Abstract
This article examines how the pay gap between men and women in the federal workforce changed over the period 1988-2007 and what factors contributed to the pay gap. To do this, we use a decomposition method to analyze the most recent data available for a representative sample of federal employees. We find that the pay gap—the difference between men’s and women’s average pay before controlling for factors that affect pay—declined dramatically over the 20-year period. Most of this decline is because women and men in the federal workforce have become more similar over time in their levels of education and work experience and the jobs they perform in the federal government.
Introduction
Over the past few decades, and particularly during the last presidential campaign, the persistent pay gap between men and women in the U.S. workforce has become a focal point for public policy. 1 The Lilly Ledbetter Fair Pay Act—a law intended to bolster protections against pay discrimination for women and other workers—was one of the first pieces of legislation enacted by the Obama presidency. The Paycheck Fairness Act—a law designed to update the Equal Pay Act of 1963 by making it easier to seek damages over pay discrimination based on gender—was proposed during the 111th Congress but did not become law. 2
Although the Equal Pay Act of 1963 and the Civil Rights Act of 1964 are widely seen as contributing to the narrowing of the pay gap between men and women over time, a persistent gap remains. Policy makers regularly cite Census data indicating women in the U.S. workforce are paid only 77 cents for every dollar that men receive. However, this statistic is misleading because it is not comparing women and men in the same occupations and with similar skill-sets. By comparing women and men with similar jobs and skills, policy makers can better understand whether the pay gap is the result of labor market discrimination or differences in other factors that affect pay—such as type of occupation, level of education, or level of work skills. Understanding the sources of differences in pay would enable policy makers to better tailor public policy to address those differences.
As the largest employer in the United States, the federal government, and public sector in general, is looked to as a leader in equal-opportunity employment. 3 In the case of equal pay for men and women, the focus of this article, researchers have found that the inequality within the federal government has existed for decades but has been lessening over time. For example, using a sample of white-collar federal workers, Lewis found that the difference between White non-Hispanic women and men fell from 40% to 37% from 1976 to 1986 (Lewis, 1988). More recently, Lewis reported that the difference was 27% in 1995. Lewis attributed the decline of the gender–pay gap in the federal workforce to the narrowing of the gap between them in their levels of education and federal experience (Lewis, 1998).
In addition, researchers have also examined the adjusted pay gap—the gap after differences in education, experience (as well as other factors), are accounted for. Using a sample of federal workers from 1979, Borjas found that white women earned 27% less than white men after controlling for differences in education and federal experience levels (Borjas, 1983). Using more recent data on federal employees from 1995, Lewis found that white women earned around 15% less than white men after controlling for differences in age, federal experience, and education (Lewis, 1998). Lewis presents some evidence that the narrowing of the adjusted pay gap could be due to narrowing differences in nonfederal experience between women and men (Lewis, 1998).
This article builds on these analyses by examining the gender–pay gap in the federal workforce over a 20-year period (1988 to 2007) and investigates whether changes in the gap were the result of changes in the characteristics of federal employees or federal reward structures. Using the most recent data available on federal workers, we employ a decomposition approach to determine how the pay gap between men and women has changed over this period and what proportion can be explained by differences in factors that might reasonably be expected to affect pay—such as type of occupation, education level, and years of federal work experience.
The article is organized as follows: We first identify variables we include in our empirical model to explain pay differences. We then describe our data and present historical statistics on the federal government from 1988 to 2007. Finally, we present our empirical approach and findings.
Measuring Pay Differences Using Empirical Techniques
To measure differences in pay between men and women one would ideally compare pay between two people with exactly the same characteristics in every respect except the potential source of discrimination (in this instance sex). Such a comparison would ensure that differences in pay between men and women are not due to differences in some other characteristic between men and women, such as education, that might affect pay. For example, some researchers have designed experiments to test for gender discrimination by sending men and women with the identical qualifications (as written on a resume, for example) to apply for a job or an audition. 4
When researchers do not have the luxury of conducting formal experiments with men and women with identical characteristics, they use regression analysis. As with an experiment, regression analysis effectively allows the researcher to analyze a data set of people with a variety of traits and then control for differences in those traits to isolate the impact of a particular trait on pay. In the case of pay differences, regression analysis allows the researcher to isolate the effect of differences in sex from the effect of differences in other factors that might affect pay.
Several factors are typically considered in empirical analyses of pay (Altonji & Blank, 1999). First, human capital characteristics that employees bring to the job can affect their productivity levels, such as education and experience. Employees with more education—and therefore higher skill levels—are more likely to command higher salaries than those with less education. Also, certain fields of education are more likely to receive higher pay than other fields. Research has shown, for example, that a significant portion of the pay gap between male and female federal employees can be explained by men’s higher likelihood to obtain education in higher paying fields such as engineering, computer science, and business (Lewis & Oh, 2009). Similarly, wages tend to increase with years of work experience, and years of experience may also differ by gender. Blau and Kahn found that models that controlled for actual work experience decreased the pay gap between men and women by 8 percentage points in 1989 (in the entire workforce) due to differences between men and women’s experience levels (Blau & Kahn, 2007).
The traits of the job itself will also affect salary level. These would include (a) the occupation or, more generally, occupational sector, (b) the geographic location of the job, as wages are influenced by local labor markets and the cost of living within a particular geographic area, and (c) whether the work is performed on a full-time or part-time basis.
In addition, personal characteristics can also affect pay. For example, family responsibilities can affect the career choices that a person makes and work performance on the job. Past research has found that such responsibilities disproportionately fall on women and can help explain part of the wage gap between women and men. For example, using the American Time Use Survey, Hersch found that hours spent performing housework had a negative effect on both male and female hourly pay and that because women generally spent more time on housework this helped explain the pay difference (Hersch, 2009). When data indicators of hours spent by activity are not available, marital status and number of children sometimes proxy for family obligations. Other personal characteristics—including gender, race, disability status, union status and, in the case of federal workers, veteran status—might be the basis for discriminatory and/or preferential treatment.
Data Source and Changes in the Federal Workforce Over Time
The data for the analysis come from the status file of the Central Personnel Data File (CPDF). The CPDF is produced by the Office of Personnel Management as a central source of information regarding the federal workforce and contains information on most federal employees who were in the federal workforce. 5 Specifically, the CPDF has data on the federal employee’s adjusted basic pay, agency, age, education level, disability status, occupation, race or national origin, gender, veteran’s preference and status, and work schedule (full-time, part-time, etc.). For our analysis, we analyzed a random sample of 20% of the workers in the CPDF during the month of September in 1988, 1998, and 2007.6,7
Table 1 shows descriptive statistics for federal workers in the 3 years used in our analysis. Over all, the table shows the federal workforce has undergone significant changes during the past 20 years, both in the type of people employed and the type of work they do. Over this period, the federal workforce has become more educated and experienced relative to 20 years ago. The proportion of federal employees with a bachelor’s degree or higher increased from 33% in 1988 to 44% in 2007. Similarly, the average years of federal service increased from 12 to 15 years over this period, and the proportion of employees with over 20 years of experience increased from 21% to 34% (not shown in table).
Changes in the Characteristics of the Federal Workforce Over Time
Note: CPDF data. Annualized salary is adjusted for inflation using the CPI. Numbers may not sum to 100 because of rounding.
There also have been dramatic shifts in the nature of the work performed in the federal government. The proportion of workers in clerical and blue-collar occupations declined steeply. At the same time, the federal workforce has become increasingly concentrated in the professional and administrative fields, which typically require a college education. 8
Over the past 20 years, the demographic composition of the federal workforce has also changed. The workforce has become older: The average ages for workers increased from 42 in 1988 to 47 in 2007. There was also a decline in the proportion of White workers in the workforce and an increase in the proportion of Hispanic and Asian/Pacific Islanders. 9 Finally, the proportion of women in the federal workforce increased from 42% to 44%.
Converging Characteristics of Male and Female Federal Workers
In addition to the changes in the characteristics of the federal workforce, men and women within the federal workforce have become more alike in characteristics that can reasonably be expected to affect pay—especially in their years of work experience, education levels, and occupations—as seen in Table 2. In 1988, men in the federal workforce had an average of 3 more years of work experience than their female counterparts. However, that difference narrowed to 2 years by 1998, and by 2007 the difference had virtually disappeared. 10 Similarly, in 1988, almost twice as many men as women in the federal workforce had bachelor’s, master’s, professional, or doctoral degrees (40% vs. 23%), but by 2007, the difference had declined to 6 percentage points (46% vs. 40%).
Differences Between Men and Women Federal Workers Over Time
Also, over the past two decades, male and female federal workers increasingly worked in similar occupations. Much of this trend is due to the diminishing clerical sector in the federal workforce. In 1988, about 38% of women in the federal workforce were in a clerical occupation. By 2007, that number was down to 13%. A similar trend occurred with men in the blue-collar sector. In 1988, almost 28% of men in the federal workforce were blue-collar workers, but by 2007 that number was down to 17%. 11 Interestingly, differences between men’s and women’s work schedules did not change significantly over the study period: The percentage of women working full-time remained at about 4 to 5 percentage points lower than men during each study year. 12
Although the CPDF does not contain variables for marital status and the number of children, it has a proxy—a variable indicating whether an individual signed up for health insurance for their family, or for themselves, or declined health insurance coverage altogether. If an employee declined coverage, it may indicate they received coverage through a spouse. In each year of our analysis, men were much more likely than women to participate in a family plan. In 1988, women were more than twice as likely as men to have declined coverage, although this gap narrowed substantially by 2007.
Over the period, there was little change in the proportion of federal workers with disabilities. Specifically, 93% of men and 95% of women were classified as having no disability in 1988, whereas in 2007 the respective numbers were 94% and 95%. 13
Empirical analysis of the federal gender -pay gap
To determine whether differences in pay between male and female federal workers are the result of differences in their (a) sex or (b) in other work characteristics that could reasonably be expected to affect pay, we used a statistical technique—the Oaxaca decomposition—that is commonly used in analyses of discrimination. 14 The goal of this technique is to separate the differences in pay between men and women into two components: One that results from differences in characteristics and a second that could result from differential treatment by sex. An advantage of this technique over simply estimating a regression that includes a dummy variable for sex is this approach allows us to see the extent to which each characteristic helps explain the pay difference between men and women, as presented in Figure 1. 15 Further discussion of our empirical methods and an example of how the decomposition method can be used to estimate the portion of the pay gap that is due to differences in particular types of characteristics are presented in the appendix.
Federal work force: Proportion of pay gap due to differences in measurable factors between men and women
Before controlling for differences in the characteristics of men and women, we find that the gender–pay gap declined significantly in the federal workforce between 1988 and 2007. Specifically, as seen in Figure 1 (and in Table 2), the gap declined from 28 cents on the dollar in 1988 to 19 cents in 1998 to 11 cents in 2007.
To perform the decomposition, we controlled for race/ethnicity, disability status, age, federal experience, educational degree, veteran status, whether in a bargaining unit, work schedule, agency, state, and occupation. 16 The results of the Oaxaca decomposition—also displayed in Figure 1—show that, for each of the 3 years we examined, all but about 7 cents of the gap can be explained by differences in work characteristics, especially differences in occupations and, to a lesser extent, differences in education levels and years of federal work experience.
Table 3 provides the detailed statistical results used to create Figure 1. The first row shows (a) the total pay gap, (b) the part of the gap resulting from differences in work characteristics, and (c) the unexplained part of the gap (the part that cannot be explained by differences in work characteristics and therefore could be the result of labor market discrimination or some other factors for which we did not have data). The other rows indicate the contribution that each of the factors that we control for in the model make to explaining the pay gap. 17
Decomposition Results (With Contributions of Key Factors)
Note: Table shows results of what we later refer to as the “main model”.
The overall conclusion drawn from the decomposition approach is that differences remain between men and women’s salaries, even after correcting for a wide range of characteristics. As the table shows, the unexplained gap has been remarkably constant—at 8 percentage (or log) points—over the past 20 years. 18 The proportion of the pay gap resulting from different work characteristics of men and women has been decreasing. For example, the percentage explained by work characteristics was 76% (–.25 / –.33) in 1988 and 41% (–.05 / –.12) in 2007. This is because as women and men have become more alike, less of the pay gap can be attributed to differences between them.
The factor that contributed the most to the pay gap (and its decline over time) was occupation. Specifically, differences in the occupations of men and women accounted for 15 of the 33 percentage point pay gap in 1988, 7 percentage points in 1988, and 3 percentage points in 2007. Other variables that accounted for a relatively large portion of the pay gap in 1988 were education level and years of federal work experience. However, as the differences between men and women in these factors also declined over this period, these factors explained progressively less of the pay gap in 1998 and 2007. In 2007, none of the gap can be attributed to differences in years of federal experience.
Alternate Specifications
To test whether our results changed dramatically depending on what variables we controlled for in the model, we performed the decomposition analysis with variations in the variables in the model. Table 4 below displays the results of variations we analyzed. With the exception of the final model, which does not control for occupation, for most of these models the unexplained pay gap remains relatively level over the study period.
Decomposition Results Using Alternate Specifications of the Model
Note: Table shows results of alternate specifications to the “main model.” Main model specification appears in text.
We estimated two models with more detailed controls for occupation. As explained above, our original model included a relatively rough measure of occupation, with six occupational categories. The two models with “job family” and “job series” used more disaggregated measures of occupation. The “job family” variable controlled for about 50 different occupation categories and “job series” controlled for more than 700 categories. As shown in Table 4, the more precise measures of occupation reduce the pay gap more than the broad measures. Specifically, the most detailed occupational variable reduces the unexplained pay gap from 8% to 5% for all 3 years of analysis.
Similarly, we ran a model with a more detailed control for geographic location of the job (measured at the county level instead of by state). This more detailed control did not have a large effect on the unexplained pay gap, which was the same as the original model for 1988 and 1998 and decreased by just one cent in 2007.
We also estimated the model with an additional variable to control for an employee’s major field of study. Of the 3 years we analyzed, this variable was only available in 2007. It had only a slight impact on the unexplained gap, accounting for an additional 1 percentage point of the gap.
We estimated the model with a proxy for family size—the type of health plan elected by the federal employee. The addition of this variable to the model has almost no impact on the unexplained pay gap. However, the low explanatory impact may be more indicative of the variable’s low correlation with family size than of the impact of having a family on pay. 19
Finally, we estimated a model that only includes variables for the individual’s characteristics and excludes all variables related to the characteristics of the worker’s job (as shown in the last row of Table 4). We did this because some researchers argue that including variables in the model that reflect a joint decision of the employee and the employer, such as a worker’s occupation or part-time status, has the potential to lead to biased estimates. This is the most limited model (insofar as it includes the fewest number of variables). The unexplained gender–pay gap from this model can be thought of as a benchmark or the upper bound of the unexplained gap. Accordingly, in this model, we found the largest unexplained gap and the greatest reduction in the gap over time.
Conclusions
The most important finding from our analysis is that the gender–pay gap in the federal workforce has declined significantly primarily because men and women in the federal workforce are more similar in work characteristics related to pay today than in past years. However, the existence of a persistent unexplained pay gap remains. Interestingly, factors that used to account for much of the gap, like education and experience, no longer do, because differences between men and women in these factors have diminished. However, despite the vast reduction in the clerical workforce in the federal government, which was largely composed of female workers, differences in the occupations that male and female workers hold is still a key contributor to the gender–pay gap in the federal workforce. Gaining a better understanding of how men and women choose occupations within the federal workforce—whether occupational choice is a function of preferences that are correlated with gender or occupational segregation—is a question that warrants additional research.
Furthermore, the existence of a persistent unexplained pay gap between men and women federal workers over a 20-year period, after we controlled for as many factors as our data allowed, means that we cannot rule out the possibility that women are being treated unequally in the federal government. The results of this analysis, however, cannot be used to test for pay discrimination or, more specifically, to determine whether the federal government is in compliance with equal opportunity policy due to several data limitations. First, we lack data on several factors, such as experience outside of the federal workforce, measures of productivity levels, the type of performance evaluation system used within different federal agencies, and personal priorities (such as commitments outside of work). Second, certain variables included in our model—such as occupation, education level, and part-time status—may have been imprecisely measured or reported. Thus, it is possible that by including additional variables in the model or by measuring them more accurately, the unexplained pay gap could be lower or higher. Improved data collection could also benefit future analyses of the gender–pay gap.
Footnotes
Appendix
Author’s note:
Benjamin Bolitzer is an Assistant Director and Erin Godtland is a Senior Economist at the U.S. Government Accountability Office (GAO) formerly known as the General Accounting office. These findings were originally presented in a U.S. Government Accountability Office report, “Women’s Pay: Gender Pay Gap in the Federal Workforce Narrows as Differences in Ocupation, Education, and Experience Diminish,” GAO-09-279. Any errors or opinions are the authors’ and should not be attributed to the GAO or its staff. We are grateful to our colleagues at the U.S. Government Accountability Office, who provided comments on preliminary versions of this paper. They include Michele Grgich, Andrew Sherrill, Tom McCool, Shana Wallace, Doug Sloane, Monique Williams, Dan Concepcion, and Greg Wilmoth. We are also grateful to participants at the Population Association of America annual meetings, who also provided helpful comments.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.