This is the html version of the file http://www.gao.gov/new.items/d0435.pdf.
Google automatically generates html versions of documents as we crawl the web.
GAO-04-35 Women's Earnings: Work Patterns Partially Explain Difference between Men's and Woman's <b style="color:black;background-color:#99ff99">Earnings</b>
Page 1
Report to Congressional Requesters
United States General Accounting Office
GAO
October 2003
WOMEN’S EARNINGS
Work Patterns
Partially Explain
Difference between
Men’s and Women’s
Earnings
GAO-04-35

Page 2
Page i
GAO-04-35 Women's Earnings
Letter
1
Appendix I
Briefing Slides
4
Appendix II
GAO Analysis of the Earnings Difference between
Men and Women
21
Review of Other Research on Earnings Differences
21
Data Used in Our Analysis
23
Results of Our Analysis
29
Limitations of Our Analysis
54
Appendix III
GAO Analysis of Women’s Workplace Decisions
56
Purpose
56
Scope and Methodology
56
Summary of Results
57
Background
57
Working Women Make a Variety of Decisions to Manage Work and
Family Responsibilities
59
Related Research
65
Appendix IV
GAO Contact and Staff Acknowledgments
75
GAO Contact
75
Staff Acknowledgments
75
Tables
Table 1: Descriptive Statistics for Selected PSID Variables
26
Table 2: Overall and Separate Model Results for Men and Women
34
Table 3: Summary of Decomposition Results
45
Table 4: Decomposition Results Using Regression Coefficients
46
Table 5: Decomposition Results Using Alternative Estimates
50
Contents

Page 3
Page ii
GAO-04-35 Women's Earnings
Abbreviations
CPS
Current Population Survey
OLS
ordinary least squares
PSID
Panel Study of Income Dynamics
This is a work of the U.S. government and is not subject to copyright protection in the
United States. It may be reproduced and distributed in its entirety without further
permission from GAO. However, because this work may contain copyrighted images or
other material, permission from the copyright holder may be necessary if you wish to
reproduce this material separately.

Page 4
Page 1
GAO-04-35 Women's Earnings
October 31, 2003
The Honorable Carolyn B. Maloney
The Honorable John D. Dingell
House of Representatives
Despite extensive research on the progress that women have made toward
equal pay and career advancement opportunities over the past several
decades, there is no consensus about the magnitude of earnings
differences between men and women and why differences may exist.
According to data from the Department of Labor’s Current Population
Survey (CPS), women have typically earned less than men.1 Specifically, in
2001, the published CPS data showed that for full-time wage and salary
workers, women’s weekly earnings were about three-fourths of men’s.2
However, this difference does not reflect key factors, such as work
experience and education, that may affect the level of earnings individuals
receive. Studies that attempt to account for key factors have provided a
more comprehensive estimate of the earnings difference. However, recent
information is lacking because many studies on earnings differences relied
on data that predated the mid-1990s. But, even when accounting for these
factors, questions remain about the size of and reasons for any earnings
difference. To provide insight into these issues, you asked that we
examine the factors that contribute to differences in men’s and women’s
earnings. On October 2, 2003, we briefed you on the results of our analysis.
This report formally conveys the information provided during that briefing
(see app. I).
To address this issue, we carried out two types of analyses. We performed
a quantitative analysis to determine differences in earnings by gender and
what factors may account for these differences. The statistical model we
1The CPS is a monthly survey that obtains key labor force data, such as employment,
wages, and occupations.
2This figure represents weekly earnings of full-time workers, but considering different
populations may result in different earnings differences. For example, according to a GAO
calculation based on CPS data from 2000 using both full-time and part-time workers,
women’s annual earnings were about half of men’s.
United States General Accounting Office
Washington, DC 20548

Page 5
Page 2
GAO-04-35 Women's Earnings
developed used data from the Panel Study of Income Dynamics (PSID),3 a
nationally representative longitudinal data set that includes a variety of
demographic, family, and work-related characteristics for individuals over
time. We tracked work and life histories of individuals who were between
ages 25 and 65 at some point between 1983 and 2000. Using our statistical
model, we estimated how earnings differ between men and women after
controlling for numerous factors that can influence an individual’s
earnings. (For more information about this analysis and its limitations, see
app. II.) To supplement this analysis, we reviewed the literature and
interviewed a variety of individuals with expertise on earnings and other
workplace issues4 to obtain a broad range of perspectives on reasons why
workers make certain career and workplace decisions that could affect
earnings. In addition, we contacted employers to discuss these issues as
well as to identify what policies employers offered to help workers
manage work and other life responsibilities. (For more information about
this analysis, see app. III.) We conducted our work from September
2002 to October 2003 in accordance with generally accepted government
auditing standards.
In summary, we found:
Of the many factors that account for differences in earnings between men
and women, our model indicated that work patterns are key. Specifically,
women have fewer years of work experience, work fewer hours per year,
are less likely to work a full-time schedule, and leave the labor force for
longer periods of time than men. Other factors that account for earnings
differences include industry, occupation, race, marital status, and job
tenure. When we account for differences between male and female work
patterns as well as other key factors, women earned, on average,
80 percent of what men earned in 2000. While the difference fluctuated in
each year we studied, there was a small but statistically significant decline
in the earnings difference over the time period. (See table 2 in app. II.)
Even after accounting for key factors that affect earnings, our model could
not explain all of the difference in earnings between men and women. Due
to inherent limitations in the survey data and in statistical analysis, we
cannot determine whether this remaining difference is due to
3The PSID is a survey of a sample of U.S. individuals that collects economic and
demographic data, with substantial detail on income sources and amounts, employment,
family composition changes, and residential location.
4These individuals will be referred to as “experts” throughout the remainder of this report.

Page 6
Page 3
GAO-04-35 Women's Earnings
discrimination or other factors that may affect earnings. For example,
some experts said that some women trade off career advancement or
higher earnings for a job that offers flexibility to manage work and family
responsibilities.
In conclusion, while we were able to account for much of the difference in
earnings between men and women, we were not able to explain the
remaining earnings difference. It is difficult to evaluate this remaining
portion without a full understanding of what contributes to this difference.
Specifically, an earnings difference that results from individuals’ decisions
about how to manage work and family responsibilities may not necessarily
indicate a problem unless these decisions are not freely made. On the
other hand, an earnings difference may result from discrimination in the
workplace or subtler discrimination about what types of career or job
choices women can make. Nonetheless, it is difficult, and in some cases,
may be impossible, to precisely measure and quantify individual decisions
and possible discrimination. Because these factors are not readily
measurable, interpreting any remaining earnings difference is problematic.
As arranged with your offices, unless you announce its contents earlier,
we plan no further distribution of this report until 30 days after the date of
this report. At that time, we will provide copies of this report to the
Secretary of Labor and other interested parties. We will also make copies
available to others upon request. In addition, the report will be available at
no charge on GAO’s Web site at http://www.gao.gov.
Please contact me or Lori Rectanus on (202) 512-7215 if you or your staff
have any questions about this report. Other contacts and staff
acknowledgments are listed in appendix IV.
Robert E. Robertson
Director, Education, Workforce, and
Income Security Issues

Page 7
Appendix I: Briefing Slides
Page 4
GAO-04-35 Women's Earnings
Appendix I: Briefing Slides
1
GAO Congressional Briefing
Representative John D. Dingell and
Representative Carolyn B. Maloney
Analysis of the Earnings Difference
between Men and Women
October 2, 2003

Page 8
Appendix I: Briefing Slides
Page 5
GAO-04-35 Women's Earnings
2
Introduction
• Despite extensive research on the progress women have
made toward equal pay, no consensus exists about the size
of any earnings difference between men and women
• Some earnings studies have not accounted for key factors
that affect earnings, such as work experience and education
• Even when accounting for such key factors, questions remain
about the size of and reasons for any difference

Page 9
Appendix I: Briefing Slides
Page 6
GAO-04-35 Women's Earnings
3
Key Question
• What factors contribute to differences in men’s and women’s
earnings?

Page 10
Appendix I: Briefing Slides
Page 7
GAO-04-35 Women's Earnings
4
Scope and Methodology
• We developed a statistical model to estimate how earnings
differ between men and women after controlling for a
comprehensive set of demographic, family, and work-related
factors that can influence an individual’s earnings
• We used the Panel Study of Income Dynamics, a nationally
representative longitudinal data set that includes a variety of
demographic, family, and work-related characteristics
• We tracked work and life histories of individuals who were
between ages 25 and 65 at any point during the period 1983
through 2000

Page 11
Appendix I: Briefing Slides
Page 8
GAO-04-35 Women's Earnings
5
Scope and Methodology (continued)
• To supplement our model, we reviewed literature and
interviewed a variety of individuals to obtain a broad range of
perspectives on why workers make certain career and
workplace decisions that could affect earnings
• Experts reviewed our work
• We conducted our work from September 2002 to October
2003 in accordance with generally accepted government
auditing standards

Page 12
Appendix I: Briefing Slides
Page 9
GAO-04-35 Women's Earnings
6
Summary of Results
• Work patterns are important when accounting for some of the
earnings difference between men and women
• After accounting for factors affecting earnings, women
earned an average of 80 percent of what men earned in 2000
• Our model could not explain all of the earnings difference
between men and women due to inherent limitations in the
survey data and in statistical analysis

Page 13
Appendix I: Briefing Slides
Page 10
GAO-04-35 Women's Earnings
7
Many Factors Account for Earnings
Difference, but Work Patterns Are Key
• While many factors account for the earnings difference
between men and women, work patterns are key
• Some of the other factors include industry, occupation, race,
marital status, and job tenure
• Some of the factors that contribute to an earnings difference
affect men and women differently, but we cannot explain why

Page 14
Appendix I: Briefing Slides
Page 11
GAO-04-35 Women's Earnings
8
Work Patterns Are Important When
Accounting for Earnings Difference
• Men’s and women’s work patterns differ:
• Women have fewer years of work experience
• Women work fewer hours per year
• Women are less likely to work a full-time schedule
• Women leave the labor force for longer periods of time

Page 15
Appendix I: Briefing Slides
Page 12
GAO-04-35 Women's Earnings
9
Work Patterns (continued)
• Years of work experience and hours worked per year differ
for men and women
9
Work Patterns (continued)
• Years of work experience and hours worked per year differ
for men and women

Page 16
Appendix I: Briefing Slides
Page 13
GAO-04-35 Women's Earnings
10
Work Patterns (continued)
• Men and women vary in terms of their full-time work and time
out of the labor force

Page 17
Appendix I: Briefing Slides
Page 14
GAO-04-35 Women's Earnings
11
Perspectives on Why Work Patterns
Differ
• Although the model could not explain why work patterns
differ, according to experts and the literature, women are
more likely to work part time or take leave from work to
manage home and family responsibilities, such as caring for
children
• According to employers, even when they offer part-time work
or leave from work to all employees, women are more likely
than men to use these options, although both men and
women use other work arrangements

Page 18
Appendix I: Briefing Slides
Page 15
GAO-04-35 Women's Earnings
12
Men’s and Women’s Earnings Differ
Even after Accounting for Key Factors
As the graph shows, there were fluctuations in the earnings difference for each year we studied. Over the time period, there
was a small but statistically significant decline in the average earnings difference between men and women.
Note: Data were collected annually through 1997 and then biennially starting in 1999.

Page 19
Appendix I: Briefing Slides
Page 16
GAO-04-35 Women's Earnings
13
Potential Reasons for the Remaining
Earnings Difference
• Our model could not explain all of the earnings difference
between men and women due to inherent limitations in the
survey data and in statistical analysis
• Some experts and literature identified potential reasons for
an earnings difference:
• some women trade off advancement or higher earnings
for a job that offers flexibility to manage work and family
responsibilities
• discrimination resulting from societal views about
acceptable roles for men and women or views about
women in the workplace may affect women’s earnings

Page 20
Appendix I: Briefing Slides
Page 17
GAO-04-35 Women's Earnings
14
Some Women Trade off Earnings for
Flexibility
• According to some experts and literature, some women trade
off career advancement or higher earnings for a flexible job
• For example, a woman may choose a human resources
job that requires less travel and time in the office than an
online position in the company, but offers less opportunity
for advancement and higher earnings
• For example, in medicine, a woman may choose family
practice because it may be more accommodating to
home and family than the surgical specialty, which offers
relatively higher earnings. Surgeons’ work is generally
less predictable because it may require treating
emergencies at all hours

Page 21
Appendix I: Briefing Slides
Page 18
GAO-04-35 Women's Earnings
15
Discrimination May Also Affect Women’s
Earnings
• According to some experts and literature, those who work in
traditionally female-dominated occupations generally receive
less earnings
• Also, according to some experts, discrimination against
women in the workplace negatively affects women’s job
opportunities, advancement, and therefore, earnings

Page 22
Appendix I: Briefing Slides
Page 19
GAO-04-35 Women's Earnings
16
Concluding Observations
• While we could account for much of the earnings difference
between men and women, we cannot explain all of the
difference due to inherent limitations in the survey data and in
statistical analysis
• It is difficult to evaluate the remaining difference without a full
understanding of what contributes to the difference
• An earnings difference resulting from individual decisions
about how to manage work and family may not be a
problem, unless the decisions are not freely made
• An earnings difference may result from workplace
discrimination or subtler discrimination about job choices
women can make

Page 23
Appendix I: Briefing Slides
Page 20
GAO-04-35 Women's Earnings
17
Concluding Observations (continued)
• It is difficult to measure and quantify individual decisions and
possible discrimination
• Because these factors are not readily measurable,
interpreting any remaining earnings difference between men
and women is problematic

Page 24
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 21
GAO-04-35 Women's Earnings
To analyze earnings differences between men and women, we conducted
multivariate regression analyses of the determinants of individuals’ annual
earnings. The regression analyses relate individuals’ annual earnings to
many variables thought to influence earnings, such as number of hours
worked, occupation, education, and experience. In an analysis of data that
included men and women, we used a variable for gender to measure the
average difference in earnings between men and women after accounting
for the influence of other variables in the model. We also analyzed both
men’s and women’s earnings in separate regressions and applied a
frequently used decomposition method to the results to identify the
important factors leading to earnings differences by gender.
This appendix provides information on (1) our findings from a review of
previous research on earnings of men and women, (2) the data we used in
our analysis, (3) the econometric model we developed, (4) the results from
our model, and (5) the limitations of our analysis.
Our literature search consisted primarily of research in peer reviewed
journals, chiefly in economics, sociology, and psychology. We
concentrated on research about gender-related earnings differences, as
opposed to, for example, race-related or age-related earnings differences.
We focused on studies of populations within the United States,
particularly, but not limited to, studies using the Panel Study of Income
Dynamics (PSID)1 or the Current Population Survey (CPS) databases, and
studies conducted within the past 10 years. We also included any seminal
work in the area. We reviewed each study’s primary methodological
approach (whether it used cross-sectional or panel data and whether it
used general regression, time series, or other analytic estimation
methods), the specific databases used, the years included in the study, the
key variables in the analysis, and the principal results.
To study earnings differences, most of the studies we reviewed estimated
a wage or earnings equation that relates individuals’ wages or earnings to
several independent variables, such as education, experience, occupation,
1The PSID is a longitudinal survey, ongoing since 1968, of a representative sample of U.S.
individuals and the families they reside in. The central focus of the data is economic and
demographic, with substantial detail on income sources and amounts, employment, family
composition changes, and residential location. PSID data were collected annually through
1997 and biennially starting in 1999. The most recent survey available is 2001, which
includes data from 2000.
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Review of Other
Research on Earnings
Differences

Page 25
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 22
GAO-04-35 Women's Earnings
industry, and region. In contrast to simple comparisons between the
average wages or earnings of men and women, these studies attempted to
determine whether a wage or earnings difference existed after accounting
for differences between men and women in these variables.
The wage or earnings difference between men and women can be
identified in two ways. Studies that pool data for men and women together
can include a variable denoting the gender of the individuals. In a
multivariate regression analysis, the coefficient on the gender variable
represents the difference in earnings between men and women, holding
constant the effects of the other variables. Alternatively, separate
regression models can be estimated for men and women and a
decomposition analysis can compare the results for the two genders.
Our review of the literature did not uncover much disagreement over the
existence of an earnings difference after holding constant the effects of
other variables. Rather, debate centered on the size of any difference and
factors that might explain it. We found that the size of a difference can
vary by model estimation procedures, the years included in the analysis,
and the data set used. The wage or earnings difference, after controlling
for several factors, varied from 2.5 percent to 47.5 percent. Few of the
studies used data more recent than the mid-1990s.
The results of some studies on wage and earnings differences used
ordinary least squares (OLS) regressions for analysis. Compared to
analyses of uncontrolled wage and earnings data, OLS regression is an
improvement because it allows for the control of some factors in the data.
The strength of findings from OLS approaches has been questioned,
however, because of at least three potentially significant biases.2 First, the
estimates can be biased if some factors that are related to individuals’
earnings and that differ between men and women are omitted from the
analysis (omitted variable bias or unobserved heterogeneity). Second,
several of the independent variables may be closely interrelated with
earnings (endogeneity). For example, earnings may be related to the
number of hours an individual works, but the number of hours one
chooses to work may depend on how much is earned by working. An OLS
analysis assumes that no such interrelationships exist. If they do exist,
OLS can produce biased estimates. Third, in the context of individuals’
2Moon-Kak Kim and Solomon W. Polachek, “Panel Estimates of Male-Female Earnings
Functions,” Journal of Human Resources 29:2 (1994): 406–28.

Page 26
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 23
GAO-04-35 Women's Earnings
work decisions, OLS estimation can produce biased estimates when
unobserved factors affect both the level of earnings and the probability
that someone chooses to work (selection bias).
To conduct our analysis, we used the PSID rather than the CPS for two
main reasons. First, by using data that follow individuals over a period of
time, we can take into account individual work and life histories more
specifically than CPS or other data sources. Several researchers have
analyzed gender wage and earnings differences and have attempted to
address potential unobserved heterogeneity bias using longitudinal data
such as the PSID. Second, the PSID includes questions that can be used to
measure actual past work experience, which may be a key factor in
explaining the gender earnings difference but is not available in the CPS.
We assessed the reliability of the PSID data by reviewing documentation
and performing electronic tests in order to check for missing data,
outliers, or other potential problems that might adversely affect our
estimates. Based on these tests we determined that the data were
sufficiently reliable for the purposes of our work.
In our sample, individuals between the ages of 25 and 65 were tracked
from 1983 to 2000.3 Data for some individuals were available for all of
these years, while data for other individuals were available for some years
only. This is because some individuals entered the sample after 1983.
Individuals were not included in the sample until they formed an
independent household and reached age 25. We did not use data on
individuals after they reached age 65.
The dependent variable we focused on is a measure of an individual’s
annual earnings. As measured in the PSID, annual earnings include an
individual’s wages and salaries as well as income from bonuses, overtime
pay, tips, commissions, and other job-related income. It also includes
earnings from self-employment and farm-related income. We took inflation
into account by using the consumer price index to adjust annual earnings
to year 2000 dollars. We also developed an alternative definition of
earnings for individuals who reported that they were “self-employed only”
in a particular industry. For these individuals, we multiplied annual hours
worked by the average hourly earnings for the particular industry they
3The lower limit of the age range was set at 25 because the PSID does not include detailed
information for dependent college students, posing potential selection bias issues.
Data Used in Our
Analysis

Page 27
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 24
GAO-04-35 Women's Earnings
worked in using U.S. Department of Labor and U.S. Department of
Agriculture data.4
To determine why an earnings difference between men and women may
exist, our model controlled for a range of variables, which can be grouped
into three variable sets. The first set of independent variables consisted of
demographic characteristics, including gender, age, and race. We also
included an education variable that indicated the highest number of years
of education each respondent attained by the end of the sample period.
Family-related demographic variables included marital status, number of
children, and the age of the youngest child in the household. We also
included other income (defined as family income minus a respondent’s
own personal earnings), the region where individuals lived (i.e., in the
South or not), and whether they lived in a rural or urban area (i.e., in a
metropolitan area or not).
The second set of independent variables pertained to past work
experience. Total work experience was defined as the actual number of
years an individual worked for money since age 18. This variable was
computed as self-reported experience as reported in 1984 (or the year the
individual entered the panel), augmented by hours of work divided by
2,000 in each subsequent year. We also included a variable measuring job
tenure, defined as the length of time an individual had spent in his or her
current job.
The third set of independent variables included labor market activity
reported in a given survey year. Variables included hours worked in the
past year, weeks out of the labor force in the past year, and weeks
unemployed in the past year. For our analysis, we considered time spent
unemployed and time out of the labor force as work “interruptions,” but
we did not include time off for one’s own illness or a family member’s
illness, vacation and other time off, or time out because of strike. We also
included a variable that accounted for an individual’s full-time or part-time
employment status, defined as the average number of hours an individual
worked per week on his or her main job. Individuals were considered to
have worked part-time if they worked fewer than 35 hours per week and
full-time if they worked 35 hours or more per week. Other variables in this
4The Department of Agriculture data are from the National Agricultural Statistics Service
data series “Annual All Hired Workers Wage Rates, U.S. Level” and the Department of
Labor data are from the Bureau of Labor Statistics data series “Average Hourly Earnings of
Production Workers.”

Page 28
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 25
GAO-04-35 Women's Earnings
category included the individual’s industry, occupation, and an indicator of
union membership. We also accounted for self-employment status, defined
as whether respondents worked for someone else, for themselves, or for
both themselves and someone else. Table 1 shows descriptive statistics for
selected PSID data used in our analysis.

Page 29
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 26
GAO-04-35 Women's Earnings
Table 1: Descriptive Statistics for Selected PSID Variables
Men
Women
Variable
Means
(averages)
Standard
deviation
Means
(averages)
Standard
deviation
All individuals (workers and nonworkers)
Annual earnings (in 2000 dollars)
35,942
34,630
16,554
18,510
Age of individual (in years)
41.3
11.3
42.0
11.5
Age of youngest child (in years)
3.3
4.9
4.0
5.2
Number of children
0.9
1.2
1.1
1.2
Married (percent)
70.1
45.8
61.2
48.7
Metropolitan area of residence
(percent)
64.7
48.1
67.1
47.0
Full-time main job (percent)
74.9
43.3
47.2
49.9
Time unemployed (in weeks)
1.9
7.0
1.8
6.9
Time out of the labor force (in
weeks)
2.4
9.9
6.1
15.3
Annual hours worked
1,931
926
1,226
957
Job tenure (in months)
80.1
102.2
55.1
80.3
Work experience (in years)
16.8
10.2
11.2
8.4
Highest education (in years)
12.9
2.7
12.7
2.4
Number of observations
42,394
54,986
Number of individuals
5,032
6,033
Workers only
Annual earnings (in 2000 dollars)
40,426
34,334
22,782
18,316
Age of individual (in years)
40.2
10.6
40.4
10.5
Age of youngest child (in years)
3.5
5.0
4.3
5.2
Number of children
1.0
1.2
1.0
1.2
Married (percent)
72.2
44.9
60.9
48.8
Metropolitan area of residence
(percent)
64.5
47.8
68.1
46.6
Full-time main job (percent)
87.6
33.0
66.8
47.1
Time unemployed (in weeks)
1.8
6.4
1.9
6.7
Time out of the labor force (in
weeks)
0.91
5.1
2.8
9.1
Annual hours worked
2,154
697
1,672
716
Job tenure (in months)
89.3
104.2
74.1
85.6
Work experience (in years)
16.4
9.8
12.1
8.0
Highest education (in years)
13.2
2.6
13.1
2.3
Number of observations
35,726
36,793

Page 30
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 27
GAO-04-35 Women's Earnings
Men
Women
Variable
Means
(averages)
Standard
deviation
Means
(averages)
Standard
deviation
Number of individuals
4,477
4,884
Source: GAO analysis of PSID data.
We used the Hausman-Taylor model to analyze the earnings difference
between men and women.5 The Hausman-Taylor model was developed to
analyze panel data and to take into account unobserved heterogeneity and
endogeneity while permitting the estimation of coefficients for factors that
do not vary over time, such as gender. As is usual practice in studies of the
determinants of earnings and earnings differences between groups, we
related the natural logarithm of the dependent variable (annual earnings in
this case) to several independent variables. The specific equation we
estimated was
5Jerry A. Hausman and William E. Taylor, “Panel Data and Unobservable Individual
Effects,” Econometrica 49:6 (November 1981). Light and Ureta use this model to analyze
the relationship between experience and wage differences (see Audrey Light and Manuelita
Ureta, “Early-Career Work Experience and Gender Wage Differentials,” Journal of Labor
Economics 13:1 (1995): 121-154).
Description of Our
Econometric Model
ln (real earningsit) = X1itβ1 + X2itβ2 + Z1iδ1 + Z2iδ2 + µi + νit
where subscripts i and t denote individuals and time periods,
X1it are exogenous time-varying variables assumed to be uncorrelated with µi
and νit,
X2it are endogenous time-varying variables possibly correlated with µi but not
with νit,
Z1i are exogenous time-invariant variables assumed to be uncorrelated with
µi and νit,

Page 31
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 28
GAO-04-35 Women's Earnings
In our specification of the model, we allowed annual hours worked, time
out of labor force, work experience, and the square of experience to be
time-varying endogenous variables. Highest education achieved was
treated as a time-invariant endogenous variable. The other independent
variables were treated as exogenous.
To account for possible selection bias arising from not accounting for an
individual’s choice of whether to work, we used a Heckman selection bias
correction. To do this, we estimated the probability of working in a
particular year for all individuals in the data set.6 We then used a term that
was estimated in this equation (the inverse Mills ratio) as an additional
independent variable in the Hausman-Taylor earnings equation. The
Hausman-Taylor model was then estimated for individuals with positive
annual hours of work and positive earnings in a given year.
Two academic labor economists reviewed a preliminary version of the
econometric model and the results. One of the reviewers has published
extensively on gender wage differences and has used the PSID in his work.
The other reviewer has published widely on labor economics topics
generally, also using the PSID. Both reviewers thought that the model and
results were sound and reasonable. To the extent possible, we have
incorporated their suggestions for clarifications and additional analysis.
6The probability that an individual worked was modeled as a function of age, the number of
children and the age of the youngest child in the household, marital status, additional
family income, work experience, education, race, region and urban-rural indicators, and a
work disability indicator. This model was estimated separately for men and women for
each of the years in the sample.
Z2i are endogenous time-invariant variables possibly correlated with µi but
not with νit,
β and δ represent coefficients on the respective variables,
µi is an individual-specific random error term designed to take unobserved
individual heterogeneity into account, and
νit is a random error term.

Page 32
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 29
GAO-04-35 Women's Earnings
We found that before controlling for any variables that may affect
earnings, on average, women earned about 44 percent less than men over
the time period we studied—1983 to 2000. However, after controlling for
the independent variables that we included in our model, we found that
this difference was reduced to about 21 percent over this time period. The
model results indicated a small but statistically significant decline in the
earnings difference over this period.
Table 2 shows the regression results for the overall model that included
observations on men and women combined and the results for men and
women separately. For each variable in each regression, the table shows
the coefficient (estimate β), the estimated standard error for the
coefficient, the p-value, and an alternative coefficient estimate. For each of
the regressions, the first column of results shows the coefficient estimates.
The standard interpretation of the regression coefficients in models of this
type is that they represent the average percentage change in earnings that
would result from a small increase in an independent variable. The
estimated standard error and the p-value are shown in the second and
third columns. A p-value of less than 0.05 indicates that the regression
coefficient is statistically significantly different from zero, which would
indicate that the variable has a statistically significant effect on earnings.
In the fourth column, we show an alternative estimate for the average
percentage change based on a transformation of the regression
coefficients, which the literature shows is a more precise measure than the
standard coefficient estimate.7 For this reason, we emphasize the
alternative estimates in the discussion of the results.
The gender coefficient in the overall model shows the difference in
earnings between men and women in each year after accounting for the
effect of the other variables in the model. As shown in the alternative
estimate column of the overall model results of table 2, the estimated
coefficient for the gender variable was –0.2025 for the year 2000. This
means that, holding all other variables in the model constant except for
gender, women earned an average of about 20.3 percent less than men in
2000. The estimated coefficients were statistically significantly different
from zero for each of the years. Overall, the model results indicated that
there was a small but statistically significant decline in the earnings
7Peter E. Kennedy, “Estimation with Correctly Interpreted Dummy Variables in
Semilogarithmic Equations,” American Economic Review, 71:4 (September 1981): 801. The
alternative estimator g = exp(β – ½ V(β)) – 1, where V(β) is the estimated variance of the
regression coefficient.
Results of Our
Analysis

Page 33
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 30
GAO-04-35 Women's Earnings
difference between 1983 and 2000. The analysis indicated that the
difference declined by about 0.3 percentage points per year, on average.
The next set of variables, included in the overall model and in the separate
regressions for men and women, deal with work patterns. In our analysis,
work patterns included years of work experience, hours worked per year,
length of time out of the labor force, and whether the individual worked a
full-time or part-time schedule. In addition, length of unemployment and
tenure were also considered to be work patterns. For the hours worked,
time out of the labor force, length of unemployment, and tenure variables,
the coefficient estimate shown represents the estimated percentage
change in earnings that would result from a one-unit change (hours or
weeks) in the particular variable. For example, as shown in table 2 in the
alternative estimate column of the overall model results, the coefficient for
time out of the labor force was –0.0226. This means that earnings would
decrease by about 2.3 percent for each additional week out of the labor
force, holding all other factors constant—including annual hours worked.
The coefficients on the experience variables indicate that each additional
year of work experience is generally associated with increased earnings,
but this increase declines as the level of experience increases.8 The
working full-time variable measures the effect of having a full-time main
job relative to having a part-time job as a main job. All the work pattern
variables are estimated to have a statistically significant effect on earnings.
The next set of variables includes other work-related characteristics.
Several of these variables are categorical in nature, such as occupation,
industry, and self-employment status. For these variables, the coefficient
for a particular category is an estimate of the effect of being in that
category relative to the omitted category. For example, as shown in
table 2 in the alternative estimate column of the overall model results, the
coefficient was -0.09 for those individuals working in service/private
household occupations. This indicates that individuals working in
service/private household occupations earned 9 percent less, on average,
8The effect of an additional year of experience on earnings is the sum of the effect of the
experience and experience-squared variables. The amount that an additional year of
experience will increase the value of the experience-squared variable will vary with the
level of experience. For example, an additional year of experience would increase
experience-squared by 1 for someone with no prior experience, and it will increase the
experience-squared variable by 41 for someone with 20 years of experience
(i.e., 441 – 400 = 41). Taking into account the effect of both variables, these estimates
would indicate that an additional year of experience would increase earnings for men with
less than 33 years of experience and for women with less than 31 years of experience.

Page 34
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 31
GAO-04-35 Women's Earnings
than individuals working in professional and technical occupations (the
omitted occupation category), holding all other variables in the model
constant. On the other hand, nonfarm managers and administrators earned
about 2.5 percent more, on average, than professional and technical
workers, holding other factors constant.
Also shown in table 2 are coefficients for demographic variables and other
independent variables that were included in the model, such as age of
individual, age of youngest child, number of children, metropolitan area,
marital status, and region. Several of the coefficients in this category, such
as age of youngest child and number of children, were not found to be
statistically significant in the overall model. However, other coefficients
were statistically significant, such as age of individual, living in a
metropolitan area, living in the South, being married, and being black. For
example, in table 2 in the alternative estimate column of the overall model
results, the coefficient for living in a metropolitan area was 0.0229. This
means that individuals living in a metropolitan area were estimated to earn
about 2.3 percent more than those living in non-metropolitan areas, and
this difference was statistically significant. Also, according to the model,
individuals living in the South were estimated to earn about 4.2 percent
less than those not living in the South, and this difference was statistically
significant.
Table 2 also shows the regression results of the separate analysis of men
and women. Most of the variables had coefficients that were both positive
or both negative for men and women, indicating that the variables affected
earnings in the same direction. This is the case for all work pattern
variables. For example, as shown in table 2 in the alternative estimate
columns for men and women, the estimated coefficients for the work
experience variable were positive for men and women (0.0264 and 0.0249
respectively) and the coefficient for the square of work experience is
negative for both men and women. As discussed above, earnings for both
men and women generally increase with additional experience, but that
increase declines the higher the level of work experience (for example, the
gain between the fifth and sixth year of work experience is larger than
between the 25th and 26th year of work experience). Estimated coefficients
for other variables were also negative for both men and women. For
example, as shown in table 2 in the alternative estimate columns for men
and women separately, the coefficients for black individuals (relative to
white—the omitted category) were as follows: -0.1385 for men and
–0.0661 for women. This means that black men earned about 13.9 percent
less than white men, while black women earned about 6.6 percent less
than white women.

Page 35
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 32
GAO-04-35 Women's Earnings
The relationship between earnings and number of children is one example
where the coefficients are not of the same sign. As shown in table 2 in the
overall model results for men and women combined, the coefficient on the
number of children variable was statistically insignificant. However, in the
separate regression analysis of men and women, number of children was
associated with about a 2.1 percent increase in earnings for men and about
a 2.5 percent decrease for women, with both estimates being significant. In
addition, married men earned about 8.3 percent more than never married
men, while the earnings difference between married and never married
women was statistically insignificant.

Page 36
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 33
GAO-04-35 Women's Earnings

Page 37
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 34
GAO-04-35 Women's Earnings
Table 2: Overall and Separate Model Results for Men and Women
Overall model
Variable
Estimate β Standard error
p-value
Alternative
estimate g
Gender: women vs. men
2000
–0.2260
0.0227
0.000
-0.2025
1999a
1998
–0.1716
0.0229
0.000
-0.1579
1997a
1996
–0.2264
0.0230
0.000
-0.2028
1995
–0.2176
0.0215
0.000
-0.1958
1994
–0.2311
0.0213
0.000
-0.2065
1993
–0.2132
0.0214
0.000
-0.1922
1992
–0.2556
0.0210
0.000
-0.2257
1991
–0.2478
0.0209
0.000
-0.2197
1990
–0.2277
0.0209
0.000
-0.2038
1989
–0.2315
0.0209
0.000
-0.2068
1988
–0.2534
0.0210
0.000
-0.2240
1987
–0.2503
0.0211
0.000
-0.2216
1986
–0.2708
0.0210
0.000
-0.2374
1985
–0.2810
0.0212
0.000
-0.2452
1984
–0.2921
0.0212
0.000
-0.2534
1983
–0.2179
0.0222
0.000
-0.1960
Work patterns
Experience
(years)
0.0231
0.0019
0.000
0.0234
Experience
squared
–0.0003
0.0000
0.000
-0.0003
Hours worked
(per year)
0.0004
0.0000
0.000
0.0004
Time out of
labor force
(weeks)
-0.0228
0.0003
0.000
-0.0226
Length of
unemployment
(weeks)
–0.0156
0.0004
0.000
-0.0155
Tenure
(months)
0.0009
0.0000
0.000
0.0009

Page 38
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 35
GAO-04-35 Women's Earnings
Men
Women
Estimate βm
Standard
error
p-value
Alternative
estimate gm
Estimate βf
Standard
error
p-value
Alternative
estimate gf
0.0260
0.0025
0.000
0.0264
0.0246
0.0031
0.000
0.0249
–0.0004
0.0000
0.000
-0.0004
–0.0004
0.0001
0.000
-0.0004
0.0003
0.0000
0.000
0.0003
0.0005
0.0000
0.000
0.0005
–0.0175
0.0006
0.000
-0.0174
–0.0224
0.0004
0.000
-0.0222
–0.0171
0.0005
0.000
-0.0170
–0.0143
0.0005
0.000
-0.0142
0.0010
0.0000
0.000
0.0010
0.0009
0.0001
0.000
0.0009

Page 39
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 36
GAO-04-35 Women's Earnings
Overall model
Variable
Estimate β
Standard
error
p-value
Alternative
estimate g
Working full time
(main job)
0.1519
0.0063
0.000
0.1640
Other work related
Mother’s education
–0.0194
0.0057
0.001
-0.0193
Father’s education
–0.0044
0.0051
0.385
-0.0044
Highest education
(years)
0.1475
0.0058
0.000
0.1590
Self-employment
status
Works for
someone else
onlyb
Self-employed
only
0.0142
0.0103
0.166
0.0142
Missing
–0.3272
0.0128
0.000
-0.2791
Both
0.0191
0.0239
0.424
0.0190
Union member
0.1435
0.0090
0.000
0.1542
Occupation
Professional,
technicalb
Service/private
household
workers
–0.0949
0.0116
0.000
-0.0906
Farm laborers
and foremen
–0.1761
0.0399
0.000
-0.1622
Farmers and farm
management
–0.3805
0.0469
0.000
-0.3172
Nonfarm laborers
–0.0907
0.0162
0.000
-0.0869
Transport
equipment
operators
–0.0869
0.0179
0.000
-0.0834
Operators,
nontransport
–0.0588
0.0136
0.000
-0.0572
Craftsmen
–0.0108
0.0122
0.376
-0.0108

Page 40
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 37
GAO-04-35 Women's Earnings
Men
Women
Estimate βm
Standard
error p-value
Alternative
estimate gm
Estimate βf
Standard
error
p-value
Alternative
estimate gf
0.1724
0.0094
0.000
0.1881
0.1180
0.0086
0.000
0.1252
–0.0107
0.0075
0.155
-0.0106
–0.0256
0.0081
0.001
-0.0253
0.0039
0.0067
0.557
0.0039
–0.0117
0.0071
0.102
-0.0116
0.1355
0.0072
0.000
0.1451
0.1603
0.0087
0.000
0.1738
–0.1056
0.0123
0.000
-0.1003
0.2168
0.0169
0.000
0.2419
–0.2823
0.0187
0.000
-0.2461
–0.3413
0.0175
0.000
-0.2892
0.0506
0.0266
0.057
0.0516
–0.0846
0.0443
0.056
-0.0820
0.1388
0.0113
0.000
0.1488
0.1405
0.0140
0.000
0.1507
–0.1061
0.0176
0.000
-0.1008
–0.0975
0.0158
0.000
-0.0930
–0.1928
0.0422
0.000
-0.1761
–0.0602
0.0850
0.479
-0.0618
–0.3434
0.0479
0.000
-0.2915
–0.1690
0.1156
0.144
-0.1611
–0.0823
0.0178
0.000
-0.0791
–0.0627
0.0380
0.099
-0.0615
–0.0576
0.0192
0.003
-0.0562
–0.1840
0.0468
0.000
-0.1690
–0.0458
0.0168
0.007
-0.0449
–0.0657
0.0217
0.003
-0.0638
0.0016
0.0138
0.909
0.0015
–0.0180
0.0290
0.534
-0.0183

Page 41
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 38
GAO-04-35 Women's Earnings
Overall model
Variable
Estimate β
Standard
error
p-value
Alternative
estimate g
Clerical workers
–0.0438
0.0104
0.000
-0.0429
Sales workers
–0.0718
0.0145
0.000
-0.0694
Nonfarm
managers,
administrators
0.0243
0.0100
0.015
0.0246
Do not
know/missing
–0.1329
0.0280
0.000
-0.1248
Industry
Wholesale/retail
tradeb
Public
administration
0.0702
0.0147
0.000
0.0726
Professional
services
0.0516
0.0107
0.000
0.0529
Entertainment
–0.0378
0.0275
0.168
-0.0375
Personal services
0.0172
0.0156
0.270
0.0172
Business and
repair services
0.0561
0.0129
0.000
0.0576
Finance,
insurance, real
estate
0.1081
0.0149
0.000
0.1141
Transportation/
communications/
public utilities
0.1692
0.0145
0.000
0.1842
Manufacturing
0.1369
0.0104
0.000
0.1467
Construction
0.1472
0.0150
0.000
0.1584
Mining/agriculture
0.0303
0.0234
0.195
0.0305
Do not
know/missing
0.0835
0.0251
0.001
0.0868
Mills ratio
–0.2834
0.0218
0.000
-0.2470
Demographic and
other controls
Age of individual
(years)
–0.0023
0.0011
0.043
-0.0023
Age of youngest
child (years)
0.0006
0.0005
0.257
0.0006

Page 42
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 39
GAO-04-35 Women's Earnings
Men
Women
Estimate βm
Standard
error p-value
Alternative
estimate gm
Estimate βf
Standard
error
p-value
Alternative
estimate gf
–0.0608
0.0178
0.001
-0.0592
–0.0497
0.0138
0.000
-0.0486
–0.0343
0.0187
0.066
-0.0339
–0.0931
0.0218
0.000
-0.0891
0.0373
0.0125
0.003
0.0379
0.0165
0.0157
0.295
0.0165
–0.1107
0.0370
0.003
-0.1054
–0.1276
0.0414
0.002
-0.1205
0.0104
0.0183
0.571
0.0102
0.1641
0.0233
0.000
0.1780
0.0172
0.0164
0.294
0.0172
0.0707
0.0146
0.000
0.0731
0.0044
0.0337
0.896
0.0039
–0.0756
0.0436
0.083
-0.0737
–0.0307
0.0301
0.308
-0.0306
–0.0097
0.0196
0.623
-0.0098
0.0705
0.0158
0.000
0.0729
0.0488
0.0208
0.019
0.0498
0.0562
0.0219
0.010
0.0575
0.1489
0.0202
0.000
0.1604
0.1713
0.0163
0.000
0.1867
0.1865
0.0280
0.000
0.2046
0.1417
0.0126
0.000
0.1521
0.1332
0.0174
0.000
0.1423
0.1708
0.0160
0.000
0.1861
0.0673
0.0384
0.079
0.0689
0.0481
0.0247
0.051
0.0489
0.0178
0.0517
0.730
0.0166
0.1106
0.0323
0.001
0.1164
0.0712
0.0378
0.060
0.0730
–0.3307
0.0285
0.000
-0.2819
–0.1584
0.0352
0.000
-0.1470
–0.0016
0.0019
0.394
-0.0016
–0.0058
0.0015
0.000
-0.0057
–0.0013
0.0007
0.048
-0.0013
0.0023
0.0007
0.003
0.0023

Page 43
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 40
GAO-04-35 Women's Earnings
Overall model
Variable
Estimate β
Standard
error
p-value
Alternative
estimate g
Number of
children
0.0004
0.0029
0.897
0.0004
Additional family
income (inflation
adjusted in
thousands of
dollars)
–0.0006
0.0001
0.000
-0.0006
Metropolitan area
0.0226
0.0067
0.001
0.0229
Excellent health
0.0088
0.0057
0.123
0.0089
Marital status
Never marriedb
Married
0.0403
0.0113
0.000
0.0410
Other
0.0245
0.0127
0.053
0.0247
Region: South
–0.0428
0.0120
0.000
-0.0420
Race
Whiteb
Black
–0.1031
0.0171
0.000
-0.0981
Other
0.0739
0.0585
0.207
0.0748
Year, compared to
1983
2000
0.0410
0.0191
0.032
0.0417
1999a
1998
–0.0223
0.0187
0.233
-0.0222
1997a
1996
–0.0837
0.0187
0.000
-0.0804
1995
–0.0705
0.0177
0.000
-0.0682
1994
–0.0794
0.0170
0.000
-0.0764
1993
–0.0664
0.0168
0.000
-0.0643
1992
–0.0477
0.0161
0.003
-0.0467
1991
–0.0867
0.0157
0.000
-0.0832
1990
–0.0839
0.0154
0.000
-0.0806
1989
–0.0569
0.0151
0.000
-0.0555
1988
–0.0277
0.0149
0.064
-0.0274
1987
–0.0318
0.0148
0.031
-0.0314
1986
–0.0205
0.0146
0.160
-0.0204

Page 44
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 41
GAO-04-35 Women's Earnings
Men
Women
Estimate βm
Standard
error
p-value
Alternative
estimate gm
Estimate βf
Standard
error
p-value
Alternative
estimate gf
0.0210
0.0037
0.000
0.0212
–0.0254
0.0047
0.000
-0.0251
–0.0009
0.0001
0.000
-0.0009
–0.0001
0.0001
0.403
-0.0001
0.0171
0.0086
0.047
0.0173
0.0305
0.0102
0.003
0.0309
0.0149
0.0072
0.038
0.0150
0.0062
0.0088
0.483
0.0062
0.0800
0.0142
0.000
0.0831
–0.0011
0.0176
0.950
-0.0013
0.0685
0.0162
0.000
0.0707
–0.0009
0.0192
0.962
-0.0011
–0.0522
0.0155
0.001
-0.0510
–0.0377
0.0173
0.030
-0.0371
–0.1487
0.0242
0.000
-0.1385
–0.0682
0.0230
0.003
-0.0661
0.0491
0.0843
0.560
0.0466
0.0972
0.0762
0.202
0.0989
0.0188
0.0192
0.328
0.0188
0.0621
0.0222
0.005
0.0638
–0.0406
0.0186
0.029
-0.0399
0.0298
0.0215
0.165
0.0300
–0.1045
0.0185
0.000
-0.0994
–0.0733
0.0205
0.000
-0.0709
–0.0813
0.0175
0.000
-0.0782
–0.0618
0.0194
0.001
-0.0601
–0.0973
0.0167
0.000
-0.0928
–0.0759
0.0188
0.000
-0.0733
–0.0854
0.0165
0.000
-0.0820
–0.0495
0.0184
0.007
-0.0484
–0.0693
0.0156
0.000
-0.0671
–0.0625
0.0180
0.001
-0.0608
–0.1023
0.0150
0.000
-0.0974
–0.0921
0.0180
0.000
-0.0881
–0.0960
0.0146
0.000
-0.0917
–0.0737
0.0174
0.000
-0.0712
–0.0691
0.0142
0.000
-0.0669
–0.0524
0.0171
0.002
-0.0512
–0.0359
0.0140
0.010
-0.0354
–0.0516
0.0169
0.002
-0.0504
–0.0389
0.0137
0.005
-0.0383
–0.0561
0.0165
0.001
-0.0546
–0.0248
0.0135
0.066
-0.0246
–0.0632
0.0164
0.000
-0.0613

Page 45
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 42
GAO-04-35 Women's Earnings
Overall model
Variable
Estimate β
Standard
error
p-value
Alternative
estimate g
1985
–0.0249
0.0145
0.086
-0.0247
1984
–0.0219
0.0144
0.127
-0.0218
Intercept
7.4055
0.0783
0.000
7.4055

Page 46
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 43
GAO-04-35 Women's Earnings
Men
Women
Estimate βm
Standard
error p-value
Alternative
estimate gm
Estimate βf
Standard
error
p-value
Alternative
estimate gf
–0.0282
0.0134
0.035
-0.0279
–0.0822
0.0163
0.000
-0.0791
–0.0237
0.0131
0.070
-0.0235
–0.0847
0.0160
0.000
-0.0813
7.5910
0.0983
0.000
7.5910
6.9846
0.1179
0.000
6.9846
Source: GAO analysis of PSID data.
aData not available.
bCategory omitted.
Tables 3, 4, and 5 show a decomposition analysis of the earnings
difference derived from the separate regression analysis for men and
women. This statistical technique—the Blinder-Oaxaca decomposition—
has been commonly used in analyses of wage or earnings differences
between men and women. The decomposition divides the (logged)
earnings difference between men and women into two parts: a part
reflecting differences in characteristics between men and women and a
part reflecting differences in parameters (or return to earnings) between
men and women.9 This decomposition is represented as follows:
We estimated the logged earnings difference between men and women
from 1983 and 2000 to be approximately 0.69 (i.e. the left hand side of the
equation above). The analysis showed that about two-thirds of this
difference, or 0.45 out of 0.69, reflected differences between men and
women’s characteristics (the first term on the right hand side of the
equation). The remaining one-third, about 0.24 out of 0.69, reflected
differences in parameters, i.e., how the variables affected earnings
9J. G. Altonji and R. M. Blank, “Race and Gender in the Labor Market,” The Handbook of
Labor Economics (Amsterdam: Elsevier Science, 1999), vol. 3C, pp. 3153–61.
where Xm and Xf represent the mean values of the independent variables for
men and women, respectively, and βm and βf are the estimated regression
coefficients for men and women for all the variables.
ln Em − ln Ef = (Xm Xf)´βm + Xf´m − βf)
ˆ
ˆ
ˆ

Page 47
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 44
GAO-04-35 Women's Earnings
differently for men and women (the second term on the right hand side of
the equation).
Table 3 summarizes how several categories of variables contributed to the
earnings difference through differences in characteristics and differences
in parameters. Positive values indicate an earnings advantage for men
while negative values indicate an advantage for women. For example, in
table 3, the difference in earnings due to characteristics from the work
pattern variables is equal to 0.2729, which indicates that men have an
earnings advantage. This figure represents the sum—for all the work
pattern variables—of the difference in men’s and women’s mean
characteristics multiplied by the men’s regression coefficients. The effect
of the work pattern variables accounted for most of the difference in
characteristics between men and women (due to different characteristics:
about 0.27 out of 0.45). Relatively little of the earnings difference was
attributable to differences in demographic characteristics (about 0.03 out
of 0.45).
Table 3 also shows the differences in earnings due to differences in
parameters (0.2446 in the total row at the bottom of table 3). The table
shows that women have a relative advantage due to parameters from the
work pattern variables. In the table, -0.2302 represents the sum—for all the
work pattern variables—of the difference in men and women’s parameters
multiplied by the women’s mean value of the variable. Women’s
advantages in the work pattern and other work-related variable categories
are outweighed by disadvantages due to the parameters for demographic
factors and from the intercept of the regressions. The relatively large
advantage to men in the intercepts of the regressions indicates that a
predictable earnings difference remains even after taking differences in
characteristics and relative returns into account.
This second part of the decomposition allows us to describe how the
remaining earnings difference results from how each factor affects
earnings differently for men and women. According to Altonji and Blank,
this component is often mistakenly attributed to the “share due to
discrimination” but actually “captures both the effects of discrimination
and unobserved differences in productivity and tastes.”10 They also point
out that it may be misleading to label only this second component as the
result of discrimination, since discriminatory barriers in the labor market
10Altonji and Blank, p. 3156.

Page 48
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 45
GAO-04-35 Women's Earnings
and elsewhere in the economy can affect the mean values of the
characteristics.
Table 3: Summary of Decomposition Results
Differences in earnings
Variable categories
Due to
characteristics
Due to
parameters
Work patternsa
0.2729
-0.2302
Other work relatedb
0.1539
-0.3218
Demographic and other controlsc
0.0272
0.1902
Intercept
N/A
0.6065
Total
0.4540
0.2446
Source: GAO Analysis of PSID data.
Note: These summary results are based on the more detailed analysis shown in table 4.
aThe work patterns category includes: work experience (years), experience squared, time out of the
labor force (weeks), length of unemployment (weeks), working full time (main job), tenure (months),
and hours worked (per year).
bThe other work related category includes: highest education (years), mother’s education, father’s
education, self-employment status, union membership, industry, occupation, and the Mill’s ratio.
cThe demographic and other controls category includes all other variables, except the intercept, which
is a parameter only.
Table 4 shows more detailed decomposition results.11 In table 4 in the
column labeled difference due to characteristics, the variables measuring
work patterns, including experience (0.108), hours worked (0.134),
working full-time versus part-time (0.036), and length of time out of the
labor force (0.034), made large contributions to explaining gender
differences in earnings. Table 4 shows that, on average, men in our sample
worked about 2,147 hours per year, women about 1,675 hours per year.
The analysis showed that the difference between men and women, based
on hours worked, resulted in a relative advantage for men of about
0.134. In other words, about one-fifth of the uncontrolled logged earnings
difference (0.134 out of 0.69) results from the greater number of hours
men worked compared to women.
11Table 5 uses the alternative estimates reported in table 2. Because the alternative
estimates are a transformation of the regression coefficients, the sum of the differences
due to characteristics and parameters need not sum to the total difference in logged
earnings as it does in the standard decomposition.

Page 49
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 46
GAO-04-35 Women's Earnings
Table 4 also shows how the variables affected earnings differently for men
and women. Positive values in the difference due to parameters column
would indicate that men would gain more from an increase in a particular
variable than would women. For example, compared to women, men
receive a greater estimated return to their earnings resulting from having
children. However, we found several large negative values indicating that
women have a relative advantage over men in terms of how other factors
affect earnings. The largest negative values in this column resulted from
the greater estimated return for each additional year of education and the
greater estimated return for an additional hour of work for women. As
mentioned above, the relative advantage for women for some of the
variables in the model is offset when the difference in the intercept terms
of the separate regressions is added. The difference in the intercept terms
captures gender differences and other unmeasured effects that we cannot
identify in the regressions.12
Table 4: Decomposition Results Using Regression Coefficients
Estimate
Means (averages)
Difference
Variable
Men
βm
Women
βf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) βm
Between
parameters
(βm βf)
Due to
parameters
(returns)
Xf (βmβf)
Work patterns
Experience (years)
0.0260 0.0246
16.2891
12.1342
4.1548
0.1081
0.0014
0.0170
Experience squared –0.0004 –0.0004 359.5914 210.6411
148.9504
–0.0558
0.0001
0.0120
Hours worked (per
year)
0.0003 0.0005 2,147.3100 1,674.8000
472.5100
0.1340
-0.0002
-0.3057
Time out of labor
force (weeks)
–0.0175 –0.0224
0.9262
2.8345
–1.9083
0.0335
0.0049
0.0139
Length of
unemployment
(weeks)
–0.0171 –0.0143
1.8149
1.8887
–0.0739
0.0013
-0.0028
-0.0054
Tenure (months)
0.0010 0.0009
91.4775
74.4278
17.0497
0.0163
0.0000
0.0015
Working full time (in
main job)
0.1724 0.1180
0.8761
0.6701
0.2059
0.0355
0.0543
0.0364
12Oaxaca and Ransom showed that the size of the intercept terms in decompositions is
sensitive to the choice of the omitted categorical variables used as reference groups in the
analysis. See Ronald L. Oaxaca and Michael R. Ransom, “Identification in Detailed Wage
Decompositions,” Review of Economics and Statistics 81:1(February 1999): 154–57.

Page 50
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 47
GAO-04-35 Women's Earnings
Estimate
Means (averages)
Difference
Variable
Men
βm
Women
βf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) βm
Between
parameters
(βm βf)
Due to
parameters
(returns)
Xf (βmβf)
Other work related
Mother’s education
–0.0107 –0.0256
3.5458
3.4941
0.0516
–0.0005
0.0150
0.0524
Father’s education
0.0039 –0.0117
3.3364
3.2447
0.0917
0.0004
0.0156
0.0506
Highest education
(years)
0.1355 0.1603
13.1455
13.0880
0.0575
0.0078
–0.0248
-0.3242
Self-employment
status
Works for some-
one else onlya
Self-employed only –0.1056 0.2168
0.1177
0.0579
0.0597
–0.0063
–0.3224
-0.0187
Missing
–0.2823 –0.3413
0.0648
0.1230
–0.0582
0.0164
0.0590
0.0073
Both
0.0506 –0.0846
0.0094
0.0042
0.0052
0.0003
0.1352
0.0006
Union member
0.1388 0.1405
0.1773
0.1187
0.0587
0.0081
–0.0017
-0.0002
Occupation
Professional,
technicala
Service/private
household workers –0.1061 –0.0975
0.0763
0.2034
–0.1271
0.0135
–0.0087
-0.0018
Farm laborers and
foremen
–0.1928 –0.0602
0.0121
0.0023
0.0098
–0.0019
–0.1326
-0.0003
Farmers and farm
management
–0.3434 –0.1690
0.0124
0.0008
0.0116
–0.0040
–0.1745
-0.0001
Nonfarm laborers
–0.0823 –0.0627
0.0547
0.0083
0.0464
–0.0038
–0.0195
-0.0002
Transport
equipment
operators
–0.0576 –0.1840
0.0680
0.0084
0.0596
–0.0034
0.1264
0.0011
Operators,
nontransport
–0.0458 –0.0657
0.0877
0.0879
–0.0002
0.0000
0.0198
0.0017
Craftsmen
0.0016 –0.0180
0.2049
0.0171
0.1879
0.0003
0.0196
0.0003
Clerical workers
–0.0608 –0.0497
0.0497
0.2565
–0.2068
0.0126
–0.0111
-0.0028
Sales workers
–0.0343 –0.0931
0.0469
0.0409
0.0059
–0.0002
0.0588
0.0024
Nonfarm
managers,
administrators
0.0373 0.0165
0.1609
0.0922
0.0687
0.0026
0.0208
0.0019
Do not
know/missing
–0.1107 –0.1276
0.0468
0.0906
–0.0439
0.0049
0.0169
0.0015

Page 51
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 48
GAO-04-35 Women's Earnings
Estimate
Means (averages)
Difference
Variable
Men
βm
Women
βf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) βm
Between
parameters
(βm βf)
Due to
parameters
(returns)
Xf (βmβf)
Industry
Wholesale/retail
tradea
Public
administration
0.0104 0.1641
0.0799
0.0607
0.0192
0.0002
–0.1538
-0.0093
Professional
services
0.0172 0.0707
0.1211
0.3467
–0.2256
–0.0039
–0.0535
-0.0186
Entertainment
0.0044 –0.0756
0.0095
0.0061
0.0034
0.0000
0.0800
0.0005
Personal services –0.0307 –0.0097
0.0130
0.0678
–0.0549
0.0017
–0.0210
-0.0014
Business and
repair services
0.0705 0.0488
0.0585
0.0340
0.0245
0.0017
0.0217
0.0007
Finance,
insurance, real
estate
0.0562 0.1489
0.0394
0.0641
–0.0248
–0.0014
–0.0928
-0.0059
Transportation/
communications/
public utilities
0.1713 0.1865
0.0976
0.0353
0.0622
0.0107
–0.0152
-0.0005
Manufacturing
0.1417 0.1332
0.2444
0.1341
0.1103
0.0156
0.0085
0.0011
Construction
0.1708 0.0673
0.0963
0.0101
0.0862
0.0147
0.1034
0.0010
Mining/agriculture
0.0481 0.0178
0.0474
0.0075
0.0399
0.0019
0.0302
0.0002
Do not
know/missing
0.1106 0.0712
0.0513
0.0954
–0.0441
–0.0049
0.0394
0.0038
Mills ratio
–0.3307 –0.1584
0.1628
0.3771
–0.2143
0.0709
–0.1723
-0.0650
Demographic and
other controls
Age of individual
(years)
–0.0016 –0.0058
40.1442
40.3309
–0.1867
0.0003
0.0041
0.1669
Age of youngest child
(years)
–0.0013 0.0023
3.4902
4.2042
–0.7140
0.0010
–0.0036
-0.0152
Number of children
0.0210 –0.0254
0.9659
1.0469
–0.0810
–0.0017
0.0464
0.0486
Additional family
income (inflation
adjusted in thousands
of dollars)
–0.0009 –0.0001
25.1172
34.9156
–9.7984
0.0086
–0.0008
-0.0284
Metropolitan area
0.0171 0.0305
0.6476
0.6806
–0.0330
–0.0006
–0.0133
-0.0091
Excellent health
0.0149 0.0062
0.2613
0.2041
0.0572
0.0009
0.0088
0.0018
Marital status
Never marrieda
Married
0.0800 –0.0011
0.7196
0.6101
0.1095
0.0088
0.0811
0.0495

Page 52
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 49
GAO-04-35 Women's Earnings
Estimate
Means (averages)
Difference
Variable
Men
βm
Women
βf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) βm
Between
parameters
(βm βf)
Due to
parameters
(returns)
Xf (βmβf)
Other
0.0685 –0.0009
0.1327
0.2424
–0.1097
–0.0075
0.0694
0.0168
Region: South
–0.0522 –0.0377
0.4142
0.4551
–0.0409
0.0021
–0.0145
-0.0066
Race
Whitea
Black
–0.1487 –0.0682
0.2666
0.3602
–0.0936
0.0139
–0.0806
-0.0290
Other
0.0491 0.0972
0.0140
0.0152
–0.0011
–0.0001
–0.0481
-0.0007
Year, compared to
1983
2000
0.0188 0.0621
0.0537
0.0538
–0.0001
–0.0000
–0.0433
-0.0023
1999b
1998
–0.0406 0.0298
0.0536
0.0515
0.0021
–0.0001
–0.0704
-0.0036
1997b
1996
–0.1045 –0.0733
0.0468
0.0514
–0.0046
0.0005
–0.0312
-0.0016
1995
–0.0813 –0.0618
0.0613
0.0622
–0.0009
0.0001
–0.0194
-0.0012
1994
–0.0973 –0.0759
0.0615
0.0655
–0.0040
0.0004
–0.0214
-0.0014
1993
–0.0854 –0.0495
0.0597
0.0641
–0.0044
0.0004
–0.0359
-0.0023
1992
–0.0693 –0.0625
0.0662
0.0684
–0.0022
0.0002
–0.0068
-0.0005
1991
–0.1023 –0.0921
0.0668
0.0675
–0.0007
0.0001
–0.0103
-0.0007
1990
–0.0960 –0.0737
0.0672
0.0686
–0.0015
0.0001
–0.0224
-0.0015
1989
–0.0691 –0.0524
0.0675
0.0680
–0.0006
0.0000
–0.0167
-0.0011
1988
–0.0359 –0.0516
0.0669
0.0667
0.0002
–0.0000
0.0157
0.0010
1987
–0.0389 –0.0561
0.0666
0.0660
0.0006
–0.0000
0.0171
0.0011
1986
–0.0248 –0.0632
0.0668
0.0654
0.0014
–0.0000
0.0384
0.0025
1985
–0.0282 –0.0822
0.0666
0.0646
0.0020
–0.0001
0.0540
0.0035
1984
–0.0237 –0.0847
0.0656
0.0631
0.0025
–0.0001
0.0609
0.0038
Sum before
intercept
-0.3618
Intercept
7.5910 6.9846
0.6065
Sum
0.4540
0.2446
Source: GAO analysis of PSID data.
aCategory omitted.
bNo data available.

Page 53
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 50
GAO-04-35 Women's Earnings
Table 5: Decomposition Results Using Alternative Estimates
Variable
Alternative
estimate
Means
(averages)
Difference
Men
gm
Women
gf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) gm
Between
parameters
(gm – gf)
Due to
parameters
(returns)
Xf (gm – gf)
Work Patterns
Experience (years)
0.0264 0.0249
16.2891
12.1342
4.1548
0.1095
0.0014
0.0175
Experience
squared
–0.0004 –0.0004 359.5914
210.6411
148.9504
–0.0558
0.0001
0.0120
Hours worked (per
year)
0.0003 0.0005 2,147.3100 1,674.8000
472.5100
0.1340
–0.0002
-0.3058
Time out of labor
force (weeks)
–0.0174 –0.0222
0.9262
2.8345
–1.9083
0.0332
0.0048
0.0136
Length of
unemployment
(weeks)
–0.0170 –0.0142
1.8149
1.8887
–0.0739
0.0013
–0.0028
-0.0053
Tenure (months)
0.0010 0.0009
91.4775
74.4278
17.0497
0.0163
0.0000
0.0015
Working full time
(in main job)
0.1881 0.1252
0.8761
0.6701
0.2059
0.0387
0.0628
0.0421
Other work
related
Mother’s education –0.0106 –0.0253
3.5458
3.4941
0.0516
–0.0005
0.0147
0.0515
Father’s education
0.0039 –0.0116
3.3364
3.2447
0.0917
0.0004
0.0155
0.0504
Highest education
(years)
0.1451 0.1738
13.1455
13.0880
0.0575
0.0083
–0.0287
-0.3757
Self-employment
status
Works for
someone else
onlya
Self-employed
only
–0.1003 0.2419
0.1177
0.0579
0.0597
–0.0060
–0.3422
-0.0198
Missing
–0.2461 –0.2892
0.0648
0.1230
-0.0582
0.0143
0.0432
0.0053
Both
0.0516 –0.0820
0.0094
0.0042
0.0052
0.0003
0.1336
0.0006
Union member
0.1488 0.1507
0.1773
0.1187
0.0587
0.0087
–0.0019
-0.0002
Occupation
Professional,
technicala
Service/private
household
workers
–0.1008 –0.0930
0.0763
0.2034
–0.1271
0.0128
–0.0079
-0.0016

Page 54
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 51
GAO-04-35 Women's Earnings
Variable
Alternative
estimate
Means
(averages)
Difference
Men
gm
Women
gf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) gm
Between
parameters
(gm – gf)
Due to
parameters
(returns)
Xf (gm – gf)
Farm laborers
and foremen
–0.1761 –0.0618
0.0121
0.0023
0.0098
–0.0017
–0.1143
-0.0003
Farmers and
farm
management
–0.2915 –0.1611
0.0124
0.0008
0.0116
–0.0034
–0.1304
-0.0001
Nonfarm
laborers
–0.0791 –0.0615
0.0547
0.0083
0.0464
–0.0037
–0.0176
-0.0001
Transport
equipment
operators
–0.0562 –0.1690
0.0680
0.0084
0.0596
–0.0033
0.1128
0.0009
Operators,
nontransport
–0.0449 –0.0638
0.0877
0.0879
–0.0002
0.0000
0.0188
0.0017
Craftsmen
0.0015 –0.0183
0.2049
0.0171
0.1879
0.0003
0.0198
0.0003
Clerical workers –0.0592 –0.0486
0.0497
0.2565
–0.2068
0.0122
–0.0106
-0.0027
Sales workers
–0.0339 –0.0891
0.0469
0.0409
0.0059
–0.0002
0.0552
0.0023
Nonfarm
managers,
administrators
0.0379 0.0165
0.1609
0.0922
0.0687
0.0026
0.0214
0.0020
Do not
know/missing
–0.1054 –0.1205
0.0468
0.0906
–0.0439
0.0046
0.0151
0.0014
Industry
Wholesale/retail
tradea
Public
administration
0.0102 0.1780
0.0799
0.0607
0.0192
0.0002
–0.1678
-0.0102
Professional
services
0.0172 0.0731
0.1211
0.3467
–0.2256
–0.0039
–0.0560
-0.0194
Entertainment
0.0039 –0.0737
0.0095
0.0061
0.0034
0.0000
0.0775
0.0005
Personal
services
–0.0306 –0.0098
0.0130
0.0678
–0.0549
0.0017
–0.0208
-0.0014
Business and
repair services
0.0729 0.0498
0.0585
0.0340
0.0245
0.0018
0.0231
0.0008
Finance,
insurance, real
estate
0.0575 0.1604
0.0394
0.0641
–0.0248
–0.0014
–0.1028
-0.0066
Transportation/
communication/
public utilities
0.1867 0.2046
0.0976
0.0353
0.0622
0.0116
-0.0178
-0.0006
Manufacturing
0.1521 0.1423
0.2444
0.1341
0.1103
0.0168
0.0098
0.0013

Page 55
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 52
GAO-04-35 Women's Earnings
Variable
Alternative
estimate
Means
(averages)
Difference
Men
gm
Women
gf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) gm
Between
parameters
(gm – gf)
Due to
parameters
(returns)
Xf (gm – gf)
Construction
0.1861 0.0689
0.0963
0.0101
0.0862
0.0160
0.1172
0.0012
Mining/
agriculture
0.0489 0.0166
0.0474
0.0075
0.0399
0.0020
0.0323
0.0002
Do not
know/missing
0.1164 0.0730
0.0513
0.0954
–0.0441
–0.0051
0.0434
0.0041
Mills ratio
–0.2819 –0.1470
0.1628
0.3771
–0.2143
0.0604
–0.1348
-0.0508
Demographic and
other controls
Age of individual
(years)
–0.0016 –0.0057
40.1442
40.3309
–0.1867
0.0003
0.0041
0.1662
Age of youngest
child (years)
–0.0013 0.0023
3.4902
4.2042
–0.7140
0.0010
–0.0036
-0.0152
Number of children
0.0212 –0.0251
0.9659
1.0469
–0.0810
–0.0017
0.0463
0.0485
Additional family
income (inflation
adjusted in
thousands of
dollars)
–0.0009 -0.0001
25.1172
34.9156
–9.7984
0.0086
–0.0008
-0.0284
Metropolitan area
0.0173 0.0309
0.6476
0.6806
–0.0330
–0.0006
–0.0136
-0.0093
Excellent health
0.0150 0.0062
0.2613
0.2041
0.0572
0.0009
0.0089
0.0018
Marital status
Never marrieda
Married
0.0831 –0.0013
0.7196
0.6101
–0.1097
–0.0091
0.0844
0.0515
Other
0.0707 –0.0011
0.1327
0.2424
0.0000
0.0000
0.0718
0.0174
Region: South
–0.0510 –0.0371
0.4142
0.4551
0.1095
–0.0056
–0.0139
-0.0063
Race
Whitea
Black
–0.1385 –0.0661
0.2666
0.3602
–0.0936
0.0130
–0.0723
-0.0260
Other
0.0466 0.0989
0.0140
0.0152
–0.0011
–0.0001
–0.0523
-0.0008
Year, compared to
1983
2000
0.0188 0.0638
0.0537
0.0538
–0.0001
0.0000
–0.0450
-0.0024
1999b
1998
–0.0399 0.0300
0.0536
0.0515
0.0021
–0.0001
–0.0699
-0.0036
1997b
1996
–0.0994 –0.0709
0.0468
0.0514
–0.0046
0.0005
–0.0285
-0.0015

Page 56
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 53
GAO-04-35 Women's Earnings
Variable
Alternative
estimate
Means
(averages)
Difference
Men
gm
Women
gf
Men
Xm
Women
Xf
Between
means
(averages)
(Xm – Xf)
Due to
characteristics
(Xm – Xf) gm
Between
parameters
(gm – gf)
Due to
parameters
(returns)
Xf (gm – gf)
1995
–0.0782 –0.0601
0.0613
0.0622
–0.0009
0.0001
–0.0181
-0.0011
1994
–0.0928 –0.0733
0.0615
0.0655
–0.0040
0.0004
–0.0196
-0.0013
1993
–0.0820 –0.0484
0.0597
0.0641
–0.0044
0.0004
–0.0335
-0.0021
1992
–0.0671 –0.0608
0.0662
0.0684
–0.0022
0.0002
–0.0063
-0.0004
1991
–0.0974 –0.0881
0.0668
0.0675
–0.0007
0.0001
–0.0093
-0.0006
1990
–0.0917 –0.0712
0.0672
0.0686
–0.0015
0.0001
–0.0205
-0.0014
1989
–0.0669 –0.0512
0.0675
0.0680
–0.0006
0.0000
–0.0157
-0.0011
1988
–0.0354 –0.0504
0.0669
0.0667
0.0002
–0.0000
0.0151
0.0010
1987
–0.0383 –0.0546
0.0666
0.0660
0.0006
–0.0000
0.0164
0.0011
1986
–0.0246 –0.0613
0.0668
0.0654
0.0014
–0.0000
0.0368
0.0024
1985
–0.0279 –0.0791
0.0666
0.0646
0.0020
–0.0001
0.0512
0.0033
1984
–0.0235 –0.0813
0.0656
0.0631
0.0025
–0.0001
0.0578
0.0036
Sum before
intercept
-0.3943
Intercept
7.5910 6.9846
0.6065
Sumc
0.4311
0.2122
Source: GAO analysis of PSID data.
aCategory omitted.
bNo data available.
cSum need not equal the log difference in earnings due to the transformation of the coefficients.
To determine whether our results would change significantly if the model
were specified slightly differently, we changed the specification in several
ways and compared those results with the results in the report. In all the
alternative specifications we developed, work patterns were important in
accounting for some of the earnings difference between men and women.
In addition, a significant gender earnings difference remained after
controlling for the effects of the variables in the model.
We developed several different specifications of the Hausman-Taylor
model presented in the report. In one particular alternative, we used a
linear time trend and the national unemployment rate instead of the year
specific dummy variables to control for the effects of national economic
conditions and other year-specific effects that are not reflected in the
other variables in the model. The results of this alternative specification

Page 57
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 54
GAO-04-35 Women's Earnings
also showed a slight narrowing of the earnings difference over time, but
they showed a decline in the difference in 1998 and 2000. We chose to
report the specification using dummy variables for each year because it is
more general than a linear time trend specification. However, this shows
that the results for certain years may be sensitive to the exact specification
chosen.
In other variants of the Hausman-Taylor model, we excluded occupation
and industry variables from the model, excluded observations from self-
employed individuals, limited the analysis to the Survey Research Center
portion of the PSID, and dropped the selection bias correction term from
the analysis. In these cases, the average earnings difference increased by
about 1 to 5 percentage points. As in the results we report, we found a
small downward trend in the difference in each case.
We also computed OLS regressions by year, using the same variables as in
the model we report. The earnings difference was smaller than the results
shown in table 2 (averaging about 14 percent over the period), and there
was a small downward trend in the difference over time.
While our analysis used what we consider to be the most appropriate
methods and data set available for our purposes, our analysis has both
data and methodological limitations that should be noted. Specifically,
although the PSID has many advantages over alternative data sets, like any
data set, it did not include certain data elements that would have allowed
us to further define reasons for earnings differences. For example, until
recently, the PSID did not contain data on fringe benefits—most
importantly, health insurance and pension coverage. Because data on
fringe benefits were not available for each year that we studied, we did not
include it for any year. If more women than men worked in jobs that
offered a greater percentage of total compensation in the form of fringe
benefits, part of the remaining gender earnings difference could be
explained by differences in the receipt of fringe benefits. Similarly, the
PSID does not contain data on job characteristics such as flexibility that
men and women may value differently.
In addition, the PSID does not contain data on education quality or field of
study, such as college major. It also does not contain data on cognitive
ability or measures of social skills, all of which may affect earnings. For
Limitations of Our
Analysis

Page 58
Appendix II: GAO Analysis of the Earnings
Difference between Men and Women
Page 55
GAO-04-35 Women's Earnings
example, studies of earnings differences that used the National
Longitudinal Survey of Youth have used a measure of ability in addition to
work experience, education, and demographic variables.13 This data set,
however, follows a specific cohort of individuals over time and is
therefore not representative of the population as a whole.
Our model is also limited in that the industry and occupation categories
that we used are broad. Gender earnings differences within these
categories are not reflected and could account for some amount of the
remaining difference. In addition, we did not explicitly model an
individual’s choice of occupation and industry and how these choices
relate to earnings differences. Also, although PSID collects information on
work interruptions, the detail of some of the survey questions limited our
ability to fully explore reasons why individuals were out of the labor force.
We used dummy variables for years to control for general economic
conditions and year-specific effects. In some specifications of the model,
we added national unemployment rate data to the PSID sample in order to
control for national labor market conditions. We did not access the PSID
Geocode Match file, which contains more detailed information on the
location of residence of survey respondents. We could not, therefore,
incorporate a measure of local unemployment rates in the analyses.
13See Altonji and Blank, pp. 3160–62, and June O’Neill, “The Gender Gap in Wages, circa
2000,” American Economic Review 93:2 (May 2003): 309-314

Page 59
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 56
GAO-04-35 Women's Earnings
Our analysis of data from the PSID identified factors that contribute to the
earnings difference between men and women, but cannot fully explain the
underlying reasons why these factors differ. For example, the model
results indicated that earnings differ, in part, because men and women
tend to have different work patterns (such as women are more likely to
work part time) and often work in different occupations. However, the
model could not explain why women worked part time more often or took
jobs in certain occupations. In addition, the analysis could not explain why
a remaining earnings difference existed after accounting for a range of
demographic, family, and work-related factors. To gain perspective on
these issues, we conducted additional work to gather information on why
individuals make certain decisions about work and how those decisions
may affect their earnings.
We conducted a multipronged effort, including a literature review,
interviews with employers as well as individuals with expertise on
earnings and other workplace issues,1 and a review of our work by
additional knowledgeable individuals. Specifically, we reviewed literature
on work-related decisions, including using alternative work arrangements,
and how these decisions may affect advancement or earnings. We also
conducted 10 interviews with a variety of experts—industry groups,
advocacy groups, unions, and researchers—to obtain a broad range of
perspectives on reasons why workers make certain career and workplace
decisions that could affect their earnings. In selecting experts, we targeted
those who have conducted research on earnings issues and have different
viewpoints.
We also interviewed employers from eight companies, as well as a group
of employees from one of these companies, about policies and practices,
including alternative work arrangements (such as part time and leave),
that may affect workers’ workplace decisions and earnings. We targeted
companies that are recognized leaders in work-life practices; for example,
those on Working Mother magazine’s “100 Best Companies for Working
Mothers” and on Fortune magazine’s “100 Best Companies to Work For”
list. In our selection, we also sought participation from a variety of sectors,
including:
financial/professional services
1These individuals will be referred to as “experts” throughout this appendix.
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Purpose
Scope and
Methodology

Page 60
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 57
GAO-04-35 Women's Earnings
health care
information technology
manufacturing
media/advertising
pharmaceuticals/biotechnology
travel/hospitality
Based on the literature and our interviews, we developed key themes
about workplace culture, decisions about work, and how these decisions
may affect career advancement and earnings. We vetted the themes with
11 experts—who are well known in the area of earnings and work-life
issues and represent views of researchers, advocacy groups, and
employers—to determine if the themes were consistent with their
experience or existing research and to identify areas of disagreement to
broaden our understanding of the issues.
According to experts and the literature, women are more likely than men
to have primary responsibility for family, and as a result, working women
with family responsibilities must make a variety of decisions to manage
these responsibilities. For example, these decisions may include what
types of jobs women choose as well as decisions they make about how,
when, and where they do their work. These decisions may have specific
consequences for their career advancement or earnings. However, debate
exists whether these decisions are freely made or influenced by
discrimination in society or in the workplace.
The tremendous growth in the number of women in the labor force in
recent decades has dramatically changed the world of work. The number
of women—particularly married women with children—who work has
increased, in many cases leaving no one at home to handle family and
other responsibilities. Single-headed households, in which only one parent
is available to handle both work and home responsibilities, are also
increasingly common. As a result, an increasing number of workers face
the challenge of trying to simultaneously manage responsibilities both
inside and outside the workplace.
At the same time, however, many employers continue to have certain
expectations about how much priority workers should give to work in
relation to responsibilities outside the workplace. While workplace culture
varies from one workplace to another, research indicates that in some
cases an “ideal worker” perception exists. According to this perception, an
Summary of Results
Background

Page 61
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 58
GAO-04-35 Women's Earnings
ideal worker places highest priority on work, working a full-time
9-to-5 schedule throughout their working years, and often working
overtime. Ideal workers take little or no time off for childbearing or
childrearing, and they appear—whether true or not—to have few
responsibilities outside of work. While this perception applies to all
workers, most experts and literature agree that it disproportionately
affects women because they often have or take primary responsibility for
home and family, such as caring for children, even when they are
employed outside of the home. However, some research indicates that
men are now more likely than in the past to participate in childcare,
eldercare, and housework and are beginning to adjust their work in
response to family obligations.
Some employers, however, have taken note of the multiple needs of
workers and have begun to offer alternative work arrangements to help
workers manage both work and other life responsibilities. These
arrangements can benefit workers by providing them with flexibility in
how, when, and where they do their work. One type of alternative work
arrangement allows workers to reduce their work hours from the
traditional 40 hours per week, such as part-time work or job sharing.2
Similarly, some employers offer workers the opportunity to take leave
from work for a variety of reasons, such as childbirth, care for elderly
relatives, or other personal reasons. Some arrangements, such as flextime,
allow employees to begin and end their workday outside the traditional
9-to-5 work hours. Other arrangements, such as telecommuting from
home, allow employees to work in an alternative location. Childcare
facilities are also available at some workplaces to help workers with their
caregiving responsibilities. In addition to benefiting workers, these
arrangements may also benefit employers by helping them recruit and
retain workers. For example, according to an industry group for attorneys,
law firms may lose new attorneys—particularly women who plan to have
children—if they do not offer workplace flexibility. This is costly to firms
due to substantial training investments they make in new attorneys, which
they may not recoup if workers quit early on.
Nonetheless, research suggests that many workplaces still maintain the
same policies, practices, and structures that existed when most workers
2Part-time work schedules allow employees to reduce their work hours from the traditional
40 hours per week in exchange for a reduced salary and possibly pro-rated benefits. Job
sharing—a form of part-time work—allows two employees to share job responsibilities,
salary, and benefits of one full-time position.

Page 62
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 59
GAO-04-35 Women's Earnings
were men who worked full time, 40-hours per week. As a result, there may
be a “mismatch” between the needs of workers with family responsibilities
and the structure of the workplace.
Working women make a variety of decisions to manage both their work
and home or family responsibilities. According to some experts and
literature, some women work in jobs that are more compatible with their
home and family responsibilities. In addition, some women use alternative
work arrangements such as working a part-time schedule or taking leave
from work. Experts indicate that these decisions may result in women as a
group earning less than men. However, debate exists about whether
women’s work-related decisions are freely made or influenced by
discrimination. Some experts believe that women and men generally have
different life priorities—women choose to place higher priority on home
and family, while men choose to place higher priority on career and
earnings. These women may voluntarily give up potential for higher
earnings to focus on home and family. However, other experts believe that
men and women have similar life priorities, and instead indicate that
women as a group earn less because of underlying discrimination in
society or in the workplace.
According to some experts and literature, some women choose to work in
jobs that are compatible with their home or family responsibilities, and
may trade off career advancement or higher earnings for these jobs. Some
experts and literature indicate that jobs that offer flexibility tend to be
lower paying and offer less career advancement.3
Women choose jobs with different kinds of flexibility based on their
needs. According to some researchers, some jobs are less demanding or
less stressful than others, which may allow women who choose these jobs
to have more time and energy for responsibilities outside of work. For
example, a woman may work in an off-line, staff position, such as a human
resources job, because it requires less travel and less time in the office
than an online position in the company. Off-line positions may offer
flexibility, but less opportunity for advancement and higher earnings. One
expert also indicated that, within a certain field, some women are more
3In contrast, other experts indicate that flexibility is often available in higher paying jobs,
particularly those where workers have more authority and autonomy.
Working Women
Make a Variety of
Decisions to Manage
Work and Family
Responsibilities
Certain Jobs May Offer
Flexibility but May Also
Affect Earnings

Page 63
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 60
GAO-04-35 Women's Earnings
likely to choose jobs that allow them more flexibility but lower earnings
potential. For example, according to this expert, within the medical field,
the family practice specialty is typically more accommodating to home and
family responsibilities than the surgical specialty, which offers relatively
higher earnings. Surgeons’ work is generally less predictable because
surgeons are often called in the middle of the night to treat emergencies.
The work is also less flexible because surgeons tend to see the same
patients throughout their treatment, while family practice doctors can rely
on other doctors in the practice to treat their patients if necessary. Experts
also noted that some women may start their own businesses, in part, to
gain flexibility in when and where they work.
According to some experts and literature, women may choose jobs that
allow them to quit (for example, to care for a child) and easily reenter the
labor force with minimal earnings loss when they return to work. Given
that job skills affect earnings, some suggest that certain women may
choose jobs in which skills deteriorate or become outdated less quickly.
As a result, this may allow women to leave and return to work while
minimizing any effect on their earnings.
Another way that women manage work and family responsibilities is by
choosing to use alternative work arrangements, which may affect their
career advancement and earnings.4 For example, some women choose to
work a part-time schedule, take leave from work, or use flextime. While
some research indicates that certain arrangements may help women
maintain their careers during times when they need flexibility, other
research suggests that there may be negative effects.
No single, national data source exists that provides information about all
workers who use alternative work arrangements. However, some data
exist from narrowly scoped studies that focus on particular types of work
arrangements, types of employees, or individual companies. Even when
employers offer alternative arrangements to all workers, some research
and the companies we interviewed indicate that women are more likely
than men to use certain arrangements, while both men and women use
others in similar proportions. Specifically, women are more likely than
men to take leave from work for family reasons and to work part time for
4Since women are more likely than men to use certain alternative work arrangements, any
effects apply disproportionately to women in these cases.
Alternative Work
Arrangements Offer
Flexibility but Some May
Affect Earnings

Page 64
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 61
GAO-04-35 Women's Earnings
family reasons even when these options are available to both men and
women. According to our interviews and some literature, some workers—
particularly men—are reluctant to use alternative arrangements because
they perceive that their advancement and earnings will be negatively
affected. This may help to explain why men tend to use personal days, sick
days, or vacation time instead of taking family leave. On the other hand,
similar proportions of men and women use flextime and telecommuting
when these options are available. However, according to some research,
men are more likely than women to work in the jobs, organizations, or
high-level, high-paying positions that have these options available.
Comprehensive, national data are lacking on how career advancement and
earnings may be affected by using alternative work arrangements, but
some limited research does exist. Certain researchers indicate that using
certain work arrangements may have some beneficial career effects if they
help workers maintain career linkages or skills that they might otherwise
lose. For example, for women who would have left the workforce or
changed jobs if they did not have access to alternative arrangements that
could help them manage work and family, part-time work5 may allow them
to maintain job skills, knowledge, or career momentum. In addition,
women who can take leave with the guarantee of returning to a similar job
benefit because they maintain links with an employer where they have
built up specific job-related skills.
Other research indicates that using certain alternative work arrangements
may have negative effects on career advancement and earnings.
Specifically, employers may view these workers as not conforming to the
ideal worker norm because they are not at work as much or during the
same work hours as their managers or co-workers. Research indicates that
some arrangements, such as leave, part-time work, and telecommuting,
reduce workers’ “face time”—the amount of time spent in the workplace.6
Given that some employers use face time as an indicator of workers’
productivity, those who lack face time may experience negative career
effects. According to some experts and literature, some employers may
5Research indicates that different types of part-time work exist. Some part-time jobs
require relatively low skills, and offer low pay and little opportunity for advancement. In
contrast, other part-time jobs are work schedules that employers create to retain or attract
workers who cannot or do not want to work full time. These jobs are often higher skilled
and higher paying with advancement potential.
6The idea of “face time” may apply primarily to certain types of jobs, such as professional,
white-collar jobs or those that require contact with clients or customers.

Page 65
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 62
GAO-04-35 Women's Earnings
view women who use alternative arrangements as less available, less
valuable, or less committed to their work. This may result in less
challenging work, fewer career opportunities, fewer promotions, and less
pay. However, one company representative that we interviewed told us
that workers using these arrangements are not necessarily less committed
and that, in some cases, they work harder. For example, several of the
women we interviewed who were scheduled to work less than full time
noted that they sometimes came into the office or worked at home on
their scheduled days off.
Although existing research is limited and often narrow in scope, following
are examples of studies that address advancement and earnings effects
that are associated with using certain alternative arrangements.
One study—which tracked a small group of working women for 7 years
after they gave birth—found that flextime, telecommuting, and reduced
work hours had some negative impact on wage growth for some
mothers. Flextime showed a neutral or mild impact on wage growth,
while telecommuting and reduced work hours—which result in less
face time—showed large pronounced negative effects, but only for
some workers. For all three arrangements, managers or professionals
experienced more negative wage effects than nonmanagerial or
nonprofessional workers.
Another study of 11,815 managers in a large financial services
organization found that leaves of absence were associated with fewer
subsequent promotions and smaller raises. This was true regardless of
the reason for the leave (i.e., a worker’s illness or family
responsibilities) or whether the leave taker was a man or woman—
though most of the managers taking leave were women. Taking leave
negatively affected workers’ performance evaluations, but only for the
year that they took the leave. Even when accounting for any potential
differences in the performance evaluations of those who did and did
not take leave, leave takers received fewer promotions and smaller
raises.
Managerial support for use of alternative work arrangements is important
when considering any effects on advancement and earnings. According to
our company interviews, some managers do not support use of these
arrangements because they are seen as accommodations to certain
workers—even though the company’s leadership views them as part of the
overall business strategy. Workers who use these arrangements may
experience negative effects if managers place limits on the types of work

Page 66
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 63
GAO-04-35 Women's Earnings
and responsibilities they receive. For example, one worker we interviewed
noted that she has not been assigned a high-profile project because she
works a part-time schedule. Most of the companies we interviewed noted
the importance of managers in implementing alternative work
arrangements, and as a result, many train managers on this topic. For
example, several companies train managers to focus on the quality of an
individual’s work rather than on when (i.e., what time of day) or where
(i.e., at home or at the workplace) they do their work. One company also
revised managers’ performance criteria to include their response to
flexible work arrangements.
On the other hand, some workers do not have the option to use alternative
work arrangements for several reasons. For example, some managers do
not allow workers to use alternative arrangements because they want to
directly monitor their workers, they fear that too many others will also
request these arrangements, or they do not understand how it relates to
the company’s bottom line. In addition, some workers—often those who
are lower paid—do not have the option to use alternative arrangements
because the nature of their job does not allow it. For example,
telecommuting may not be feasible for administrative assistants because
they must be in the office to support their bosses. Furthermore, low-paid
workers often cannot afford to choose a work arrangement that reduces
their pay. For example, some women in lower-paying jobs cannot afford to
take any unpaid maternity leave, or to take it for an extended period of
time, because of their financial situation.
Debate exists whether decisions that women make to manage work and
family responsibilities are freely made or influenced by underlying
discrimination. Some experts believe that women are free to make choices
about work and family, and willingly accept the earnings consequences.
Specifically, certain experts believe that some women place higher priority
on home and family, and voluntarily trade off career advancement and
earnings to focus on these responsibilities. Other experts believe that
some women place similar priority on family and career. Alternatively,
other women place higher priority on career and may delay or decide not
to have children. However, other experts believe that underlying
discrimination exists in the presumption that women have primary
responsibility for home and family, and as a result, women are forced to
make decisions to accommodate these responsibilities. One example of
this is a woman who must work part time for childcare reasons, but would
have preferred to work full time if she did not have this family
responsibility. In addition, some experts also suggest that women face
Potential for Direct Or
Indirect Discrimination

Page 67
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 64
GAO-04-35 Women's Earnings
other societal and workplace discrimination that may result in lower
earnings. However, according to other experts, although women may still
face discrimination in the workplace, it is not a systematic problem and
legal remedies are already in place. For example, Title VII of the Civil
Rights Act of 1964 prohibits employment discrimination based on gender.
According to some experts and literature, women face societal
discrimination that may affect their career advancement and earnings.
Some research suggests that the career aspirations of men and women
may be influenced by societal norms about gender roles. For example,
parents, peers, or institutions (such as schools or the media) may teach
them that certain occupations—such as nursing or teaching, which tend to
be relatively lower-paying—are identified with women while others are
identified with men. As a result, men and women may view different fields
or occupations as valuable or socially acceptable. According to some
experts, societal discrimination may help explain why men and women
tend to be concentrated in different occupations. For example, some
research has found that women tend to be over-represented in clerical and
service jobs, while men are disproportionately employed in blue-collar
craft and laborer jobs.7 Other research suggests that gender differences
exist even among those who are college educated. For example, men tend
to be concentrated in majors such as engineering and mathematics, while
women are typically concentrated in majors such as social work and
education. Research indicates that men and women who work in female-
dominated occupations earn less than comparable workers in other
occupations.
Additionally, some experts and literature suggest that women face
discrimination in the workplace. This type of discrimination may affect
what type of jobs women are hired into or whether they are promoted. In
some cases, employers or clients may underestimate women’s abilities or
male co-workers may resist working with women, particularly if women
are in higher-level positions. Employers may also discriminate based on
their presumptions about women as a group in terms of family
responsibilities—rather than considering each woman’s individual
situation. For example, employers may be less likely to hire or promote
7Notably, research indicates that women tend to be concentrated in service-producing
occupations, such as retail trade and government, which lose relatively few jobs or actually
gain jobs during recessions. However, men tend to be concentrated in goods-producing
industries, such as construction and manufacturing, which often lose jobs during
recessions.

Page 68
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 65
GAO-04-35 Women's Earnings
women because they assume that women may be less committed or may
be more likely to quit for home and family reasons. To the extent that
employers who offer higher-paying jobs discriminate against women in
this way, women may not have the same earnings opportunities as men.
Finally, other experts suggest that both men and women who are parents
face discrimination in the workplace due to their family responsibilities in
terms of hiring, promotions, and terminations on the job.
According to some literature, discrimination may occur if employers enact
policies or practices that have a disproportionately negative impact on one
group of workers, such as women with children. For example, if an
employer has a policy that excludes part-time workers from promotions,
this could have a significant effect on women because they are more likely
to work part time. Other experts suggest that workplace practices
reflecting ideal worker norms—such as requiring routine overtime for
promotion—could be considered discrimination. This could impact
women more (particularly mothers) and may result in a disproportionate
number of men in high-level positions.
Anderson, Deborah J., Melissa Binder, and Kate Krause. “The Motherhood
Wage Penalty Revisited: Experience, Heterogeneity, Work Effort, and
Work-Schedule Flexibility.” Industrial and Labor Relations Review
56:2 (2003): 273-295.
Appelbaum, Eileen, ed. “The New Realities of Family Life and the
Workplace: Is There a Mismatch?” in Balancing Acts: Easing the Burdens
and Improving the Options for Working Families, 1-9. Washington, D.C.:
Economic Policy Institute, 2000, http://www.lights.com/epi/virlib/
Studies/2000/balancinga/.
Appelbaum, Eileen, and Lonnie Golden. “The Standard Workday or the
Highway: Employers Stall in Delivery of More Flexible Arrangements That
Can Help Relieve Workers’ Time Squeeze.” Washington, D.C.: The Center
for Designing Work Wisely, 2002, http://www.cdww.org/pressroom.htm.
Appelbaum, Eileen. “The Transformation of Work and Employment
Relations in the U.S.” Paper presented at the NICHD/Sloan Foundation
Conference on Workplace/Workforce Mismatch? Work, Family, Health,
and Well-Being, Washington, D.C.: June 16-18, 2003,
http://www.popcenter.umd.edu/conferences/nichd/agenda.html.
Related Research

Page 69
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 66
GAO-04-35 Women's Earnings
Bailyn, Lotte, Robert Drago, and Thomas A. Kochan. “Integrating Work
and Family Life: A Holistic Approach.” Cambridge, Mass.: MIT Sloan
School of Management, 2001, http://icdl.uncg.edu/ft/070202-01.html.
Bardasi, Elena and Janet Gornick. “Women and Part-Time Employment:
Workers’ ‘Choices’ and Wage Penalties in Five Industrialized Countries.”
United Kingdom: Institute for Social and Economic Research, 2000,
http://netec.mcc.ac.uk/WoPEc/data/Papers/eseiserwp2000-11.html.
Barnett, Rosalind C. “A New Work-Life Model for the Twenty-First
Century.” Annals of American Academy of Political and Social Science
562 (1999): 143-158.
Bianchi, Suzanne M., and Sara Raley. “Changing Work and Family
Demographics.” Paper presented at the NICHD/Sloan Foundation
Conference on Workplace/Workforce Mismatch? Work, Family, Health,
and Well-Being, Washington, D.C.: June 16-18, 2003,
http://www.popcenter.umd.edu/conferences/nichd/agenda.html.
Blair-Loy, Mary, and Amy S. Wharton. “Employees’ Use of Work-Family
Policies and the Workplace Social Context.” Social Forces 80:3 (2002):
813-845.
Blank, Rebecca M. “The Dynamics of Part-Time Work.” (Working Paper
Number 4911). Cambridge, Mass.: National Bureau of Economic Research,
1994, http://papers.nber.org/papers/W4911.
Boden, Richard J. Jr. “Flexible Working Hours, Family Responsibilities,
and Female Self-Employment.” The American Journal of Economics and
Sociology 58:1 (1999): 71-83.
Bond, James T., Ellen Galinsky, and Jennifer Swanberg. “The 1997
National Study of the Changing Workforce.” New York: Families and Work
Institute, 1998.
Boston College Center for Work and Family. “Measuring the Impact of
Workplace Flexibility.” Chestnut Hill, Mass.: Center for Work and Family,
2000, http://www.bc.edu/centers/cwf/research/highlights/.
Bravo, Ellen, Mark Greenberg, and Cindy Marano. “Investing in Family
Well-Being, a Family-Friendly Workplace and a More Stable Workforce: A
‘Win-Win’ Approach to Welfare and Low-Wage Policy,” 2002,
http://www.economythatworks.org/PDFs/ford_policyfinal.pdf.

Page 70
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 67
GAO-04-35 Women's Earnings
Brown, Charles, and Mary Corcoran. “Sex-Based Differences in School
Content and the Male/Female Wage Gap.” (Working Paper 5580.)
Cambridge, Mass.: National Bureau of Economic Research, 1996,
http://ideas.repec.org/p/nbr/nberwo/5580.html.
Budig, Michelle J., and Paula England. “The Wage Penalty for
Motherhood.” American Sociological Review 66:2 (2001): 204-225.
Caputo, Richard K. “Race and Marital History as Correlates of Women’s
Access to Family-Friendly Employee Benefits.” Journal of Family and
Economic Issues 21:4 (2000): 365-385.
Carr, Deborah. “The Psychological Consequences of Work-Family Trade-
Offs for Three Cohorts of Men and Women.” Social Psychology Quarterly
65:2 (2002): 103-124.
Catalyst. Flexible Work Arrangements III: A Ten-Year Retrospective of
Part-Time Arrangements for Managers and Professionals. New York:
Catalyst, 2000.
Catalyst. A New Approach to Flexibility: Managing the Work/Time
Equation. New York: Catalyst, 1997.
Cooper, Marianne. “Being the ‘Go-To-Guy’: Fatherhood, Masculinity, and
the Organization of Work in Silicon Valley.” Qualitative Sociology 23:4
(2000): 379-405.
Correll, Shelley J. “Gender and the Career Choice Process: The Role of
Biased Self-Assessments.” American Journal of Sociology 106:6 (2001):
1691-1730.
Costello, Cynthia B., Vanessa R. Wight, and Anne J. Stone, eds. The
American Woman 2003-2004: Daughters of a Revolution – Young
Women Today. New York: Palgrave Macmillan, 2003.
Crittenden, Ann. The Price of Motherhood: Why the Most Important Job in
the World is Still the Least Valued. New York: Henry Holt and Company,
2001.

Page 71
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 68
GAO-04-35 Women's Earnings
Darrah, Charles N. “Anthropology, Ethnography, and the Thesis of a
Workplace/Workforce Mismatch.” Paper presented at the NICHD/Sloan
Foundation Conference on Workplace/Workforce Mismatch? Work,
Family, Health, and Well-Being, Washington, D.C.: June 16-18, 2003,
http://www.popcenter.umd.edu/conferences/nichd/agenda.html.
Dodson, Lisa, Tiffany Manuel, and Ellen Bravo. “Keeping Jobs and Raising
Families in Low-Income America: It Just Doesn’t Work.” Cambridge, Mass:
Harvard University, Radcliffe Institute for Advanced Study, 2002,
http://www.radcliffe.edu/pubpol/boundaries.pdf.
Eaton, Susan C. “If You Can Use Them: Flexibility Policies, Organizational
Commitment, and Perceived Productivity.” (Draft Working Paper).
Cambridge, Mass.: Harvard University, John F. Kennedy School of
Government, 2001, http://ksgnotes1.harvard.edu/research/wpaper.nsf/
rwp/RWP01-009?OpenDocument.
Employment Policies Institute. “Correcting Part-Time Misconceptions.”
Washington, D.C.: Employment Policies Institute, 2000.
Ferber, Marianne A., and Jane Waldfogel. The Long-Term Consequences of
Nontraditional Employment. Monthly Labor Review, 121:5 (1998): 3-12.
Ford Foundation. “Relinking Life and Work: Toward a Better Future.” New
York: Ford Foundation, 1997, http://www.fordfound.org/publications/
recent_articles/life_and_work/relink_toc.cfm.
Firestone, Juanita M., Richard J. Harris, and Linda C. Lambert. “Gender
Role Ideology and the Gender Based Difference in Earnings.” Journal of
Family and Economic Issues 20:2 (1999): 191-215.
Fletcher, Joyce K. “Gender Perspectives on Work and Personal Life
Research.” Paper presented at the NICHD/Sloan Foundation Conference
on Workplace/Workforce Mismatch? Work, Family, Health, and Well-
Being, Washington, D.C.: June 16-18, 2003, http://www.popcenter.umd.edu/
conferences/nichd/agenda.html.
Friedman, Dana E. “Employer Supports for Parents with Young Children.”
The Future of Children 11:1 (2001): 63-77.
Furchtgott-Roth, Diana, and Christine Stolba. The Feminist Dilemma:
When Success is Not Enough. Washington, D.C.: The American Enterprise
Institute Press, 2001.

Page 72
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 69
GAO-04-35 Women's Earnings
Galinksy, Ellen, and James T. Bond. “The 1998 Business Work-Life Study:
A Sourcebook.” New York: Families and Work Institute, 1998.
Galinsky, Ellen, Stacy S. Kim, and James T. Bond. “Feeling Overworked:
When Work Becomes Too Much.” New York: Families and Work Institute,
2001.
Gerstel, Naomi, and Katherine McGonagle. “Job Leaves and the Limits of
the Family and Medical Leave Act.” Work and Occupations 26:4 (1999):
510-534.
Glass, Jennifer. “Blessing or Curse? “Work-Family Policies and Mothers’
Wage Growth Over Time.” Updated version of a paper presented at the
Population Association of America meeting, March 2000.
Glass, Jennifer. “The Impact of Occupational Segregation on Working
Conditions.” Social Forces 68:3 (1990): 779-796.
Glass, Jennifer. “Sociological Perspectives on Work and Family.” Paper
presented at the NICHD/Sloan Foundation Conference on
Workplace/Workforce Mismatch? Work, Family, Health, and Well-Being,
Washington, D.C.: June 16-18, 2003, http://www.popcenter.umd.edu/
conferences/nichd/agenda.html.
Glass, Jennifer, and Valerie Camarigg. “Gender, Parenthood, and Job-
Family Compatibility.” American Journal of Sociology 98:1 (1992):
131-151.
Glass, Jennifer L., and Sarah Beth Estes. “The Family Responsive
Workplace.” Annual Review of Sociology 23 (1997): 289-313.
Glass, Jennifer L., and Lisa Riley. “Family Responsive Policies and
Employee Retention Following Childbirth.” Social Forces 76:4 (1998):
1401-1435.
Golden, Lonnie. “Flexible Work Schedules: What Are We Trading Off to
Get Them?” Monthly Labor Review, 124:3 (2001): 50-67.

Page 73
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 70
GAO-04-35 Women's Earnings
Gornick, Janet C., and Marcia K. Meyers. “Parental Care of Children:
Family Leave Policy and the Regulation of Working Time,” 2002,
http://depts.washington.edu/crfam /Symposium1/
Gornick_Meyers_chap5.pdf. (Excerpt from: Janet C. Gornick and Marcia
K. Meyers, eds., forthcoming. Earning and Caring: What Government
Can Do to Reconcile Motherhood, Fatherhood, and Employment. New
York: Russell Sage Publications.)
Grover, Steven L., and Karen J. Crooker. “Who Appreciates Family-
Responsive Human Resource Policies: The Impact of Family-Friendly
Policies on the Organizational Attachment of Parents and Non-Parents.”
Personnel Psychology 48:2 (1995): 271.
Hakim, Catherine. Work-Lifestyle Choices in the 21st Century: Preference
Theory. New York: Oxford University Press, 2000.
Hammonds, Keith H., and Ann Therese Palmer. “The Daddy Trap.”
Business Week, September 21, 1998, 56.
Hartmann, Heidi, Young-Hee Yoon, and Diana Zuckerman. “Part-Time
Opportunities for Professionals and Managers: Where Are They, Who Uses
Them, and Why.” Washington, D.C.: Institute for Women’s Policy Research,
2000.
Hecker, Daniel E. “Earnings of College Graduates: Women Compared to
Men.” Monthly Labor Review, 121:3 (1998): 62-71.
Hewlett, Sylvia Ann. Creating a Life: Professional Women and the Quest
for Children. New York: Talk Miramax Books, 2002.
Hirsch, Barry. “The Relative Compensation of Part-Time and Full-Time
Workers.” Washington, D.C.: Employment Policies Institute, 2000.
Holzer, Harry J. “Work and Family Life: The Perspective of Employers.”
Paper presented at the NICHD/Sloan Foundation Conference on
Workplace/Workforce Mismatch? Work, Family, Health, and Well-Being,
Washington, D.C.: June 16-18, 2003, http://www.popcenter.umd.edu/
conferences/nichd/agenda.html.
Hundley, Greg. “Male/Female Earnings Differences in Self-Employment:
The Effects of Marriage, Children, and the Household Division of Labor.”
Industrial and Labor Relations Review 54:1 (2000): 95-115.

Page 74
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 71
GAO-04-35 Women's Earnings
Jacobs, Jerry A. “Changing Hours of Employment in American Families.”
Paper presented at the NICHD/Sloan Foundation Conference on
Workplace/Workforce Mismatch? Work, Family, Health, and Well-Being,
Washington, D.C.: June 16-18, 2003, http://www.popcenter.umd.edu/
conferences/nichd/agenda.html.
Jones, Ray, and Audrey J. Murrell. “Signaling Positive Corporate Social
Performance.” Business and Society 40:1 (2001): 59-78.
Judiesch, Michael K., and Karen S. Lyness. “Left Behind? The Impact of
Leaves of Absence on Managers’ Career Success.” Academy of
Management Journal 42:6 (1999): 641-651.
Kalleberg, Arne L., Edith Rasell, Naomi Cassirer, Barbara F. Reskin, Ken
Hudson, David Webster, Eileen Appelbaum, and Roberta M. Spalter-Roth.
“Nonstandard Work, Substandard Jobs: Flexible Work Arrangements in
the U.S.” Washington, D.C.: Economic Policy Institute, 1997,
http://www.lights.com/epi/virlib/Studies/1997/nonstandardw.PDF.
Keene, Jennifer Reid, and John R. Reynolds. “Gender Differences in the
Job Consequences of Family-to-Work Spillover,”
http://garnet.acns.fsu.edu/~jreynold/ keene_reynolds_asa2002.pdf.
Kelly, Erin, Frank Dobbin, Alexandra Kalev, and Samantha Ammons. “The
Princeton-Minnesota Study of Employers’ Family Policies: A Report to
Respondents,” http://www.soc.umn.edu/~elkelly/FinalReport1002.pdf.
King, Jerome E. “Part-time Workers’ Earnings: Some Comparisons.”
Compensation and Working Conditions 5:2 (2000): 27-36.
Klerman, Jacob Alex, and Arleen Leibowitz. “Job Continuity Among New
Mothers.” Demography 36:2 (1999): 145-155.
Kossek, Ellen Ernst. “Workplace Policies and Practices to Support Work
and Families: Gaps in Implementation and Linkages to Individual and
Organizational Effectiveness.” Paper presented at the NICHD/Sloan
Foundation Conference on Workplace/Workforce Mismatch? Work,
Family, Health, and Well-Being, Washington, D.C.: June 16-18, 2003,
http://www.popcenter.umd.edu/conferences/nichd/agenda.html.
Lambert, Susan J. “Lower-Wage Workers and the New Realities of Work
and Family.” Annals of the American Academy of Political and Social
Science 562 (1999): 174-190.

Page 75
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 72
GAO-04-35 Women's Earnings
Leibowitz, Arleen A. “An Economic Perspective on Work, Family, and
Well-Being.” Paper presented at the NICHD/Sloan Foundation Conference
on Workplace/Workforce Mismatch? Work, Family, Health, and Well-
Being, Washington, D.C.: June 16-18, 2003, http://www.popcenter.umd.edu/
conferences/nichd/agenda.html.
Levine, Linda. “The Gender Wage Gap and Pay Equity: Is Comparable
Worth the Next Step?” Washington, D.C.: Congressional Research Service,
2001.
MacDermid, Shelley M., Leon C. Litchfield, and Marcie Pitt-Catsouphes.
“Organizational Size and Work-Family Issues.” Annals of the American
Academy of Political and Social Science 562 (1999): 111-126.
McCrate, Elaine. “Working Mothers in a Double Bind: Working Moms,
Minorities Have the Most Rigid Schedules, and Are Paid Less for the
Sacrifice.” Washington, D.C.: Economic Policy Institute, 2002.
Meiskins, Peter, and Peter Whalley. Putting Work in its Place: A Quiet
Revolution. New York: Cornell University Press, 2002.
Miree, Cynthia E., and Irene Hanson Frieze. “Children and Careers: A
Longitudinal Study of the Impact of Young Children on Critical Career
Outcomes of MBAs.” Sex Roles 41:11/12 (1999): 787-808.
Osterman, Paul. “Work/Family Programs and the Employment
Relationship.” Administrative Science Quarterly 40:4 (1995): 681.
Perlow, Leslie A. “Putting the Work Back into Work/Family.” Group and
Organization Management 20:2 (1995): 227-239.
Reskin, Barbara F. “Getting it Right: Sex and Race Inequality in Work
Organizations.” Annual Review of Sociology 26 (2000): 707-709.
Roehling, Patricia V., Mark V. Roehling, and Phyllis Moen. “The
Relationship Between Work-Life Policies and Practices and Employee
Loyalty: A Life Course Perspective.” Journal of Family and Economic
Issues 22:2 (2001): 141-170.

Page 76
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 73
GAO-04-35 Women's Earnings
Ruhm, Christopher J. “How Well Do Parents With Young Children
Combine Work and Family Life?” Paper presented at the NICHD/Sloan
Foundation Conference on Workplace/Workforce Mismatch? Work,
Family, Health, and Well-Being, Washington, D.C.: June 16-18, 2003,
http://www.popcenter.umd.edu/conferences/nichd/agenda.html.
Saltzstein, Alan L., Yuan Ting, and Grace Hall Saltzstein. “Work-Family
Balance and Job Satisfaction: The Impact of Family-Friendly Policies on
Attitudes of Federal Government Employees.” Public Administration
Review 61:4 (2001): 452-467.
Schneer, Joy A., and Frieda Reitman. “Effects of Employment Gaps on the
Careers of M.B.A.’s: More Damaging for Men Than for Women?” Academy
of Management Journal 33:2 (1990): 391-406.
Schwartz, Debra B. “An Examination of the Impact of Family-Friendly
Policies on the Glass Ceiling.” New York: Families and Work Institute,
1994, http://www.ilr.cornell.edu/library/e_archive/gov_reports/
GlassCeiling/papers/Family-FriendlyPolicies.pdf.
Sharpe, Deanna L., Joan M. Hermsen, and Jodi Billings. “Factors
Associated with Having Flextime: A Focus on Married Workers. Journal of
Family and Economic Issues 23:1 (2002): 51-72.
Shore, Rima. “Ahead of the Curve: Why America’s Leading Employers Are
Addressing the Needs of New and Expectant Parents.” New York: Families
and Work Institute, 1998.
Stolba, Christine. “The Wage Gap.” Arlington, Va.: Independent Women’s
Forum, 2001, http://www.iwf.org/news/IUWageGap.pdf.
The Council of Economic Advisors. “Explaining Trends in the Gender
Wage Gap.” Washington, D.C.: The Council of Economic Advisors, 1998,
http://clinton4.nara.gov/WH/EOP/CEA/html/gendergap.html.
Thompson, Cynthia A., Jeanine Andreassi, and David Prottas. “Work-
Family Culture and Climate.” Paper presented at the NICHD/Sloan
Foundation Conference on Workplace/Workforce Mismatch? Work,
Family, Health, and Well-Being, Washington, D.C.: June 16-18, 2003,
http://www.popcenter.umd.edu/conferences/nichd/agenda.html.

Page 77
Appendix III: GAO Analysis of Women’s
Workplace Decisions
Page 74
GAO-04-35 Women's Earnings
Tilly, Chris. “Short Hours, Short Shrift: Causes and Consequences of Part-
Time Work.” Washington, D.C: Economic Policy Institute, 1990,
http://www.lights.com/epi/virlib/Studies/1990/shorth.PDF.
U.S. General Accounting Office. The Changing Workforce: Comparison of
Federal and Nonfederal Work/Family Programs and Approaches.
GAO/GGD-92-84. Washington, D.C.: April 23, 1992.
U.S. Department of Labor. “Balancing the Needs of Families and
Employers: The Family and Medical Leave Surveys—2000 Update.”
Washington, D.C., 2001, http://www.dol.gov/asp/fmla.
U.S. Department of Labor, Bureau of Labor Statistics. “Highlights of
Women’s Earnings in 2001.” Washington, D.C., 2002.
Waldfogel, Jane. “Understanding the ‘Family Gap’ in Pay for Women with
Children. Journal of Economic Perspectives 12:1 (1998): 137-156.
Waldfogel, Jane. “The Effect of Children on Women’s Wages.” American
Sociological Review 62:2 (1997): 209-217.
Williams, Joan. Unbending Gender: Why Family and Work Conflict and
What to Do About It. New York: Oxford University Press, 2000.
Williams, Joan, and Nancy Segal. “The New Glass Ceiling: Mothers – and
Fathers – Sue for Discrimination.” Washington, D.C.: The Program on
Gender, Work and Family, American University Washington College of
Law, 2002 (2nd edition), http://www.wcl.american.edu/gender/workfamily/
chilly_climate0211.pdf.
Women’s Bar Association of Massachusetts, Employment Issues
Committee. “More Than Part-time: The Effect of Reduced-Hours
Arrangements on the Retention, Recruitment, and Success of Women
Attorneys in Law Firms.” Boston, Mass.: Women’s Bar Association of
Massachusetts, 2000.
Wootton, Barbara H. “Gender Differences in Occupational Employment.”
Monthly Labor Review 120:4 (1997): 15-24.

Page 78
Appendix IV: GAO Contact and Staff
Acknowledgments
Page 75
GAO-04-35 Women's Earnings
Linda Siegel, Analyst in Charge (202) 512-7150
The following individuals also made important contributions to this report:
Patrick DiBattista, R. Scott McNabb, Corinna Nicolaou, and Caterina
Pisciotta, Education, Workforce, and Income Security Issues. In addition,
the following individuals played a key role in developing the statistical
model and conducting the analysis: Brandon Haller, Ed Nannenhorn,
MacDonald Phillips, and Wendy Turenne, Applied Research and Methods;
Scott Farrow, Chief Economist; and Robert Parker, Chief Statistician.
Appendix IV: GAO Contact and Staff
Acknowledgments
GAO Contact
Staff
Acknowledgments
(130187)

Page 79
The General Accounting Office, the audit, evaluation and investigative arm of
Congress, exists to support Congress in meeting its constitutional responsibilities
and to help improve the performance and accountability of the federal
government for the American people. GAO examines the use of public funds;
evaluates federal programs and policies; and provides analyses,
recommendations, and other assistance to help Congress make informed
oversight, policy, and funding decisions. GAO’s commitment to good government
is reflected in its core values of accountability, integrity, and reliability.
The fastest and easiest way to obtain copies of GAO documents at no cost is
through the Internet. GAO’s Web site (www.gao.gov) contains abstracts and full-
text files of current reports and testimony and an expanding archive of older
products. The Web site features a search engine to help you locate documents
using key words and phrases. You can print these documents in their entirety,
including charts and other graphics.
Each day, GAO issues a list of newly released reports, testimony, and
correspondence. GAO posts this list, known as “Today’s Reports,” on its Web site
daily. The list contains links to the full-text document files. To have GAO e-mail
this list to you every afternoon, go to www.gao.gov and select “Subscribe to e-mail
alerts” under the “Order GAO Products” heading.
The first copy of each printed report is free. Additional copies are $2 each. A
check or money order should be made out to the Superintendent of Documents.
GAO also accepts VISA and Mastercard. Orders for 100 or more copies mailed to a
single address are discounted 25 percent. Orders should be sent to:
U.S. General Accounting Office
441 G Street NW, Room LM
Washington, D.C. 20548
To order by Phone: Voice: (202) 512-6000
TDD:
(202) 512-2537
Fax:
(202) 512-6061
Contact:
Web site: www.gao.gov/fraudnet/fraudnet.htm
E-mail: fraudnet@gao.gov
Automated answering system: (800) 424-5454 or (202) 512-7470
Jeff Nelligan, Managing Director, NelliganJ@gao.gov (202) 512-4800
U.S. General Accounting Office, 441 G Street NW, Room 7149
Washington, D.C. 20548
GAO’s Mission
Obtaining Copies of
GAO Reports and
Testimony
Order by Mail or Phone
To Report Fraud,
Waste, and Abuse in
Federal Programs
Public Affairs