The Impact of Veterans’ Preference on the Composition and Quality of the Federal Civil Service

Abstract

Government preference for military veterans in hiring, a means to honor them for their service and sacrifices, potentially conflicts with other values of the public service, including its diversity and quality. Census data for 1990, 2000, and 2006–2009 show that veterans are at least three times as likely to hold federal jobs as, but only 10% more likely to hold state and local government jobs than, comparable individuals without military service. Veterans’ preference has substantially increased the percentage of federal employees who are men and has probably decreased the percentages who are Asians, gay men, and immigrants, but its effects on the composition of state and local governments are small. Federal personnel data for the past decade show that veteran new hires are older and less educated than nonveteran new hires, and that they do not advance as far in the first 15 years of their careers as nonveterans hired into the same grades at the same time, suggesting that veterans’ preference may be lowering the performance of the federal civil service.

Abstract

《退伍军人优惠政策对于联邦公共服务质量的影响》

作者：格雷戈里B. 刘易斯

退伍军人在报考美国政府公务员时可享有一定的优先级 (veteran preferences)。政府通过此类优惠政策来认可和奖励他们为国家作出的牺牲和贡献。然而退伍军人优惠政策可能会造成与其他公共服务价值理念的冲突，因而不利于维持政府部门人员的多样性和公共服务的质量。1990、2000和2006至2009年的人口普查数据表明，在其他条件均等的情况下，退伍军人被联邦政府录用的机率是普通报考者的三倍。而州和地方政府中退伍军人公务员的比例只比普通公务员多百分之十。退伍军人在报考公务员中的优惠政策导致联邦政府中男性雇员的比例大幅增长，而女性、亚裔、同性恋和移民雇员的比例则可能会减少。这一优惠政策对于州和地方政府却影响甚微。过去二十年联邦政府的人事数据表明，联邦政府雇用的退伍军人公务员通常比普通公务员的学历低。他们在联邦政府中工作的前十五年内晋升的空间也小于普通公务员。这些比较数据意味着，退伍军人优惠政策可能会降低联邦服务的质量。

Commonly recognized goals of the federal personnel system include increasing the quality, equity, diversity, representativeness, responsiveness, and managerial effectiveness of the civil service. One-quarter of federal employees, however, are hired through a mechanism designed to fulfill a different goal: to recognize and reward veterans for their service to and sacrifices for the nation. Fulfilling this goal may require trade-offs with other goals. By explicitly preferring a group that has traditionally been very disproportionately male, white, native born, and heterosexual, veterans’ preference creates obstacles to the diversity and may decrease the representativeness of the federal civil service. The U.S. General Accounting Office (GAO 1977, 1), for instance, concluded, “Veterans’ preference … severely limits job opportunities for people who are not veterans. It particularly diminishes the employment chances of women.” Further, by creating red tape to protect veterans’ rights, preference limits managerial discretion in hiring and may push managers toward alternative hiring mechanisms. GAO (1995, 5–6) reported, “Agencies prefer using noncompetitive hiring mechanisms where they do not have to apply veterans’ preference points and the Rule of Three.” By crediting military service in hiring, veterans’ preference challenges merit principles and may lower the quality of the civil service.

Although veterans’ preference has attracted occasional scholarly attention and government reports, we still know strikingly little about (1) how much it impacts who gets federal jobs, (2) how it alters the composition of the federal workforce, and (3) how it affects the qualifications and quality of federal employees. After briefly recounting the history of veterans’ preference and reviewing the findings of a handful of federal and academic studies, I provide a more systematic analysis of the effect of veterans’ preference on the federal civil service. Census data for 1990, 2000, and 2006–2009 show that veterans are three to four times as likely to hold federal jobs as comparable nonveterans, but only a little more likely to hold state and local government jobs. This preference for veterans dramatically shrinks women’s representation in the federal civil service and decreases the representation of Asians, naturalized citizens, and gay men, though it has minimal impact on the composition of the state and local public workforce. Federal personnel records show that veterans’ preference leads to a less educated but more experienced federal civil service and one that probably does not perform as well as it would in the absence of veterans’ preference.

HOW MUCH DOES VETERANS’ PREFERENCE AFFECT WHO GETS HIRED?

The federal government has preferred disabled veterans in hiring at least since the Civil War. Congress expanded preference to cover non-disabled veterans and their widows after World War I. President Warren G. Harding implemented a point system for awarding preference to disabled and other veterans, and Congress codified the point system as World War II was winding down (US Civil Service Commission (CSC) 1955). Although at times only veterans who had served in combat qualified, preference was gradually widened to cover those who received honorable or general discharges after at least 180 days of service in wartime, including the Cold War and the Global War on Terror.

Traditionally, federal job applicants have been rated on a 100-point scale, typically based on a written examination or an “unassembled” assessment of the applicants’ qualifications conducted by the US Office of Personnel Management (OPM), or its predecessor (CSC), or the department’s personnel department. OPM or CSC then created a register of qualified applicants (those with scores of 70 or above), added 10 points to the scores of disabled veterans and 5 points to the scores of other veterans, and ranked applicants by their final scores. Veterans ranked ahead of nonveterans with the same final scores, so that disabled veterans with raw scores of 90 and other veterans with raw scores of 95 were placed ahead of nonveterans with perfect scores of 100. Veterans with compensable, service-connected disabilities of 10% or more and with scores of at least 70 “floated” to the very top of the register, ahead of everyone else. Under the traditional rule of three, federal hiring officials could only consider the top three candidates at the top of the civil service register.

Under the newer “category rating” system (which was first tried as a US Department of Agriculture “demonstration project” and expanded as an option for all agencies in 2002, before completely replacing the rule of three in 2010), agencies define two or more categories of qualified applicants (e.g., highly qualified, well qualified, and qualified). Hiring officials can consider anyone in the top category. Veterans, however, are placed ahead of all nonveterans in the same category. Under both systems, hiring officials cannot pass over veterans to hire nonveterans lower on the register or in the same category without a written explanation approved by OPM.

All 50 states and the District of Columbia award veterans’ preference. State programs vary on several dimensions. Massachusetts, New Jersey, Pennsylvania, and South Dakota provide absolute preference in hiring: veterans meeting minimum requirements must be hired ahead of nonveterans (Fleming and Shanor 1977; Virelli 2004). Forty states and the District of Columbia award points to veterans with passing scores, though the number of points awarded varies somewhat, and seven states are vague about the nature of the preference offered (Virelli 2004, 1086–87).

How much does veterans’ preference affect one’s chances of obtaining a government job? Some researchers consider it merely a “limited preferential hiring program” (Berger and Hirsch 1983, 459), but others find a major impact. The GAO (1977) indicated that 30% of federal and only 15% of nonfederal workers at that time were veterans. Blank (1985) found that veterans were much more likely to work for government, especially the federal government, than nonveterans of the same sex, race, and experience and educational levels. Sanders (2007, 412) found that veteran status raised the odds of a government job by about 40% for native-born citizens, and the odds nearly doubled for immigrants.

Sanders’s failure to examine federal and other public employment separately, however, probably understates the impact of veterans’ preference on the federal civil service. Preferential treatment should increase the representation of veterans to the extent that people prefer government jobs, and state government jobs tend to be less desirable. Public administration scholars generally argue that federal workers are underpaid (the President’s Pay Agent [2011] estimates that the current pay disparity is at least 30%). Labor economists, however, find that federal workers earn substantially more than comparable private sector workers, but that state and local government pay tends to be quite similar to private sector pay (Krueger 1988; Moulton 1990; Smith,1976). Although job security and opportunities to serve the public also attract people to government jobs even in the absence of pay advantages, the clear evidence of higher pay in the federal than in the state sector suggests that veterans’ preference will disproportionately attract veterans to the federal civil service.

WHO BENEFITS?

Weighting military service in hiring decisions is likely to benefit men, whites, heterosexuals, and native-born citizens over others. In pushing to weaken veterans’ preference when Congress was considering the Civil Service Reform Act of 1978, CSC Chairman Alan Campbell argued that “the present limitations favoring preferential hiring from a labor pool which is 98% male and 92% nonminority places an inordinate burden on Federal agencies trying to implement affirmative action” (quoted in Lewis and Emmert 1984, 329). Women could not make up more than 2% of the armed forces until 1967 (Elliott 1986), and GAO (1977) reported that large numbers of highly qualified women could not be considered for federal jobs because, even with perfect scores on civil service examinations, they ranked lower than less educated and experienced veterans with lower test scores. Women now make up 14% of the US military, but men still outnumber them by 6-to-1 (Clemmitt 2009). Several studies have found a negative impact of veterans’ preference on women in the federal civil service (Emmert and Lewis 1982; Keeton 1994; Mani 1999) and in state governments (see several studies in Hale and Kelly 1989).

The impact on minorities has attracted less attention, though the military clearly limited opportunities for African Americans and Asians in World War II. The military has become much more diverse, of course. In 2000, 66% of active duty military were white, 20% were black, 8% were Hispanic, and 4% were Asian, but whites and blacks were still over-represented, and Latinos and Asians were under-represented, relative to their shares of the military-age population (Lutz 2008, 177). “Don’t Ask, Don’t Tell” ended in 2011, but the military long prohibited the service of homosexuals (Berube 1990; Shilts 1993) and only partially loosened the restriction in 1993 (Rimmerman 1996). Partnered gay men, but not partnered lesbians, are less likely than comparable married people to hold federal jobs (Lewis and Pitts 2011), consistent with gay/lesbian patterns of under/over-representation in the military (Gates 2004). Sanders (2007, 407) reports that 6.1% of employed male immigrants were veterans, but rates of military service for native-born males are substantially higher. Naturalized citizens who immigrated as adults may be hurt in their prospects for federal employment, due to a smaller probability of qualifying for veterans’ preference (Lewis, Liu, and Edwards 2011).

Further, as federal hiring has been fairly limited since the 1980s, the composition of the military in the past—when it was more white, male, and heterosexual—matters more for the composition of the federal civil service. With federal hiring accelerating as Baby Boomers retire, the increasing diversity of the military will mean that a more representative group of Americans will benefit from veterans’ preference, but the composition of the federal civil service will evolve much more slowly.

MANAGERIAL DISCRETION AND THE QUALITY OF FEDERAL EMPLOYEES

Veterans’ preference has repeatedly raised concerns about merit principles and managerial hiring discretion. While Congress was considering expansion of veterans’ preference at the end of World War I, the CSC responded, “The civil-service law is based upon the principle that every citizen should have equal opportunity to appointment in the public service and that in each case the most efficient should be appointed” (quoted in Emmert and Lewis 1982, 48). A survey of state employees and personnel specialists found much greater dissatisfaction with veterans’ preference among the personnel specialists, presumably due to their greater commitment to the merit principle (Elliott 1986).

The US Merit Systems Protection Board (MSPB 1994, xii) argued that “veterans preference and the ‘rule of three’ [are] widely viewed as an impediment to good hiring practices. In fact, a good portion of the evolution in hiring methods that has occurred may have been molded by agency reaction to the combined effects of these two requirements rather than by a desire to use the best selection tools available.” GAO (1992) reported that agencies frequently sent back registers that were headed by veterans in hopes that other agencies would hire the veterans (allowing the original agency to go deeper on the registers to reach nonveterans they liked better) and were increasingly hiring through the Outstanding Scholar Program (OSP), which allowed them to hire college graduates in the top 10% of their class or with grade point averages above 3.5 and to ignore veterans’ preference. OSP proved popular with hiring officials: OSP hires rose from 1% to 9% of new hires in entry-level professional and administrative positions between 1984 and 1992 (MSPB 1994), and Tsugawa (2011) finds that 30% to 40% of entry-level professionals were hired through OSP in the mid- to late-1990s. In 2006, however, MSPB ruled that OSP violated the rights of veterans, and OPM strongly advised agencies to stop using OSP.

In 2000, President Clinton created the Federal Career Intern Program (FCIP) to allow agencies to hire into entry-level professional and administrative positions without public notice or competition and, following private sector practice, to recruit and hire on college campuses and at job fairs (MSPB 2008, 15). FCIP hiring exploded under President Bush: by 2007, about 70% of entry-level professional and administrative new hires entered under that hiring authority (Tsugawa 2011). Subsequently, however, a federal court and the MSPB found that filling positions through FCIP without posting them competitively and awarding veterans’ preference violated veterans’ rights (Losey 2010; Rosenberg 2009), and President Obama dismantled it (E.O. 13562, December 2010). He has recently implemented a new “Pathways for Students and Recent Graduates to Federal Careers” program that explicitly applies veterans’ preference.

Thus, veterans’ preference probably decreases the employment of college graduates in the federal civil service, and some hiring officials apparently believe the program forces them to hire less-qualified applicants. Several studies of career patterns in the federal civil service have tested the hypothesis that veterans’ preference leads to veterans earning more than comparably educated and experienced nonveterans (Grandjean 1981; Johnson and Libecap 1989; Lewis 1988; Lewis and Emmert 1984; Taylor 1979). Surprisingly, veterans actually earn less, leading to speculation that “veterans’ preference, applied at entry to the federal civil service, places individuals of lower ability into jobs for which they would otherwise not qualify, while subsequent promotions are affected more by ability than by the veterans’ preference”(Taylor 1979, 473). If veterans’ preference does lead to lower quality applicants being hired, the quality of the civil service is likely to be unnecessarily poor.

In sum, veterans’ preference may have major impacts on the civil service. Veterans should be dramatically more likely than comparable nonveterans to obtain federal jobs but only somewhat more likely to hold lower-paying state and local government jobs. Veterans’ preference probably has the most negative impact on the employment of women, but it may also disadvantage Latinos, Asians, naturalized citizens, and people with same-sex partners, at least in the federal government. It probably decreases educational levels in the federal civil service and may lead to less-qualified job applicants being hired.

DATA AND METHODS

I work with two types of data: Census data on the full-time labor force and a 1% sample of personnel records for full-time, white-collar federal employees. The 5% Public Use Microdata Samples (PUMS) from the 1990 and the 2000 Census and the combined 2006–2009 American Community Surveys (ACS) provide information on military service, sector of employment, and a variety of demographic and work characteristics for over 3 million full-time (36+ hours per week), full-year (50+ weeks) employees, for more than 9 million cases. (I drop the self-employed and part-time workers.)

Trying to determine whether people qualified for veterans’ preference proved problematic. Respondents check all the periods in which they served in the US military, from a list of 8 to 11 periods that closely resemble the periods OPM uses to determine preference eligibility,1 but OPM (2011) reports that 90% of federally employed veterans receive preference, and only 70% of federally employed veterans in the ACS sample reported military service during the periods that qualify for preference. Thus, two-thirds of the veterans I initially coded as not qualifying for preference probably did qualify, and they were nearly as likely to hold federal jobs as veterans who clearly qualified. I chose to use a single dummy variable to identify those who had served in the military. This will still include some who do not qualify for preference and will miss some who qualify for preference due to the service of their deceased or disabled spouse.

The Census asks whether each employed person works for a private for-profit company, a private not-for-profit or charitable organization, a local government, a state government, or the federal government. I simplified these into three sectors: federal, state and local, and private (including both for-profit firms and nonprofit organizations). To test the effects of military service on the probability of public sector employment, I begin with simple comparisons of the percentages of veterans and nonveterans who held jobs in each sector in each period (1990, 2000, and 2006–2009).

I then run multinomial logit models to test whether veteran–nonveteran differences in education, age, race/ethnicity gender, relationship status (married, living with an unmarried partner of the same or opposite sex, or unpartnered), citizenship status, and English language proficiency can explain differences in sector of employment. Because the huge sample sizes require no simplifying assumptions about the effects of age and education on the probability of federal employment, I restrict the sample to employees between the ages of 21 and 65 and use 45 dummy variables for year of birth and eight dummy variables for level of education. I also use sets of dummy variables for race/ethnicity and gender, for relationship status (e.g., man with male partner), and for citizenship status and English proficiency. Most of these are shown in table 1.

Table 1

Veteran–Nonveteran Differences in Percentage with Federal Jobs, 2006–2009

	Veterans	Nonveterans	Ratio
Race/ethnicity/gender
White women	14.8	2.7	5.5
Black women	23.1	5.4	4.3
Hispanic women	16.0	2.3	7.1
Asian women	20.8	3.7	5.7
Native American women	17.2	10.5	1.6
Other/mixed race women	19.0	4.2	4.5
White men	8.8	2.4	3.8
Black men	14.9	3.1	4.8
Hispanic men	13.5	1.3	10.6
Asian men	20.5	3.3	6.1
Native American men	13.8	5.6	2.5
Other/mixed race men	12.4	2.9	4.3
Married women	18.0	3.1	5.8
Women with male partner	11.9	2.2	5.5
Women with female partner	14.7	2.8	5.3
Single women	16.7	3.2	5.2
Married men	10.6	2.5	4.2
Men with female partner	6.5	1.5	4.4
Men with male partner	8.4	2.6	3.3
Single men	9.5	2.1	4.6
Educational attainment
No high school	3.7	0.4	8.7
Some high school	3.7	0.7	5.3
High school graduate	7.9	2.2	3.7
Some college	11.7	2.9	4.1
College graduate	13.4	3.3	4.1
Master’s degree	15.0	3.8	4.0
Professional degree	12.8	4.6	2.8
Doctoral degree	12.4	5.5	2.2
Year of birth
1940–1944	7.4	3.4	2.2
1945–1949	8.5	3.6	2.3
1950–1954	12.2	3.9	3.1
1955–1959	14.1	3.7	3.8
1960–1964	12.1	3.2	3.7
1965–1969	10.3	2.6	4.0
1970–1974	9.2	2.1	4.3
1975–1979	9.0	1.8	4.9
1980–1984	7.9	1.6	4.9
1985–1989	4.4	1.3	3.5

With the private sector as the base category, coefficients show how one-unit increases in the independent variables affect the log-odds of federal (or state–local) employment relative to private sector employment. To make this more understandable, I calculate everyone’s predicted probabilities of working in each sector twice, once as a veteran and once as a nonveteran (coding the veteran variable first as a 1, then as a 0), and calculate the mean probabilities of employment in each sector for veterans and nonveterans. The difference between these mean probabilities is the average partial effect (Wooldridge 2009, 582). (Probabilities are nonlinear functions of the independent variables in logit models; a variety of methods can translate logit coefficients into probability differences. This method compares veterans and nonveterans across the full distribution of the variables in the sample.)

Using a single veteran variable implicitly assumes that being a veteran has the same impact on sector of employment for everyone. To allow the effects of veteran status to vary with race, sex, age, educational level, age, and sexual orientation, I repeat the multinomial logit models, adding interaction terms between veteran status and the other independent variables. I again estimate each person’s probability of working in each sector twice, once as a veteran and once as a nonveteran, and calculate the mean probabilities.

To test whether veterans’ preference disproportionately benefits men, whites, heterosexuals, and native-born citizens, I first compare the characteristics of veterans and nonveterans, beginning with Census data for the three periods. I then look at my second data source: a 1% sample of the Central Personnel Data File (CPDF), which OPM maintains as the federal government’s personnel records. I focus on all white-collar federal employees in 2009 and on employees hired into white-collar jobs from April 1999 through March 2009.2 I focus on race/ethnicity and sex, as my CPDF sample includes no information on marital status, sexual orientation, or citizenship status.

Next, to estimate what the composition of the federal civil service would have been in the absence of veterans’ preference, I rerun the multinomial logit models, restricting the sample to nonveterans. I then calculate the probabilities of employment in each sector for everyone (both veterans and nonveterans) as if the patterns that currently apply for nonveterans held for everyone. I then calculate mean probabilities by race/ethnicity and gender, by relationship status, and by citizenship status. I multiply those probabilities times the number of people of that characteristic in the sample to calculate the expected number and percentage of federal employees who would have each characteristic. This approach assumes that, in the absence of veterans’ preference, individual characteristics would affect sector of employment the same way they currently do for nonveterans. I expect the representation of women, Asians, gay people, and naturalized citizens would increase.

Testing the possibility that veterans’ preference leads to a less-qualified federal work force is tricky. I begin by examining the educational levels and ages (as a proxy for work experience) of veterans and nonveterans, on the expectation that the veterans are more experienced but less educated. Labor economists have demonstrated, however, that both education and experience increase the productivity of workers; so this tradeoff could increase or decrease the quality of the civil service. Those who have studied the impact of veterans’ preference have typically looked at pay differences between veterans and equally educated and experienced nonveterans of the same race and sex, expecting that preference leads veterans to earn more than comparable nonveterans. There are two problems with this approach for assessing the impact of veterans’ preference on the quality of the civil service. First, the consistent finding that veterans earn less than comparable nonveterans would imply either that the federal civil service discriminates against veterans, which does not seem plausible, or that veterans are less qualified than equally experienced and educated nonveterans. Second, we want to compare the productivity of veterans not to nonveterans with the same characteristics but to the nonveterans who would have been hired in the absence of veterans’ preference.

As an approximation to this, I restrict the sample to full-time employees in the General Schedule (GS) and equivalent pay schedules. GS positions are classified into 15 grades, based on the difficulty and responsibility of the work involved. I focus on employees who were hired into the four most common entry grades and track the mean grades of veterans and nonveterans for 15 years. If veteran and nonveteran new hires in the same grade are equally qualified, they should progress at approximately the same rate, on average.

FINDINGS

Impact on Probability of Government Employment

Veterans are far more likely than nonveterans to have federal jobs, but they are only a bit more likely than nonveterans to hold state and local government jobs (table 2). In the 2006–2009 ACS, for instance, 10.5% of veterans and only 2.9% of nonveterans worked for the federal government, and 14.6% of veterans and 14.2% of nonveterans worked for state and local governments. That is, veterans were 3.6 times as likely as nonveterans to be federal employees but only 3% more likely than nonveterans to work for state and local governments. The percentages vary somewhat across the years, but veterans are always 2.9 to 3.6 times as likely as nonveterans to hold federal jobs and never more than 10% more likely than nonveterans to hold state and local government jobs.

Table 2

Veteran–Nonveteran Differences in Sector of Employment

		1990				2000				2006–2009
		Veteran		Nonveteran		Veteran		Nonveteran		Veteran		Nonveteran
Percent in sector
Federal government		10.0		3.4		9.3		2.9		10.5		2.9
State/local governments		12.9		11.7		13.4		12.2		14.6		14.2
Private sector		77.1		84.9		77.3		84.8		82.9		75.0
Multinomial odds ratios
Federal government		3.56*** (0.007)				3.52*** (0.009)				4.18*** (0.012)
State/local governments		1.09*** (0.005)				1.16*** (0.006)				1.22*** (0.009)
Predicted probabilities from logit
Federal government		10.7		3.4		9.3		2.9		9.7		2.8
State/local governments		11.7		11.9		12.4		11.9		13.1		13.8
Private sector		77.6		84.7		78.4		85.2		77.2		83.4
Predicted probabilities from logit with interaction terms
Federal government		15.1		3.5		11.4		3.0		11.6		2.8
State/local governments		11.7		12.0		12.7		12.3		14.2		13.6
Private sector		73.2		84.4		75.9		84.7		74.2		83.6

Note: ***Odds ratio is significant at the .0001 level.

Potentially, veteran–nonveteran differences in gender, race, age, education, and sexual orientation could account for those differences in sector of employment. The multinomial logit models, however, control for all those variables and still find that veterans are far more likely than comparable nonveterans to hold federal jobs. The second panel of table 2 shows that veterans’ odds of federal rather than private sector employment are 3.5 to 4.2 times as high as the odds for comparable nonveterans. Veterans’ odds of state and local employment, however, are only 9% to 22% higher than those of comparable nonveterans.

The third panel uses the multinomial logit findings to predict each person’s probabilities of employment in each sector twice, once as a veteran and once as a nonveteran, and then calculates the mean probabilities of employment in each sector for veterans and nonveterans, holding all other variables at their actual distribution in the sample. In 2006–2009, for instance, the average predicted probability of holding a federal job was 9.7% if employees were coded as veterans and only 2.8% if the same employees were coded as nonveterans. In 2006–2009, this controlled difference (9.7% versus 2.8%) is somewhat smaller than the uncontrolled difference (10.5% versus 2.9%), but the two differences are the same in 2000 and the controlled difference is somewhat larger in 1990. That is, differences in the individual characteristics of veterans and nonveterans can explain some of the differences in their probabilities of federal employment in 2006–2009, but not in 1990 or 2000.

Veterans’ higher probabilities of federal employment hold across all sub-groups. Figure 1 shows that the percentage of veterans who held federal jobs in 2006–2009 was substantially higher than the percentage of nonveterans of the same sex who did so at each year of birth.3Table 2 shows the percentages of veterans and nonveterans who hold federal jobs within several sub-groups, plus the ratio of those percentages. In general, veteran status makes more difference for women and minorities than for white men. Among whites, for instance, 14.8% of female and only 8.8% of male veterans held federal jobs, with veterans being 5.5 times as likely as nonveterans to have federal jobs among women and only 3.8 times as likely to do so among men. Veterans’ preference increases the probability of federal employment more for naturalized than for native-born citizens, for those born since 1970 than for those born earlier, and for less-educated Americans. Althogh Native Americans and partnered gay men are more likely to hold federal jobs if they are veterans than if they are not, however, the boost appears to be smaller than for white and married men.

Figure 1

Probability of Federal Employment, 2006–2009

Given that veterans’ preference affects sector of employment differently for different groups, the fourth panel of table 1 allows its impact to vary with individual characteristics, by working from multinomial logit models that include interaction terms between veteran and all the other independent variables. Holding the other variables constant, veteran status increases the log-odds of federal employment more for women and minority men than for white men, more for married people than for those in other relationship categories, more for naturalized than for native-born citizens, and more for younger than older Americans (results not shown). Allowing for this differential effect of veteran status on federal employment, the unexplained veteran–nonveteran difference in probability of federal employment is actually wider in each period than the difference without controlling these other variables. In sum, veterans are much more likely than comparable nonveterans to be federal employees. The simple differences of percentages are not overstating the difference.

I also ran the multinomial logit models separately for each state in 2006–2009. Veterans were significantly more likely than comparable nonveterans to hold federal jobs in every state, but they were significantly more likely to hold state and local government jobs in only 10 states (California, Connecticut, Florida, Illinois, Massachusetts, New Jersey, New York, Pennsylvania, Vermont, and Washington). Preference might be stronger in these states (Massachusetts, New Jersey, and Pennsylvania have absolute veterans’ preference [Virelli 2004]), or the statistical significance may be a function of sample size (all except Vermont are in the 15 most populous states), but the veteran coefficients are split almost equally between positive and negative across the 50 states, and veterans are significantly less likely to hold state and local government jobs in Georgia, Oklahoma, South Carolina, and Virginia. The over-representation of veterans in some state and local governments may indicate that their public sector jobs are particularly desirable.

Veterans’ Preference and the Composition of the Federal Civil Service

The Census data confirm that some groups—especially men—are more likely than others to be veterans (table 3). In 2006–2009, for instance, 91% of veteran and only 52% of nonveteran full-time employees are men. In addition, higher percentages of veterans than nonveterans are white (81% versus 72%) and black (10% versus 9%), but smaller percentages are Latino (6% versus 12%) and Asian (1% versus 5%). Men with male partners are only half as likely as married men to have served in the military (7.8% versus 15.6%), whereas women with female partners are four times as likely as married women to be veterans (5.4% versus 1.4%). (The numbers with same-sex partners are too small to show well in table 1.) Naturalized citizens are only one-third as likely as native-born citizens to have military service (3.1% versus 9.6%). The veteran population has become more diverse over time. The male percentage dropped from 97% in 1990 to 91% in 2006–2009, and the white male percentage dropped from 84% to 76%.

Table 3

Characteristics of Veterans and Nonveterans

		Census Data												Central Personnel Data File
		1990				2000				2006–2009				2009				New Hires, 2000–2009
		Vets		Nonvets		Vets		Nonvets		Vets		Nonvets		Vets		Nonvets		Vets		Nonvets
White female		2.4		41.0		4.0		35.5		5.3		34.2		11.6		34.2		12.8		33.0
Black female		0.4		5.5		1.1		5.4		1.5		5.2		7.5		15.5		8.1		11.0
Hispanic female		0.2		4.7		0.3		3.9		0.4		4.9		1.6		4.3		2.0		4.3
Asian female		*		1.4		0.1		1.7		0.1		2.4		0.6		3.0		0.8		3.9
White male		83.6		36.5		78.4		39.5		75.5		37.9		53.7		30.2		53.1		32.4
Black male		6.7		3.7		8.5		3.8		8.6		3.5		14.6		4.2		13.9		4.5
Hispanic male		5.1		5.0		4.4		6.0		5.1		7.2		6.5		3.6		5.9		4.5
Asian male		0.7		1.4		0.9		2.0		1.3		2.8		2.8		2.3		2.6		3.5
No high school		3.3		4.9		1.0		3.2		0.4		3.5		0.4		0.1		0.1		0.0
Some high school		9.4		11.5		5.0		8.3		1.9		5.1		0.3		0.6		0.1		0.4
High school graduate		58.3		53.2		41.8		36.5		37.5		33.0		39.0		33.1		45.7		32.4
Some college		8.0		7.8		29.8		23.6		33.2		24.3		23.6		16.8		21.8		13.7
College graduate		13.6		15.4		14.4		18.9		17.2		22.3		24.2		31.8		21.6		33.1
Master’s degree		5.0		4.9		5.7		6.4		7.4		8.4		11.2		12.7		9.3		14.2
Professional degree		1.4		1.5		1.5		1.8		1.5		2.0		0.7		2.7		0.5		3.1
PhD		0.9		0.9		0.8		1.2		0.8		1.4		1.0		3.3		0.8		3.2
Education (mean years)		12.9		12.8		13.6		13.5		13.9		13.9		14.3		14.9		14.0		15.0
Age (mean)		47.2		37.7		46.9		39.8		49.6		42.2		48.5		45.2		40.2		34.6
Married man		73.9		29.8		69.7		34.1		64.1		31.4		—		—		—		—
Man with female partner		2.6		1.8		4.0		2.8		4.5		3.4		—		—		—		—
Man with male partner		0.1		0.1		0.2		0.2		0.2		0.3		—		—		—		—
Single man		20.3		15.2		20.4		15.4		23.0		18.2		—		—		—		—
Married woman		1.6		30.9		2.7		26.8		3.8		23.7		—		—		—		—
Woman with male partner		0.1		2.2		0.3		2.7		0.4		3.0		—		—		—		—
Woman with female partner		*		0.1		0.1		0.2		0.2		0.3		—		—		—		—
Single woman		1.4		19.9		2.6		17.9		3.8		19.8		—		—		—		—
Native-born citizen		98.0		90.3		97.1		88.7		96.8		82.6		—		—		—		—
Naturalized citizen		1.6		4.0		2.2		5.6		2.8		7.8		—		—		—		—
Non-citizen		0.4		5.5		0.7		6.7		0.5		9.6		—		—		—		—

Note: *Less than 0.05%.

The final two columns of table 3 use the federal personnel data for white-collar federal employees. Veterans who are civil servants are more diverse than veterans generally. Over 21% are women, more than twice the percentage of veterans overall who are female. Still, white men make up nearly twice as big a share of veterans as of nonveterans in the federal civil service (54% versus 30%), and women of each race/ethnicity compose more than twice as high a percentage of nonveterans as of veterans. Patterns for new hires over the past decade are similar to those for all federal white-collar employees in 2009, but more of newly hired veterans than of all federally employed veterans are women.

Table 4 reports the percentages of federal employees who belong to various sub-groups and the percentages of employees who would be predicted to belong to those sub-groups if the probabilities of holding federal jobs were predicted by the patterns for nonveterans. In 2006–2009, for instance, white women make up 24.9% of federal employees but are predicted to compose 28.8% in the absence of veterans’ preference. The model predicts that, if veterans’ preference had never existed, the federal civil service would have been almost perfectly split between men and women in all three periods; instead, the female percentage of federal employees rose from 39.3% in 1990 to 43.2% in 2006–2009. The racial composition, however, would have been nearly unchanged, because gains for white and black women would have been offset by losses for white and black men. (Asian and other/mixed race men would gain from the absence of veterans’ preference, as would Hispanic men in the most recent period.) In the absence of veterans’ preference, representation would also rise for naturalized and non-citizens and for partnered gay men (but not partnered lesbians), though the latter effects are too small to show up in table 4.

Table 4

Current and Predicted Composition of Federal Workforce

		1990				2000				2006–2009
		Current		Predicted		Current		Predicted		Current		Predicted
Female		39.3		50.0		41.1		49.3		43.2		49.9
Male		60.7		50.0		58.9		50.7		56.8		50.1
White		72.6		71.8		69.0		68.5		63.7		63.1
Black		14.8		15.3		16.3		16.3		18.4		18.1
Hispanic		7.6		7.5		6.8		6.6		8.1		8.9
Asian		3.2		3.4		4.3		4.5		5.8		6.5
Other		1.8		2.0		3.7		4.0		3.2		3.3
White female		26.5		33.5		26.1		31.2		24.9		28.8
Black female		7.7		9.9		8.9		10.6		10.4		11.9
Hispanic female		3.0		3.8		2.7		3.3		3.7		4.4
Asian female		1.3		1.6		1.6		2.0		2.5		3.0
Other female		0.9		1.1		1.8		2.3		1.6		1.9
White male		46.1		38.4		43.0		37.3		38.8		34.4
Black male		7.2		5.4		7.4		5.7		8.0		6.3
Hispanic male		4.6		3.7		4.1		3.4		5.2		4.6
Asian male		1.9		1.8		2.5		2.4		3.4		3.5
Other male		0.9		0.8		1.9		1.8		1.5		1.3
Native-born citizen		94.5		93.5		92.7		92.0		90.3		89.4
Naturalized citizen		3.7		4.2		5.4		5.6		7.3		7.7
Non-citizen		1.8		2.3		2.0		2.4		2.4		2.9

Because veterans’ preference has much less impact on whether one works for state and local governments than for the federal civil service, it has little impact on the composition of the state and local workforce. I ran the analysis separately for the 10 states, where being a veteran had a significant positive impact on getting a state or local government job, and for the other 40 states. Although men composed a higher percentage of the state and local workforce in the 10 states, statistically removing veterans’ preference from the system changed the gender composition in each group of states by less than half a percentage point.

Veterans’ Preference and the Quality of the Federal Civil Service

On average, veterans and nonveterans are equally educated in the population, but mostly because veterans’ educational levels vary less: they are more likely than nonveterans to have graduated from high school but less likely to have graduated from college. In 2006–2009, for instance, the mean years of education for both veterans and nonveterans was 13.9, but 8.6% of nonveterans and only 2.3% of veterans had not finished high school (table 3). In contrast, 34.1% of nonveterans and only 26.9% of veterans held bachelor’s or graduate degrees.

Federal white-collar employees are more educated than the general workforce—over 99% are high school graduates and nearly half are college graduates. CPDF data show that veterans are older and less educated than nonveterans, with the patterns stronger for new hires than for the civil service as a whole. Among those hired into white-collar jobs since 1999, over half of the nonveterans and only one-third of the veterans had completed college, and nonveterans were twice as likely to hold graduate degrees (20% versus 10%). On average, the newly hired nonveterans had one more year of education than the newly hired veterans. However, the veteran new hires were nearly 6 years older than the nonveterans.

Thus, veterans’ preference appears to increase experience levels at the expense of educational levels in the federal civil service. As both education and experience increase productivity, this tradeoff could increase or decrease the quality of the civil service.4 Nonveteran new hires, however, are advancing more quickly than nonveterans hired into the same entry grades (figure 2). Within 2 years, the mean grades of nonveterans are noticeably higher than veterans starting at the four most common entry grades, and the pattern persists over the first 15 years of their federal civilian careers. Among employees who entered in GS-5, for instance, mean grades 5 years later were 7.9 for veterans and 8.4 for nonveterans; after 10 years, the means were 9.0 and 9.6, respectively. Most strikingly, veterans who entered in GS-7 had a mean grade of only 9.8 after 5 years, whereas nonveterans’ mean was 11.0; after 10 years, nonveterans remained 1.2 grades higher (11.7 versus 10.5).5 Perhaps because veterans begin their federal civilian careers later in life, their careers do not progress as rapidly as nonveterans entering at the same levels. Ten years later, nonveterans are 0.6 to 1.3 grades higher, on average, than veterans who entered at the same levels. If advancement accurately reflects performance, nonveteran new hires tend to have more potential than veterans.

Figure 2

Career Progression by Entry Grade

CONCLUSION

Veterans’ preference has a powerful impact on who gets hired in the federal civil service. Americans who have served in the military are three to four times as likely to obtain federal jobs as Americans who have not. The strikingly higher probabilities of federal employment hold for men and women, whites and minorities, heterosexuals and gay/lesbians of every age and education level. The basic pattern holds after controlling for all these variables simultaneously.

As one-quarter of federal employees receive veterans’ preference, at least 16% of federal employees would be different people if veterans’ preference had never existed. Because men still outnumber women in the US military by 6-to-1, preferential treatment of veterans in hiring has very disproportionately benefited men. If—in the absence of veterans’ preference—gender, race/ethnicity, age, education, and sexual orientation affected probabilities of federal employment for everyone the way they currently affect nonveterans, federal jobs might have been split almost equally between men and women since 1990. The numbers of Hispanics, Asians, and gay men might also be 20% higher.

Questions about how the quality of the federal civil service would differ are more difficult to answer. Veterans’ preference clearly leads to older, less-educated people being hired. Veteran new hires appear to have lower potential, on average, than nonveterans hired into the same entry grades. The nonveterans move ahead of the veterans within the first 2 years and remain in somewhat higher grades over the first 15 years of the career.

Although veterans’ preference programs in the states are similar to (and sometimes stronger than) the federal program, veterans are only slightly more likely than comparable nonveterans to work for state and local governments and only in some states. This provides additional evidence that federal jobs are desirable. Even after decades of bureaucrat-bashing and numerous reports that federal workers are underpaid, Americans still want federal jobs, and groups that get preferential treatment take advantage of it. Worries that the federal civil service cannot attract highly qualified applicants seem overstated.

Is veterans’ preference worth its obvious costs to the diversity and the less-clear costs to the quality of the federal civil service? The answer is a political one, weighing the relative values of veterans’ sacrifices against other goals of the federal personnel system. One decade into the wars in Iraq and Afghanistan, with the Obama administration apparently firmly committed to increasing employment of veterans and with partisan polarization making any policy change that could be considered as disrespecting our troops extremely dangerous politically, we are unlikely to see any changes in veterans’ preference in the near future. We should be clear, however, that preference is a powerful tool that does not come without costs.

1

The ACS uses the following periods: November 1941 or earlier, World War II (December 1941 to December 1946), January 1947 to June 1950, Korean War (July 1950 to January 1955), February 1955 to February 1961, Vietnam era (August 1964 to April 1975), May 1975 to August 1980, September 1980 to July 1990, August 1990 to August 2001 (including Persian Gulf War), and September 2001 or later. To qualify for veterans’ preference in the federal civil service, one must have 180 or more consecutive days of active duty service since September 11, 2001; or between August 2, 1990, and January 2, 1992; or between January 31, 1955, and October 15, 1976; or between April 28, 1952, and July 1, 1955; or in World War II or the Cold War (between December 7, 1941, and April 28, 1952). http://www.fedshirevets.gov/job/vetpref/index.aspx.

2

The CPDF does not explicitly identify new hires, so I use employees who first appear in the CPDF in 2000 or later. (OPM draws the sample in April of each year, based on the final three digits of the social security number, so that employees in the sample one year appear in every year in which they are federal employees in March.) This method introduces some error, as one-quarter of these “new hires” (71% of those with veterans’ preference and 15% of those without) already have more than 1.2 years of federal civil service on their first personnel record. The most likely explanation is military service (my CPDF sample has no direct measure of it, but it is included in federal civil service for retirement and other purposes), but these employees may also have returned to federal civil service after being out since 1973, transferred in from a non-reporting federal agency (e.g., the USPS or the CIA), or had their hire paperwork delayed in reaching the CPDF.

3

Multinomial logit analyses done separately for each sex for each year of birth and for each sub-group (not shown) confirm that the patterns hold even after controlling for the other variables. The veteran coefficient is always highly significant and generally larger for groups for whom the ratios are larger.

4

Using the 2006–2009 ACS on a sample restricted to white males working in the private, for-profit sector (the group among whom pay differences should be most likely to reflect productivity), an additional year of education raises expected salary by 12%. During the early career, an additional 3 years of experience would have the same impact (though its impact dwindles quickly and nearly disappears after about 15 years). If these pay differences reflect real productivity differences, getting an additional 6 years of experience in return for giving up 1 year of education among new hires could potentially increase the quality of the federal civil service.

5

Following the previous research, I also ran regression models for veteran–nonveteran grade differences (controlling for education, age, experience, race, and sex) for each year and for each level of federal civilian service. Veterans had lower expected grades than nonveterans in every model, typically by about one-half grade. I also ran logit models, controlling for the same variables plus GS level, for whether comparable veterans and nonveterans were equally likely to be promoted and to receive outstanding performance ratings. They were.