information at the establishment level for employees across a large num- ber of establishments in well-defined industries. Unfortunately, the level of needed detail is not readily available in more standard data sets that comprise a wider spectrum of occupations.


American Association of University Women. 1992. AAUW Report: How Schools Shortchange Girls. Annapolis Junction, Md.: American Association of University Women.

Beller, Andrea H. 1984. "Trends in Occupational Segregation by Race and Sex." Pp. 11-26 in Sex Segregation in the Workplace: Trends, Explanations, Remedies, edited by Barbara F. Reskin. Washington, D.C.: National Academy Press.

Bielby, William T., and James N. Baron. 1984. "A Woman's Place Is with Other Women: Sex Segregation within Organizations." Pp. 27-55 in Sex Segregation in the Workplace: Trends, Explanations, Remedies, edited by Barbara F. Reskin. Washington, D.C.: National Academy Press.

Blau, Francine D. 1977. Equal Pay in the Offcce. Lexington, Mass.: Lexington Books. Collinson, David L., David Knights, and Margaret Collinson. 1990. Managing to Discriminate. London: Routledge. Duncan, Otis Dudley, and Beverly Duncan. 1955. "A Methodological Analysis of Segregation Indices." American Sociological Review 20 (2): 200-17. England, Paula. 1992. Comparable Worth: Theories and Evidence. Hawthorne, N.Y.: Aldine. Goldin, Claudia. 1990. Understanding the Gender Gap: An Economic History of American Women. New York: Oxford University Press.

Groshen, Erica L. 1991. "The Structure of the FemaleIMale Wage Differential: Is It Who You Are, What You Do, or Where You Work?" Journal of Human Resources 26 (3): 457-72.

Hartmann, Heidi I. 1987. "Internal Labor Markets and Gender: A Case Study of Promotion." Pp. 59-92 in Gender in the Workplace, edited by Clair Brown and Joseph Pechman. Washington, D.C.: Brookings Institution.

Hochschild, Arlie. 1989. The Second Shift. New York: Viking. Jacobs, Jerry A. 1989. "Long-Term Trends in Occupational Segregation by Sex." American Journal of Sociology 95 (1): 160-73.

Judge, George G., W. E. Griffiths, R. Carter Hill, Helmut Liitkepohl, and Tsoung- Chao Lee. 1985. The Theory and Practice of Econometrics, 2d ed. New York: Wiley.

Marini, Margaret M. 1989. "Sex Differences in Earnings in the United States." Annual Review of Sociology 15:343-80.

Petersen, Trond. 1989. "The Earnings Function in Sociological Studies of Earnings Determination: Functional Form and Hours Worked." Research in Social Stratifi- cation and Mobility 8:220-49. . 1991. "Reward Systems and the Distribution of Wages." Journal of Law, Economics, and Organization 7 (special issue): 130-58. . 1992. "Payment Systems and the Structure of Inequality: Conceptual Issues and Analysis of Salespersons in Department Stores." American Journal of Sociology 98 (1): 64-107.

Petersen, Trond, and Laurie A. Morgan. 1995. "Occupation-Establishment Sex Seg-

regation and the Gender Wage Gap." Working Paper no. 64. Institute of Industrial Relations, University of California, Berkeley. Reskin, Barbara F. 1988. "Bringing the Men Back In: Sex Differentiation and the Devaluation of Women's Work." Gender and Society 2 (1): 58-81. Reskin, Barbara F., and Heidi I. Hartmann, eds. 1986. Women's Work, Men's Work: Sex Segregation in the Workplace. Washington, D.C.: National Academy Press.

Reskin, Barbara F., and Patricia Roos, eds. 1990. Job Queues, Gender Queues: Ex- plaining Women's Inroads into Male Occupations. Philadelphia: Temple University Press.

Spilerman, Seymour. 1986. "Organizational Rules and the Features of Work Ca- reers." Research in Social Stratification and Mobility 5:41-102. Stolzenberg, Ross M. 1975. "Occupations, Labor Markets, and the Process of Wage Attainment." American Sociological Review 40 (5): 645-55. Tomaskovic-Devey, Donald. 1993. Gender and Race Inequality at Work: The Sources and Consequences of Job Segregation. Ithaca, N.Y.: ILR Press. Treiman, Donald J., and Heidi I. Hartmann, eds. 1981. Women, Work, and Wages: Equal Pay for Jobs of Equal Value. Washington, D.C.: National Academy Press.

Bureau of Census. 1984. 1980 Census of Population. Vol. 2: Subject Reports: Earnings by Occupation and Education. Supplementary Report PC80-2-8B. Washington, D.C.: Government Printing Office (GPO).
Department of Labor. 1976a. Industry Wage Survey: Miscellaneous Plastics, September 1974. Bureau of Labor Statistics (BLS), Bulletin 1914. Washington, D.C.: GPO.

. 19766. Industry Wage Survey: Wood Household Furniture, November 1974. BLS, Bulletin 1930. Washington, D.C.: GPO. . 1977a. Dictionary of Occupational Titles, 4th ed. Bureau of Employment Security. Washington, D.C.: GPO. . 19776. Industry Wage Survey: Textiles, May 1975. BLS, Bulletin 1945. Washington, D.C.: GPO.

. 1977c. Industry Wage Survey: Textile Dyeing and Finishing, June 1976. BLS, Bulletin 1967. Washington, D.C.: GPO. . 1977d. Industry Wage Survey: Nonferrous Foundries, May 1975. BLS, Bulle- tin 1952. Washington, D.C.: GPO. . 1977e. Industry Wage Survey: Paints and Varnishes, November 1976. BLS, Bulletin 1973. Washington, D.C.: GPO. . 1977f. Industry Wage Survey: Industrial Chemicals, June 1976. BLS, Bulletin 1978. Washington, D.C.: GPO. . 1977g. Industry Wage Survey: Fabricated Structural Steel, November 1974. BLS, Bulletin 1935. Washington, D.C.: GPO. . 1979. Industry Wage Survey: Men's Shirts and Separate Trousers, May 1978. BLS, Bulletin 2035. Washington, D.C.: GPO. . 1980. Industry Wage Survey: Hotels and Motels, May 1978. BLS, Bulletin 2025. Washington, D.C.: GPO. . 1981a. National Survey of Professional, Administrative, Technical, and Cleri- cal Pay, March 1981. BLS, Bulletin 2108. Washington, D.C.: GPO. . 1981b. Industry Wage Survey: Banking, February 1980. BLS, Bulletin 2099. Washington, D.C.: GPO. . 1981c. Industry Wage Survey: Life Insurance, February 1980. BLS, Bulletin 2119. Washington, D.C.: GPO.

. 1982. Labor Force Statistics Derived from the Current Population Survey: A Databook. Vol. 1, September 1982. BLS, Bulletin 2096. Washington, D.C.: GPO.

. 1983. Industry Wage Survey: Computer and Data Processing Services, Octo- ber 1980. BLS, Bulletin 2184. Washington, D.C.: GPO.

. 1984. Industry Wage Survey: Hospitals. BLS, Bulletin 2204. Washington, D.C.: GPO.

. 1985. Industry Wage Survey: Machinery, November 1983. BLS, Bulletin 2229. Washington, D.C.: GPO.

U.S. Executive Office of the President. 1987. Standard Industrial Classi$cation Man- ual. Office of Management and Budget. Washington, D.C.: GPO.

  • Recommend Us