Measuring and Explaining the Gender Wage Gap in the Federal Government

Transcription

1 Measuring and Explaining the Gender Wage Gap in the Federal Government Alexander Bolton John M. de Figueiredo Draft: December 15, 2016 Abstract Although the gender wage gap in the private sector has received substantial attention over the past fifty years, the gender wage gap in the public sector has received less focus in the literature. This paper brings together the largest dataset on public sector employees, covering over 5.6 million individuals during a 24-year period, to examine the size and causes of gender wage gap in the U.S. federal government. We find that the unconditional gender wage gap has been large but steadily declining over the time period. However, after controlling for many factors, the gap is almost half the magnitude of the private sector, and has declined from 6.5% to 3.9% over the past 24 years. Two main factors appear to drive gender wage disparities. First, entry wages for women are less than entry wages for men, and although promotions are similar for both groups, the wage gap grows during employees tenure. Second, the gap is particularly large for administrators and the top percentiles of wage earners. The magnitude of the wage gap is significantly smaller in STEM occupations and traditionally gendered occupations and does not seem to be caused by occupational mobility or mix. Overall, the paper demonstrates that the gender wage gap in the federal government is smaller than in the private sector, but substantial pockets of wage differences persist. This research was financially supported by the National Science Foundation (Award # , , and ). We would like to thank Tom Balmat, John Johnson, and Sam Rosso for research assistance. This is a preliminary and incomplete draft please do not quote, cite, or distribute without permission. Assistant Professor, Department of Political Science, Emory University, Atlanta, GA 30322; www. alexanderbolton.com, abolton@emory.edu Edward and Ellen Marie Schwarzman Professor of Law and Professor of Strategy and Economics, Duke University, Durham, NC 27708; National Bureau of Economic Research; jdefig@duke.edu.

2 1 Introduction An extensive literature has shown the gender wage gap to be pervasive in the private sector (Blau and Kahn 2016). While the literature on the public sector is less robust, there is a consensus that a gender wage gap exists in government employment as well (Gregory and Borland 1999). The literature has identified a number of reasons for potential pay disparities between men and women. First, differences in individual observable characteristics, such as education, age, work experience, and race have all been shown to affect the magnitude the wage gap (Altonji and Blank 1999; Blau and Kahn 2006; Blau, Ferber and Winkler 2014, Chapter 8). Second, more women than men tend to participate the in the workforce as part-time or seasonal workers, having more interruptions in their work. This may affect the gender pay gap (Blau and Beller 1988; Blau and Kahn 2006; Mulligan and Rubinstein 2008). Third, women tend to be in different occupations than men, occupations that pay different wages. Previous authors have argued that women, for example, tend to be highly represented in fields such as health, education, and welfare, and poorly represented in STEM and regulatory occupations (Blau, Ferber and Winkler 2014; Ginther, Kahn, and Williams 2014; England and Li 2006). Fourth, the literature has argued that women encounter positional segregation, gaining entrance to the private and public workforces through entry-level positions, such as clerical positions, and facing limited promotion opportunities and encountering glass ceilings in the government (Lewis 1986; Alkandry and Tower 2006; Guy 1993; Newman 1994). This paper addresses all of these issues, measuring the effect of these factors on the gender wage gap in the U.S. Federal Government over a 24-year period. Using a dataset of over 5.6 million full time, non-seasonal, federal workers, and their career histories from , we bring to bear the most extensive data on the gender wage gap in governments. The paper begins by estimating the gender wage gap. Numerous studies have documented a gender wage gap in public sector employment (Lewis 1996; Gunderson 1989; Baron and Newman 1989; Sorenson 1989; Wharton 1989; Bridges and Nelson 1989). (The public sector represents 14% of U.S. employment and a higher percentage in other countries). The gender wage gap has been estimated to be between 8% and 13% in the public sector. One of the most comprehensive 1

3 recent analysis for the U.S. federal government is a General Accountability Office (GAO) (2009) study. Using a series of cross-sectional regressions from 1988, 1998, and 2007 using a random sample of Office of Personnel Management (OPM) data, this study found that the average gender pay gap declined from 4.6% in 1988 to 4.5% in 2007 (p. 65) after controlling for human capital and other factors. This study controls for observables, such as education, age and experience, race/ethnicity, agency, and detailed occupational information. OPM itself published a study of the years 1992, 2002, and 2012 and estimated that by 2012, the gender wage gap in the federal government was 13%. 1 We replicate and extend the analysis of this study and Blau and Kahn (2016) including all federal workers from 1988 to We use a variety of models to estimate the wage gap over time. We show that gender wage gap has decreased from 6.5% in 1988 to 3.9% in (The gap was also about 3.9% in 2007). Overall, when controlling for observable individual, occupational, and agency characteristics, the wage gap is estimated to be 4.5%. This is almost half of the size of the 8.4% gender wage gap found in the U.S. as a whole by Blau and Kahn (2016: 68). 2 One of the largest concerns in the literature in identifying the gender wage gap is that occupations may mask the true effect of wage discrimination against women (Blau and Kahn 2016; Duncan and Duncan 1955; Blau, Brummund, Liu 2013a,b; Groshen 1991). This is often argued as the main source of discrimination. We conduct an extensive analysis of occupations, occupational categories, men s and women s occupations, and STEM occupations. While generally decreasing across time in all analysis, the gender wage gap is substantially different across occupational categories. Women enjoy a substantial wage advantage in clerical positions, but a substantial wage disadvantage in administrative, professional, technical positions, and blue collar positions. For example, the gender wage gap in the fully specified model ranges from 7.5%-8.6% for administrative employees. 1 The discrepancy in the estimates between GAO and OPM largely comes from the ways in which occupation are treated in the analysis. We discuss this further below. 2 There are reasons one might expect the government would have a lower gender wage gap than the private sector. Governments are more affected than private firms by procedural fairness that makes it harder than the private sector for managers to favor men over women in wage setting. In addition, the public sector has many unions, which will tend to standardize wages across seniority, and thus allow there to be a smaller wage gap. 2

4 A second category source of the gender wage gap in the literature is gendered occupations (Levanon, England, and Allison 2009; Blau 1977, Lewis 1996). Gendered occupations are those in which women (or men) are substantially over-represented in certain occupations. We examine the effect of gendered occupations (with less than 25% of workers of one gender) and compare them to gender neutral occupations (25%-75% of workers from one gender) and show that the gender wage gap is 5% for gender neutral occupations (27 million observations), but 4% for male occupations (14 million obs) and -1% (women are at an advantage over men) in female occupations (7 million obs). We also note that the percentage of employees working in gendered occupations has declined substantially over the past 24 years, from 64% to under 40%. This has occurred simultaneously with a decline in the gender wage gap of gender neutral and male occupations, as well. STEM occupations have also been identified as a location where one finds a substantial gender wage gap in the literature. In the government data analyzed in this paper, however, we see a slightly smaller gender wage gap than average. STEM occupations have also seen a decline in the gender wage gap from 4.2% in 1988 to 3.3% in What causes the largest gender wage gap in the U.S. Federal Government? There are two factors. First, a cohort analysis demonstrates that the wage gap expands over the tenure of a cohort. Controlling for other factors, women enter the government on average approximately two steps lower than men. 3 This means that women s salaries are on average 0.5% to 3.5% lower than men s upon entry depending upon the year. After entry, women receive similar percentage raises and are promoted with roughly the same frequency and same magnitude as men. 4 However, because women start on a lower base salary than men, roughly equal promotion rates combined with equal cost-of-living wage increases for both groups cause the wage gap to increase (in percentage terms) over the tenure of employment. 3 The grade and step system in the U.S. Federal Government General Schedule system is a way to assign rank and salary to each employee. There are fifteen grades (ranks), GS1 to GS15. Within each grade there are ten steps (Step 1 to Step 10). Each higher grade means a promotion and higher salary. Salaries are also increasing within grades for each step. 4 Another source of wage variation under investigation by researchers is promotions (Noonan, Corcoran, and Courant 2005, Blau, Ferber, and Winkler 2014: Chapter 7; Gayle, Golan, and Miller 2014). They suggest women are less likely to be promoted than men; thus the gender wage gap effects are masked by limited promotion opportunities and glass ceilings. 3

5 Second, quantile regressions demonstrate that while the gender wage gap in the government has been declining across all parts of the wage distribution, the highest percentiles of the wage distribution account for the largest gender wage gap. The 90th percentile of the wage distribution, heavily populated with administrators, has a gender wage gap which is 20-27% larger than the 50th percentile. 5 This finding is largely consistent with the literature on the private sector (Blau and Kahn 2016; Kassenboehmer and Simming 2014). What is different about the government sector is that the very top of the wage distribution, where the gender wage gap is the largest, is also the part of the distribution with the largest percentage of political appointees present. On average, male political appointees make 8.7% more than female appointees, even after controlling other factors. Hence, political appointments seem to drive the gender wage gap as well. Overall, the paper shows that the gender wage gap in the U.S. Federal Government is 3.9% in 2011, substantially less than the gender wage gap for the overall country. The highest gender wage gap seems to occur because of 1) lower entry wages for women which creates a wage gap than gets larger over time, and 2) a gender wage gap at the very top percentiles of the wage distribution, particularly concentrated among administrative employees. The remainder of the paper is organized as follows. In Section 2, we describe the data employed to conduct our analysis. In Section 3, we estimate the average gender wage gap. In Section 4, we tackle the question of occupations. In Section 5, we estimate the gender wage gap across the wage distribution. The analysis in Section 6 examines the gender gap for employees in supervisory and agency leadership roles. Section 7 focuses on career dynamics, including the starting wages for women relative to men, differences in promotion propensities, and cohort analyses. Finally, in Section 8, we summarize our findings and conclude. 5 While the Senior Executive Service exhibits almost no gender wage gap, supervisors do exhibit a much larger wage gap. 4

6 2 Data In order to examine the underlying dynamics of wage growth in the United States government, we use the Office of Personnel Management s (OPM) Central Personnel Data File and Enterprise Human Resources Integration (CPDF-EHRI). This dataset contains employee records of all nonsensitive civilian employees employed by the U.S. Federal Government from 1988 through The dataset contains information on employee careers (wages, work schedules, awards earned, supervisory status, receipt of monetary incentives, occupation, supervisory status, etc.) and their individual characteristics (gender, race, educational background, geographic location, etc.). There 5,609,493 unique full-time, non-seasonal employees in over 42 million observations of data. Individuals work in 381 different agencies and 874 identifiable unique sub-agency organizations. This dataset is substantially larger than most other papers which employ this dataset, as others usually rely on 1% or 10% samples of parts of the data (Borjas 1983, Lewis 1996, Katz and Krueger 1991; GAO 2009). With the larger dataset, we are able to examine relatively small segments of the workforce in detail without concerns about sampling error. We are also able to generate substantially more power from our statistical tests and conduct analyses which are difficult to assess with substantially smaller datasets. Moreover, our data is both cross-sectional and longitudinal, at one year intervals, allowing us to link individuals and their career progressions over time so we can examine the dynamics of wage growth in the context of an overall career. The average age of employees is 44.4, and, on average, they have 3.53 years of education beyond eleventh grade. The median wage for all employees was $55,826 over Further summary statistics for the main variables used throughout the paper can be found in Appendix A. 3 Estimates of the Federal Gender Wage Gap We begin by estimating the overall gender wage gap in the federal government in Table 1. In later sections, we break this gap down to examine how it varies across subgroups. In all of our analyses, the measure of pay is based on an employee s annual basic pay, which is defined by 5

7 OPM as: The employee s rate of basic pay. Exclude supplements, adjustments, differentials, incentives, or other similar additional payments. Thus, this measure of pay does not include locality adjustments, for instance, that tie individuals pay to their geographic locations or payments for bonuses. All dollar amounts have been converted to September 2011 dollars. Overall, we find that there is significant convergence in male and female federal wages over time. Figure 1 below displays the median annual wage for both men and women. Both groups saw substantial growth over the last twenty five years, though women s wages show significantly faster growth. In particular, over the period , the female median real wage grew nearly 50%, rising from $38,300 in 1988 to $56,991 in During the same period, the comparable statistic for men grew about 12%, from $55,296 in 1988 to $62,777. [Figure 1 about here.] As another measure of women s increasing pay, we can also examine the gender breakdown of the top decile of wage earners in each year over the period of our study. Women have increasingly made up an increasing share of this group, but there is still a significant skew toward men. Figure 2 shows that in 1988, only 11.4% of the employees in the top decile of the wage distribution were female. By 2011, this number had nearly tripled to 32.8%. [Figure 2 about here.] While these broad summary statistics suggest a lessening gap in wages between men and women over time, they do not control for a number of important factors, including human capital, demographic variables, and the types of work employees do that might differ across men and women. In order to more rigorously measure the gender wage gap and its dynamics, we conduct five different regression analyses in Table 1. Model 1 is a simple regression of logged wages onto an indicator for whether or not an employee is female. Model 2 includes a battery of human capital related variables in the specification, including an employee s age (Age); their years of education after 11th grade (Education); an indicator for their race, as defined by OPM: American Indian or 6

8 Alaska Native (AI/AN), Asian (Asian), Black (Black), Hispanic (Hispanic), or White (omitted); and a variable that captures an employee s length of tenure government as well as the square of their tenure (Tenure and Tenure 2 ). This model closely tracks human capital regression models used in the larger gender wage gap literature (see Blau and Kahn 2016 for a review). Model 3 adds indicators for an individual s bureau as well as the year (Bureau FE and Year FE) to the human capital variables in Model 2. Models 4 and 5 estimate the gender gap controlling for occupation in different ways. Controlling for occupations is a fraught issue in the gender wage gap literature (see Blau and Kahn 2016 for an extensive discussion). In essence, we might think of entrance into occupations as being a causal consequence of an individual s gender. In particular, women have long faced discrimination even entering into certain occupations. Thus, by controlling for occupation in a very specific way, we may understate the extent of the gender wage gap. However, at the same time, the types of work that individuals do has important implications for their pay and we would also like to answer the question of whether there is a wage gap for individuals that are similarly situated in terms of human capital and doing similar types of work. For this reason, we examine the types of work in which employees are engaged with two different types of indicators of occupation. First, in Model 4, we use OPM s occupational category variable. This divides occupations into six broad, aggregated categories: professional, administrative, technical, clerical, other white collar, and blue collar occupations (Occ. Cat. FE). Then, in Model 5, we use an extremely disaggregated measure of occupation, that divides workers into 817 distinct occupations (Occupation FE). Finally, we also performed a two-fold Oaxaca-Blinder decomposition for each of the specifications. This decomposition provides insights into how much of the overall gender gap in the sample is explained by the variables in a given specification (and their differences in levels across men and women) and how much is left unexplained. We report the percentage of the gap that is unexplained at the bottom of Table 1. It is important to note that the percentage of the gap that is unexplained may be attributed to a number of different factors, including unobserved variables or possibly discrimination. 7

9 [Table 1 about here.] The results of these initial analyses are displayed in Table 1. As can be seen, across all five specifications, there is a persistent gender gap, with women earning noticeably less than men. The unconditional results suggest that overall, women, on average, earn about 18% less than men in the federal government. Moving to Model 2, we find that after controlling for age, education, race, and tenure in the federal government, the gap shrinks to about 11%. Notably, the unexplained difference between male and female wages drops considerably, to 55%. This is actually fairly low relative to other attempts to measure the gender wage gap in the broader economy. After controlling for similar variables, Blau and Kahn (2016) find that between 71.4% and 85.2% of the gap remains unexplained. The gap decreases further after including indicators for an employee s bureau and the observation s year, dropping to about 10%. It shrinks considerably, however, once we take into account the types of work that an employee does. In Model 4, with broad occupation indicators, we find a 6.9% gender wage gap and that just over half of the gap is explained by the variables included in the specification. This is smaller than the gap estimated by GAO (2009), which reports a gap of 10.9% in 1988 and 8.3% in 2007 using a similar model with the same level of occupational aggregation. When we include the detailed occupation indicators in Model 5, the gap decreases by a third relative to Model 4, to 4.5%. Furthermore, only 20.5% of the gap remains unexplained after taking into account this disaggregated occupation information. For comparison, a similar model reported by Blau and Kahn (2016) for the broader American workforce found that between 38 and 48.5% of the gap was unexplained. In addition to measuring the gender wage gap on average over the period we examine, it is also important to examine temporal trends in the gap. Reports on the larger economy suggest that the wage gap has decreased substantially over time. We find a similar result in the federal government. In Figure 3, we plot the estimated coefficient for the female indicator in Models 1 and 5 over the time period of our study, [Figure 3 about here.] 8

10 As can be seen, the gap narrowed significantly over the time period of our study. The overall unconditional gap has been more than halved, from 27.8% in 1988 to 10.1% in We see a similar story when we examine the estimate from the full model. There, the estimated gap decreased 40% from about 6.5% in 1988 to 3.9% in These results are in line with the earlier descriptive statistics presented, which suggested that the percentage of top decile earners who were women increased substantially over this time period. This trend toward a decreasing gender gap is consistent with those reported in Lewis (1998), GAO (2009), and OPM (2014). However, it is notable that the decline in the wage gap has slowed over time as well, at least in the case of the full model. One concern that may arise from this analysis is employee departures. Over the long-run, selected types of employees may depart, creating an upward or downward bias in the wage gap over time. This bias occurs if departures are systematic (highly successful women/men or highly unsuccessful women/men). In order to control for this effect, we also conducted another analysis to assess differences in year-to-year wage growth for men and women in the federal government. In particular, we estimated a series of year-by-year regressions in which the dependent variable in the analysis was the difference in logged basic pay for an employee from the previous year. The logic here is that between any two years, the profile of employees is very similar and systematic departures are less likely to affect the coefficient estimates. In addition to the female indicator used in the models above, we also included indicators for an employee s race; an indicator for an employees lagged grade and step (since year-to-year wage growth differs across the wage distribution); both lagged values and differenced values of an employee s educational attainment beyond eleventh grade; lagged age; lagged tenure (and its square); lagged bureau; and, finally, a lagged indicator an employees detailed occupation. The estimated percentage point difference in year-to-year wage growth for women relative to men is plotted in Figure 4. Note that the analysis starts in 1989 given that it requires lags from 1988, the first year in our dataset. First, in terms of magnitude, the differences in year-to-year wage growth for men and women is very small, never estimated to be more than 0.15 percentage points. 9

11 While the difference appears to be decreasing from 1988 through the mid-2000s (with women actually experiencing higher year-to-year growth in some years), more recently the trend has been downward. However, the differences that we estimate are fairly negligible on a year-to-year basis (though they could, of course, compound over time in a way that makes them more substantial). [Figure 4 about here.] Overall, then, these results suggest a convergence in men and women s wages during the period of our study, though there is a persistent and significant gap in earnings in all of the years we study, whether we look at an unconditional difference or one that includes both human capital and detailed occupation indicators. In the following sections, we seek to provide a more detailed characterization of the gender wage gap in the federal government across the wage distribution and among different sets of occupation and employees. 4 Occupations and the Gender Gap We begin our deeper exploration of the gender wage gap by describing variation across the types of work in which employees are engaged. Different types of occupations require different levels of human capital and experience. This variation should have predictable effects on the wages of employees. In order to better understand how the gender wage gap differs across types of work, we examine each of the aggregated occupational categories used in the analyses above separately. Figure 5 graphs the results of this analysis. We examined the effects of occupation in two ways. First, we ran the equivalent of Model 5 in Table 1 (including detailed occupation indicators) for each broad occupational category separately. These estimates are denoted with solid circles in Figure 5. Second, we ran the models for each occupational category separately with no additional occupation variables. The estimated female indicator variables is plotted with empty triangles in Figure 5. [Figure 5 about here.] 10

12 As can be seen there is substantial variation in the gender wage gap across occupational categories. Across both model specifications, administrative occupations are among those with the highest gender gaps, about 9% in the model without the detailed occupation indicators and 7.5% in the model with these indicators. These occupations are the most common, making up 31.8% of all full-time non-seasonal employees from , and among the most highly paid in the government, suggesting that they are an important source of the overall gap observed in Section 3. Administrative occupations do not necessarily require a four-year college degree, but they do tend to require skills that can be attained at that educational level. The three largest occupations in the administrative category are miscellaneous program and administration, management and program analysis, and criminal investigating. Blue collar occupations, which tend to be dominated by men (less than 10% of blue collar employees in the dataset are female), also appear to have a larger than average gender gap (10.7%), particularly in the models without detailed occupation information. However, the estimated gap shrinks to 2.4% when we control for an employee s detailed occupation within the blue collar category. Furthermore, the proportion of blue collar employees in the government is relatively small (only 14% of all employee observations), suggesting that they may be of limited influence in the overall estimated gender gap. Clerical positions are distinctive in that women, on average, actually earn between 1.5 and 5.3% more than men in these positions, even after controlling for employee human capital and other demographic information. However, clerical positions are relatively limited in number in the government (making up just 10.1% of all employee-years), and they have been declining in number over time. The fourth occupational category other white collar employees shows the smallest magnitude gender gap of any occupational category. With the detailed occupation indicators included in the model, women are estimated to hear 1% more than men each year, though this effect reverses when these additional occupation controls are excluded (to a 1.5% wage disadvantage for women). It is a relatively small category, with a variety of jobs ranging from human resources management 11

13 to emergency management specialists to student trainees across a wide range of academic disciplines. Overall, only about 3% of employees from were part of this category, suggesting that it is not hugely influential in the overall governmental analysis. Professional occupations are among the most highly skilled in the government, requiring substantially higher levels of education than other categories. The three largest professional occupations are nurses, contracting, and attorneys. Most STEM occupations also tend to be in the professional category. Overall, professionals make up 23% of employees in the period In the model without disaggregated occupational indicators, the estimated gender gap for professional employees, 6.5%, is slightly smaller than the estimate in Model 4 of Table 1. In the model with detailed occupation indicators, the estimated gap is only about two thirds of the average estimated gap in Model 5 of Table 1: 2.9%. Finally, the wage gap for technical employees is estimated to be just 2.2% in the model with detailed occupation indicators and as high as 9.3% (above the overall government average) when those indicators are not included. Technical positions are a fairly large proportion of federal jobs (18.8%) and are also relatively low human capital. These jobs do not typically require college degrees. The three largest technical occupations are miscellaneous clerk and assistant, engineering technical, and contact representative. In addition to examining the average wage gap within occupations, we can also examine whether the gender gap temporal dynamics that we observed in Figure 3 are constant across occupational categories. Figure 6 displays the estimated coefficient for the female indicator within each occupational category for each year of the study. Note that these results are from models that do not include the additional detailed occupation indicators, i.e. they are the equivalent of Model 3 in Table 1 estimated within occupation-years. [Figure 6 about here.] The results from this analysis for most occupations (with the exception of clerical occupations) largely mirror the dynamics that we observed in Figure 3. There is a significant decrease in the 12

14 gender wage gap during the first part of the study through the mid-1990s, and the narrowing slows after that. For clerical occupations, however, women s wage advantage over men actually shrinks over time, from 4.9% in 1988 to 3.2% in Overall, the trend across all occupations has been toward convergence in the wages of men and women. We can also look at a less aggregated version of occupation to get more of a sense of how the gender gap varies across different types of work in the federal government. In particular, we use the Office of Personnel Management s two-digit occupational codes to assess this question. This yields 59 occupational groups that correspond to the types of work carried out by an employee. See the appendix for a full list of these codes. We ran separate wage regressions for each of these occupational groups, regressing logged basic pay onto a female indicator variable, as well as including controls for race, age, education, tenure, bureau, and year. [Figure 7 about here.] The results of these regressions are plotted in Figure 7. In particular, we plot the estimated coefficient for the female indicator as well as 95% confidence intervals. Additionally, we include the number of observations in each occupational category on the right axis. As can be seen, there is wide variation in the gender gap across these different occupational categories. Particularly notable, is the fact that the largest gender gaps appear to be among white collar occupations (i.e. those with a two digit occupational code less than or equal to 22). This more or less comports with the large gender gaps we found above for administrative occupations. These suggest that there may be relatively high levels of gender wage disparities among the highest earners in the government. We now turn to to another line of inquiry whether or not women fare better in occupations that are predominantly female. In particular, as the results by occupation group demonstrate, women actually made more than comparable men in clerical positions. This is significant because 82% of clerical employees over the time period we examine were female. This connects to a larger hypothesis in the gender wage gap literature that traditionally" female occupations yield better career and wage outcomes for women. 13

15 In order to test this in the context of the federal government and evaluate it as an explanation for the gender wage gap in the federal government, we divide occupation-years into three different groups: predominantly female, in which more than 75% of employees in the occupation in that year are female; predominantly male occupations, which have fewer than 25% female employees; and gender neutral occupations that have between 25 and 75% female employees. Table 2 lists the largest occupations within each of these three categories. The predominantly female occupations tend to be ones that are traditionally female for example, clerical and secretarial work, nursing, and typing. Similarly, predominantly male occupations in the government tend to skew toward stereotypically male work, such as law enforcement and engineering. [Table 2 about here.] Over time, the percentage of federal workers in predominantly male or female occupations has declined fairly significantly. Figure 8 displays the percentage of individuals within each of the three occupation types over the course of our study. In particular, in 1988, there were actually more employees in male-dominated occupations than in neutral ones. However, by 2011, more than 60% of federal employees were working in gender neutral occupations, while just about 30% were in male-dominated ones and 10% in female-dominated occupations. Just as wages have converged for men and women, so has the type of work carried out by both genders in the federal government. [Figure 8 about here.] In order to characterize the gender gaps in each of these occupational categories, we reran the five regression analyses on each of the three occupational groups we have identified. The results of these analyses are reported in Tables 3, 4, and 5. First, beginning with traditionally female occupations, it indeed appears to be the case that women do fare better in terms of wages. In particular, the results suggest that women earn between one and five percentage points more than comparable men in these occupations, depending upon the specification. This largely tracks the results above for clerical occupations, in which women are extremely over-represented. However, 14

16 it should be noted that the proportion of employees working in predominantly female occupations is overall quite low about one-sixth of all observations in the dataset. [Table 3 about here.] [Table 4 about here.] [Table 5 about here.] The results for predominantly male occupations show a consistent negative gap for female employees. Notably, however, this gap is actually smaller than the average gap estimated in Table 1 or the gender gap for gender neutral occupations reported in Table 5. Indeed, women on average earn 4% less than man in predominantly male fields but 5% less than men in gender neutral fields. One explanation for this may be that there are positive selection effects. In fields where women are discriminated against in terms of entry, those that do choose to enter the field and are able to secure employment may be high quality and perform exceptionally. However, if this is the case, then that would suggest that this is an asymmetric effect across genders, because men perform worse in terms of wages in female-dominated fields. The over time trends in the gender wage gap for these three occupational groups largely mirror the broader trends that we found for the government as a whole. Figure 9 displays the year-by-year estimates of the gender gap for employees in each of the three groups. For predominantly male and neutral occupations, we see familiar (and parallel) trends. The gap decreased steadily until the mid-1990s at which point the narrowing levels of off fairly significantly. We do not see the same trend, however, for female-dominated occupations, which with the exception of 2001, hover fairly steadily around 1%. [Figure 9 about here.] Finally, to conclude our examination of occupations and the gender wage gap, we examine one particular example of an occupational group where women s representation has lagged science, technology, engineering, and mathematics (STEM) occupations (Ceci, Ginther, Kahn, and 15

17 Williams 2014). We estimate our five standard regression models to characterize the gender wage gap in this particular, critical field of work. Women made up only 19.9% of STEM employees during the period of our study. However, it is not clear whether this underrepresentation is also associated with a larger-than-average wage gap given the results reported above. Indeed, it seems that it is not. We ran the standard five regression analyses on the subset of employees working in STEM occupations as designated by the Office of Personnel Management. 6 The results of these analyses (reported in Table 6 largely comport with the findings above about gender-segregated occupations. In particular, we find that the gender wage gap is actually smaller in STEM occupations than on average across occupational groups in the government. For instance, the unconditional difference in wages is about half that of the government as a whole 10.3%. Furthermore, in Model 5, which contains detailed occupational indicators, the estimated gap (3.2%) is more than a quarter lower than that of the government as a whole. The over time trend in the gender wage gap for this class of occupations largely parallels the government-wide trends (see Figure 10). [Table 6 about here.] [Figure 10 about here.] 5 The Gender Wage Gap Across the Wage Distribution In this section, we examine whether the gender wage gap and its dynamics are consistent across the entire wage distribution. In order to do this, we begin by examining quantile regression analyses 6 These occupations include: general natural resources management and biological sciences; microbiology; pharmacology; ecology; zoology; physiology; entomology; toxicology; botany; plant pathology; plant physiology; horticulture; genetics; rangeland management; soil conservation; forestry; soil science; agronomy; fish and wildlife administration; fish biology; wildlife refuge management; wildlife biology; animal science; general physical science; health physics; physics; geophysics; hydrology; chemistry; metallurgy; astronomy and space science; meteorology; geology; oceanography; cartography; geodesy; land surveying; information technology management; general engineering; safety engineering; fire protection engineering; materials engineering; landscape architecture; architecture; civil engineering; environmental engineering; mechanical engineering; nuclear engineering; electrical engineering; computer engineering; electronics engineering; bioengineering and biomedical engineering; aerospace engineering; naval architecture; mining engineering; petroleum engineering; agricultural engineering; chemical engineering; industrial engineering; general mathematics and statistics; actuarial science; operations research; mathematics; mathematical statistics; statistics; cryptanalysis; and computer science. 16

18 of the gender wage gap. While the previous analyses reported in Table 1 modeled mean wages as a function of gender and other variables, we now turn our attention to examining different parts of the wage distribution. In particular, we ran quantile regressions modeling the 10th, 50th, and 90th percentiles of the distribution in order to examine whether the differences in conditional mean wages for men and women are replicated both in magnitude and trend across the wage distribution. To begin, we can examine the unconditional quantiles of the wage distribution for men and women and how they have changed over time. Figure 11 plots the 10th, 50th, and 90th percentile wages of the wage distributions for men and women in the federal government. Overall, we see that across the wage distribution, men earn significantly more than women, and that this difference appears to be most pronounced at higher levels of the distribution. For instance, the 90th percentile of the female wage distribution was 12% less than the 90th percentile of the male distribution in 2011, whereas at the 50th and 10th percentiles the difference was 10% and 7% respectively. The gap between men s and women s wages do appear to have narrowed across the wage distribution over time, particularly at the top. [Figure 11 about here.] In order to more rigorously characterize the gender across the wage distributions, we run a series of year-by-year quantiles regressions at the 10th, 50th, and 90th percentiles. The results of these analyses are presented in Figure 12. The models included the same covariates used in Model 5 in Table 1 individual human capital data, demographic variables, agency fixed effects, year fixed effects, and detailed occupation indicators. The models were estimated using the Frisch- Newton algoritm developed by Koenker and Ng (2005) for sparse quantile regression. [Figure 12 about here.] The results of these analyses largely mirror the results for the government as a whole as well as the unconditional quantiles discussed above. Across all parts of the federal government s wage distribution, women earn significantly less than comparable men. The gaps identified in Figure 11 17

19 become smaller, but are not eliminated after including controls for human capital, demographics, organizations, and the types of work carried out by employees. Furthermore, the familiar trend identified earlier in the paper holds here as well. Across the whole wage distribution, women gained until about the mid-1990s, at which all three lines hit an inflection point, with much slower growth in relative wages for women. 6 Supervisors and Executives We now turn our attention the gender wage gap among top executives and supervisors across the federal government. In particular, we examine two groups of high-level officials in the government individuals with designated supervisory roles and members of the Senior Executive Service. OPM identifies seven types of groups with supervisory status during the period of our study, some of which changed over time. These groups are: Supervisor ( ) Manager ( ) Supervisor or Manager ( ) Supervisory Position as designated by the Civil Service Reform Act ( ) Managerial Position as designated by the Civil Service Reform Act ( ) Leader ( ) Team Leader ( ) In general, positions designated as supervisory or managerial under the CSRA tend to have less responsibility and oversee fewer employees than those that are classified in this way by OPM in the first three categories. Leaders and team leaders have the smallest purviews of authority and tend to be in charge of small groups of employees. The Senior Executive Service (SES) was created by the Civil Service Reform Act of The SES was designed to be an elite cadre of administrators, and its members undergo extensive quality vetting by both OPM and the hiring agency when they are appointed. Members of the SES occupy 18

20 some of the top managerial positions and policy-determining roles across the federal government. Across both groups of employees, women have made enormous strides in terms of representation during the period of our study. Figure 13 plots the percentage of each group that is made up of women over time, demonstrating this point. [Figure 13 about here.] We can also examine the gender wage gap within each of these groups and their temporal dynamics using the same tools as above. First, we run the same five regression models for the Senior Executive Service as we did for the government as a whole. The results of those regressions are reported in Table 7. As can be seen, the gender wage gap is significantly smaller for SES employees than for the government as a whole. In the full model (i.e. Model 5), the gap is estimated to be just 0.6%. While statistically significant, substantively this gap is very low relative to the rest of the estimates we have reported in this paper. This gap has not been substantively high over time either. Figure 14 plots the estimated year-by-year gender wage gap for SES employees. As can be seen it reaches its largest point in 1994, when it is estimated to be about 1.7%, but for most of the period of our study, it hovers at less than 1% and in some cases is actually indistinguishable statistically from zero. [Table 7 about here.] [Figure 14 about here.] A similar story holds for supervisors in the government as well, though there is some variation across the different types of supervisory status. Figures 15 and 16 plot the estimated gender wage gap for the full model for each of the seven supervisory categories over the period and year-by-year, respectively. As can be seen in Figure 15, leaders and team leaders have by far the smallest gender wage gaps, with both hovering near 1%. The largest observed wage gaps appear to be among managers and supervisors. However, recall that these categories also existed only during the period , at which point they were collapsed into one category (i.e. managers and 19

21 supervisors ). Thus, this appears to be a function of the more general over time convergence in male and female wages. [Figure 15 about here.] [Figure 16 about here.] Indeed, the trends displayed in Figure 16 appear to confirm this. With the exception of the leader category, all other supervisory categories have shown a move toward smaller wage gaps, which occurred at an accelerated pace until the mid-1990s. The leader category, appears to show a different trend. Women fare the best relative to men in this category across the entire time period, but the trend has been toward a larger gender wage gap for leaders over the period of this study. Finally, we turn our attention to political appointees, tend to hold the highest level positions in government and tend to be among the highest paid employees, with a mean real wage of $116,900. Fully 35.5% of political appointees are in the top 1% of the wage distribution, and 64.4% are in the top 10%. Comparatively, however, political appointees are a relatively small group, making up 5.4% of the top percentile of the wage distribution and 1% of the top 10%. There are three types of political appointees PAS appointees, who require confirmation by the Senate; non-career members of the Senior Executive Service; and, what are know as Schedule C appointees (Lewis 2008). Table 8, below, reports the five basic models we have used to measure the gender wage gap. As can be seen, the gender wage is significantly higher among political appointees than for overall government. The unconditional gap is 22%. In Model 5, which includes disaggregated indicators for occupation, we estimate a gap of 8.7%, significantly larger than in the government as a whole. Such a large gap among very high level appointees may help to explain at least some of the relatively high estimated gap at higher levels of the wage distribution, as reported in the quantile regression analyses. [Table 8 about here.] 20

22 7 Career Dynamics Finally, we turn our attention to examining the career dynamics for male and female employees in the federal government. In particular, we investigate their starting points to see whether there are differences in the starting wages of comparable men and women. Then we examine whether there are differences in the propensities of men and women to be promoted once in the government. In order to conduct these analyses, we focus on the largest pay system in the federal government the General Schedule (GS). The GS covers 69% of the employees in our dataset, however, it does exclude blue collar workers. The GS system is comprised of fifteen grades, that are increasing in pay, and ten steps within each grade. In our study, a promotion is defined as moving up in grade. We begin by examining the relative initial entry points into the GS scale for men and women. To conduct this analysis, we created a continuous scale of all the 150 possible step-grade combinations. This is the dependent variable in the analysis. We then used the five regression model specifications as in the gender wage gap analyses to characterize the gap in initial GS positions of comparable men and women on the GS scale. 7 The results of this analysis are recorded in Table 9. The sample for this analysis is new GS employees during the years because we do not observe the starting grade-step for individuals in the dataset in [Table 9 about here.] The results of this analysis suggest that there is a persistent gap in the starting positions of male and female employees that are part of the General Schedule pay system. Depending upon the specification, women start between 1.68 and 3.11 steps below men doing similar work and with the same levels of human capital. These differences can have a significant impact on pay. For instance, in the mean grade level (9), a difference of three steps, from say step 1 to 4, has a pay difference of ten percentage points. This initial lower position for women on the GS scale can have significant, career-long implications for their pay relative to men. These results correspond with the findings by the Office of Personnel Management (2014), which found similar results. They 7 Note that the tenure variables are omitted because we are examining the first year for employees. 21