Printer Friendly

Educational benefits and military service: an analysis of enlistment, reenlistment, and veterans' benefit usage 1991-2005.

I. INTRODUCTION

Educational benefits for military veterans have become a prime recruiting tool in the All-Volunteer military force (AVF). The primary education benefit for most of the AVF era has been the Montgomery GI Bill (MGIB). This version of the GI Bill began in July 1985 and bears the name of its sponsor, Alabama representative Sonny Montgomery. Under this program, eligible veterans who enroll in an education program approved by the Veterans Administration are eligible to receive up to 36 mo of benefits if they begin usage within 10 yr of separating from service. Veterans are only eligible for MGIB benefits if they contribute $100 per month during their first year of service. (1) From time to time, three of the four military branches have supplemented the basic MGIB benefit with College Fund (CF) benefits. Research has indicated that educational benefits do in fact attract enlistments of "high-quality" youth, defined by the military services to be youth who possess a high school degree and who score above the youth population average of 50 on the Armed Forces Qualification Test (AFQT). (2)

Although educational benefits attract more highly qualified youth into the military, those benefits may be costly for at least two reasons. First, such youth may be more likely to separate from the military to use their benefits, thus requiring higher rates of recruitment of new enlistees to maintain the size and quality of the force. Second, higher benefits may increase the tendency of youth who separate from the military to use those benefits as well as increase the duration of usage. Of course, this would not be surprising; a growing body of research finds that college tuition subsidies raise college attendance. Interest in the potential behavioral effects of military educational benefits was recently heightened by a vigorous debate in Congress over veterans' benefits. This debate culminated in the Post-9/11 Veterans Education Assistance Act of 2008, which was passed in July 2008, becoming effective on August 1, 2009. The act provides future veterans with substantially higher educational benefits than those available through the MGIB.

This paper analyzes how GI Bill benefits affect military retention and veterans' benefit usage. The analysis is useful for two reasons. First, as our review of the literature below indicates, existing studies of the retention and usage effects of military education benefits are now quite dated. Second, prior studies have all ignored issues relating to selection; as a result the estimates of retention and usage effects found in them cannot be cleanly interpreted as pure incentive effects arising from benefit changes. Our analysis accounts for the possibility that higher veterans' education benefits may attract youth who are more likely to separate and use those benefits, conditional on their observables. Over the period of our analysis, the nominal value of GI Bill benefits were set legislatively and changed unpredictably. We use the unpredictable evolution of the value of veterans' education benefits over our data period to disentangle selection effects and incentive effects of benefit changes on retention and benefit usage.

Our analysis makes use of administrative data that span the period 1988-2005 and contains annual information on every individual who entered active military service from the time of entry to the time of separation or, if still in service, until the end of fiscal year (FY) 2004. As shown below, this was a period during which the real value of military educational benefits fluctuated considerably. For those who separated, details of educational benefit usage were available to us as of June 2005. If educational benefits do affect the retention behavior of military personnel or the educational decisions of military veterans, their effects should be apparent in data over this period.

To highlight our key findings, we estimate that a $10,000 increase in MGIB benefits would increase the fraction of Army veterans who use them within 2 yr of separation by 5-8 percentage points. The estimated effects for the other services are also, as a rule, positive and statistically significant, although smaller than for the Army. We find that higher education benefits increase separation from the Army and the Air Force. The association between separation and educational benefits is weaker for the Navy and Marine Corps. Enlistees with higher ability as measured by AFQT scores are more likely to separate after their initial enlistment and more likely to use their educational benefits. Finally, unobservable factors that affect separation are found to be positively correlated with educational benefit usage, indicating that components of ability other than that measured by AFQT may be playing a role in enlistees' decision making.

This paper is organized as follows. Section II reviews previous studies. Section III provides a qualitative description of our model and its predictions. Section IV outlines the current military educational benefit programs and documents the evolution of educational benefits and recruit participation in the military educational benefit programs over the decade-long period between 1993 and 2003. Section V details the data used for this study and documents trends in veterans' educational benefit usage. Section VI discusses identification issues. Section VII presents simultaneous estimates of separation from the military and education benefit usage. Section VIII presents hazard estimates. Section IX presents an analysis of the duration of usage. Section X concludes the paper with a discussion of the policy implication of our findings and outlines the path of future work.

II. PREVIOUS STUDIES

Existing studies suggest that education subsidies have significant, though modest effects on enrollment. It is easiest for the purposes of comparison to convert the various estimated effects to elasticities. Hilmer (1998) examined the responsiveness of individuals in the High School and Beyond survey to interstate variation in tuition and fees and found that a 1% increase in university fees would lead to a roughly 0.9% decline in the probability of enrollment in higher education. An early study (McPherson and Schapiro 1991) used time-series data to examine the effects of college costs on the attendance of low-income students. Their estimates implied that a 1% rise in tuition would reduce enrollment by about 0.67%. Dynarski's (2002) study of the Georgia Hope Scholarship program implied an elasticity of 0.25, and Dynarski's (2003) study of the effects of discontinuing the Social Security Student Benefit implied a slightly larger elasticity of 0.45. Seftor and Turner's (2002) study of the impact of Pell grants on college enrollment decisions of "nontraditional" students found elasticities of 0.12 and 0.29 for males and females, respectively. Finally, Nielsen, Sorensen, and Taber (2008), using a regression discontinuity design on Danish data, estimated an elasticity of about 0.5. (3)

We are aware of only two previous studies of educational benefit usage of AVF-era veterans. (4) Angrist (1993) examined the effects of veterans benefits on education and earnings using a small sample of early AVF-era veterans who were part of the 1987 Survey of Veterans. Table 5 of Angrist (1993) indicates that Vietnam veterans, covered by a more generous educational benefit program than early AVF veterans who entered military service after January 1977 and were covered by the Veterans Educational Assistance Program (VEAP), were between 22 and 33 percentage points more likely to have used their educational benefits by the time of the survey. Hogan, Smith, and Sylwester ( 1991) also studied educational benefit usage among Army veterans who entered in FY 1982 and separated by 1987. They estimated that a 1% increase in educational benefits would increase usage by 1.5%, for an elasticity of 1.5, much higher than typically estimated in studies of civilian populations and much higher than the estimates we obtain below.

Studies of the retention effects of military educational benefits are limited to three studies sponsored by the U.S. Army Research Institute (Hogan, Smith, and Sylwester 1991; Smith, Sylwester, and Villa 1991; and Warner and Solon 1991). These studies used various samples of Army entrants over the period FY 1974-1983--a period that spanned the end of the Vietnam-era GI Bill, the VEAP period (1977-1981), and a period of experimentation with the Army College Fund (ACF; 1982-1983)--to estimate the retention effects of educational benefits. Collectively, they found evidence of negative retention effects of educational benefits. Although these studies did not directly report elasticities, they may be computed from data reported in them. Calculations from the Hogan, Smith, and Sylwester study, for example, indicate retention education benefit elasticities in the range of -0.1 to -0.15. Using a base reenlistment rate of 40% (the mean rate in our data for the Army), these estimates indicate that a doubling of education benefits would reduce Army retention by about 4-6 percentage points. Calculations from the Warner-Solon study indicate an elasticity of about -0.2 for 3-yr enlistees; however, they found no significant effects of education benefits on retention of 4-yr enlistees.

III. THEORETICAL DISCUSSION

Changes in education benefits affect not only the payoff to separating from the military to attend college, but also the returns to enlisting in the first place, and hence can affect the composition of military recruits. To clarify our thinking, we constructed a dynamic programming model. The model, while simple conceptually, contains a large number of variables. We therefore merely outline the model here; the full model will be furnished to interested readers on request. (5)

Each individual i chooses at each time t between three states: (1) military service, (2) working in the civilian sector; and (3) going to school in the civilian sector. Denoting these choices with a superscript m, n, and c, each individual maximizes the expected value of lifetime wealth:

[E.sub.t]([V.sub.i,t+1]) = E[max(E([V.sup.m.sub.i,t+1]) + [[epsilon].sup.m.sub.i,t+1], (E([V.sup.n.sub.i,t+1]) + [[epsilon].sup.n.sub.i,t], (E([V.sup.c.sub.i,t+1]) + [[epsilon].sup.c.sub.i,t+1]

Wealth in sector j is equal to the sum of a deterministic component [V.sup.t.sub.it] and a random shock [[member of].sup.j.sub.it]. The [V.sup.j.sub.it] are (partly) functions of the stream of pecuniary compensation in j, which in turn are affected by an individual's level of ability. In the military sector, more able individuals are more likely to be promoted, and in the civilian sector, more able individuals are more highly paid, with ability having a larger impact on the civilian earnings of college graduates than of nongraduates. The value of military service is also a function of a random individual-specific nonpecuniary taste factor. Finally, education is equally cosily for both civilians and military veterans, but veterans are eligible to receive college benefit payments.

The key results of the model are as follows. Because individual-level ability affects college payoffs more than noncollege civilian or military payoffs, more able individuals are more likely to go to college upon graduating from high school and less likely to join the military. Among those who do join, higher ability individuals are more likely to separate to attend college.

An increase in the expected future college benefit has four effects.

1. It increases a given youth's likelihood of enlistment. We call this the enlistment incentive effect of the educational benefit.

2. It increases the average ability of enlistees because the sensitivity of enlistment to expected future college benefits is greater for high-ability youth. We call the resulting alteration in the ability mix and consequent higher probability of attending college later on the enlistment selection effect.

3. It reduces the probability of reenlistment, with stronger effects for high-ability individuals, who have a higher propensity to attend college.

4. Finally, it increases the likelihood that any individual separated from the military uses those benefits to attend college. This is what we call the pure incentive effect.

In principle, it is possible to estimate the dynamic programming model to analyze selection into the military as a function of the level of college benefits (and other variables in the model). However, doing so would require using data on a broader sample of individuals than is used in the current analysis because our data here consist of only the self-selected sample of military enlistees. We therefore take an alternative approach to dealing with the potential endogeneity bias between college benefits and selection on unobservables (ability) into the military. Our identification strategy exploits the distinct way in which education benefits have evolved over time. As will be seen in the next section, over the period of our data military college benefits did not change smoothly or perfectly predictably over time. As shown in Section VI, the fact that at least some portion of these changes was unpredictable at the time of enlistment provides the identifying variation necessary to estimate the pure incentive effect of education benefits.

IV. OVERVIEW OF MILITARY EDUCATION BENEFIT PROGRAMS

A. The Basic Benefit

An important feature of the MGIB program was that the value of the basic benefit was set in nominal terms by the U.S. Congress, participating veterans receiving the same benefit regardless of when they entered military service or began benefit usage. Unlike many federal programs in which benefits are adjusted annually for inflation, nominal MGIB benefits were adjusted haphazardly, the consequence of which was substantial and unpredictable fluctuations in the real value of MGIB benefits.

When the MGIB program was first implemented in 1985, the monthly benefit for a veteran who had three or more years of service was $300 per month; such an individual qualified for a maximum total benefit of $10,800 for 36 mo of usage. Individuals serving a 2-yr enlistment qualified for a monthly benefit of $272 per month and were eligible for a maximum entitlement of $9,800. Importantly, between 1985 and 1992, Congress did not adjust these benefit amounts despite the fact that college tuition costs as measured by the consumer price index (CPI) for tuition and fees increased at an annual rate of 7%. As a result, the real value of MGIB benefits, which did not change in nominal terms between 1985 and 1992, declined at about a 7% annual rate in real terms over that period.

Congress reversed the decline in MGIB purchasing power in 1992, raising nominal benefits by nearly 20%. Over the next 6 yr, benefits increased periodically at about the rate of overall CPI inflation, but college tuition inflation was nearly double that of the overall CPI. By 1998, the MGIB was worth less in real terms than it was in 1991, and, indeed, less than at its inception in 1985. Concerned about these trends, in late 1996, Congress appointed a commission headed by Anthony J. Principi to study the GI Bill. The commission's main recommendations were twofold: (1) the GI Bill should cover the full tuition and fees at the veteran's educational institution of choice and (2) the requirement that the recruit contributes $1,200 to the MGIB fund during the first year of service should be abolished (Principi 1999). Although neither recommendation was adopted, Congress did increase MGIB benefits substantially in January 1999 and again in January 2000. Further increases were programmed in July of 2000 to go into effect in years 2001 through 2003. By 2003, MGIB benefits were higher in real value than at any time since 1991. (6)

Figure 1 shows nominal and real MGIB amounts between 1991 and 2004 for individuals who enlisted for a term of 2 yr (called 2-yr obligors, or 2-YOs) and individuals who enlisted for terms of 3 or more years (3+ YOs). The real amounts were constructed using the CPI for college tuition and fees and are in 2005 dollars.

B. Service CFs

The Army, Navy, and to a lesser extent the Marine Corps offer supplemental CFs, known as "kickers," to channel recruits into harder-to-fill military occupations. During our data period, kicker amounts for new entrants were equal to the difference between a predetermined total nominal dollar value of educational benefit and the MGIB amount then in effect. For example, if a CF recipient was promised an educational benefit of $40,000 at a time when the basic MGIB amount was $15,000, the kicker amount was fixed nominally at $25,000. If the basic MGIB amount rose to $20,000, the kicker received by new recruits awarded CF would be reduced by $5,000. The real kicker amounts therefore exhibit a sawtooth pattern, falling sharply when MGIB amounts rose and rising when total CF amounts were increased.

[FIGURE 1 OMITTED]

The percentage of Army recruits receiving CF kickers ranged from a high of 29% in 1991 to a low of 7% in 1999. CF enrollment in the Navy increased from less than 10% before 1994 to about 30% in the late 1990s. Typically, fewer than 5% of Marine Corps recruits received CF kickers between 1988 and 2001.

CF kickers generate both anticipated and unanticipated variation in the real value of education benefit received by recruits in each military service. Although the higher college benefits that CF recipients expect at the time they enter the military can affect the composition of recruits, they also generate important unanticipated changes in the real value of those benefits as well. There is uncertainty about the real value of those kickers to the extent that there is uncertainty about the future course of college tuition inflation.

Figure 2 shows average real Army kicker amounts received by 2-YOs and 4-YOs over the period FY 1988-2001. Recruits who obligated for 4 yr of service received larger CF kicker amounts than recruits who obligated for just 2 yr. Both 2-YO and 4-YO CF recipients, in turn, receive higher education benefits than their non-CF counterparts. For 4-YOs who entered in FY 1992, the average value of the kicker received by 4-YOs who enlisted in 1992 was nearly $45,000, much higher than the average amount of $26,000 received by 4-YOs who enlisted in 1999. Despite increases in 2000 and 2001, by 2004, the real value of the Army CF kickers was at its lowest point since the start of our data.

[FIGURE 2 OMITTED]

Because of college tuition inflation, the real value of kickers, which are fixed in nominal terms, declines between the time a recruit enters the military and actually use. Figure 2 shows that the decline in purchasing power of the kickers was largest for recruits entering service in the early to mid-1990s, and larger for 4-yr recruits than 2-yr recruits due to the shorter amount of time between entry and potential usage.

V. DATA AND TRENDS

The data used in this study were provided by Defense Manpower Data Center, and include information on every active duty enlistment contract signed at a Military Enlistment Processing Center between the FY 1988 and 2001. The data contain information on college benefit usage through June 2005, including for each veteran the date that benefit usage began (if ever), the date that usage ended, and the duration of usage in months. Our data also include information from annual in-service records that provide annual snapshots on military occupation, pay grade, date of latest enlistment, expiration of time in service, and other information through the end of FY 2004.

Our master data set contains 3.5 million records on individuals who signed enlistment contracts between FY 1988 and FY 2001. We constructed two data sets for this paper. The first contains individuals who reached the end of their first-term enlistment contracts by FY 2003. For example, of some 870,000 Army entrants, 439,309 stayed to the end of their initial enlistment contract, called Expiration of Term of Service (ETS), 262,132 separated, and 177,177 (40%) remained at least 1 yr beyond the first-term ETS. (7) Of those who separated, 146,378 (56%) used the GI Bill between the time of separation and June 2005.

The second data set contains all individuals, including those who served beyond their first term, who separated between 1988 and 2003 and were eligible to use the MGIB. Table 1 documents the total number of separations and MGIB use by YOS (years of service) category.

From Table 1 it can be seen that the percentage of separatees who use the MGIB, between 48 and 51%, is remarkably stable across services. MGIB usage rates are lower among individuals who served more than 6 yr, ranging from 34% for the Marine Corps to 39% for the Air Force.

Veterans who separated after 1995 are not observed for the full 10-yr eligibility window. This means that completed usage rates will be higher than those reported in Table 1. It also complicates comparisons of MGIB usage across separation cohorts. We therefore focus our attention on usage during the first 2 yr after separation.

[FIGURE 3 OMITTED]

Figure 3 shows GI Bill usage within 2 yr of separation among veterans with 3 or 4 yr of completed service, who comprise the bulk of GI Bill users. Two-yr usage rates hovered around 35% between 1993 and 2000 and rose sharply in all four services, reaching 45% for Army veterans in 2002.

Two-year usage trends were lower but otherwise similar for groups with longer completed periods of service. For example, 2-yr usage rates among veterans who separated before 1999 with seven or eight completed years ranged between 20 and 25%, and rose between 2000 and 2002 to about 35%.

[FIGURE 4 OMITTED]

Figure 4 displays the cumulative MGIB usage profiles of selected separation year cohorts of Army veterans who had three or four completed years of service. Because our usage data end in 2005, it is possible to track the 1993, 1994, and 1995 separation cohorts for the full 10 yr of their eligibility window; other cohorts can only be tracked for shorter periods. The figure shows that cumulative benefit use increases at a decreasing rate with time since separation. The figure also suggests that usage rose in response to increases in educational benefits for the 2001 and 2002 cohorts.

VI. IDENTIFYING THE EFFECTS OF EDUCATION BENEFITS

From a policy perspective, it is straightforward to measure the effect of education benefit changes on separation from the military and on education benefit usage. However, estimation of the pure incentive effect of education benefits on benefit usage is more difficult. Denote [CB.sub.i,t] the educational benefit available to the ith individual in the tth year after entry to military service. Higher [CB.sub.i,t] may attract military entrants who are more interested in going to college in the first place, more likely to separate, and more likely to use the benefit. The estimated effect of [CB.sub.i,t] may therefore overstate the pure incentive effect.

To estimate the incentive effect of [CB.sub.i,t], we exploit the fact that the long lags and vagaries of the legislative process should have introduced considerable uncertainty on the part of potential enlistees about the real value of benefits they would actually receive. Even a wellinformed recruit who followed the legislative process closely would have been unable to determine which of the competing visions and pieces of legislation would ultimately prevail. For example, neither of the key recommendations of the Principi (1999) Commission was adopted by the Congress.

The incentive effect can be estimated if the researcher can plausibly decompose [CB.sub.i,t] into two components: (1) the future real value expected when the veteran entered service in year [tau], denoted E([CB.sub.i,[tau]], and (2) the unanticipated change, or "benefit shock," that occurred between entry year [tau] and period t, denoted [DELTA][CB.sub.i,t-[tau]. By definition, benefit shocks are uncorrelated with unobservable individual characteristics. To do such a decomposition, one must make assumptions about how entrants form expectations about future benefits. One assumption is that youth have static expectations, projecting the real value of benefits in place at the time of entry to all future periods. Such a scenario assumes that recruits are unaware that basic benefit adjustments are not always equal to college cost increases and it assumes that recruits who received kickers are unaware that they are not indexed to college cost inflation. We experimented with several alternative expectations scenarios, all of which yielded similar results. For brevity, all estimates below are based on the scenario in which enlistees forecast future nominal MGIB benefits based on a 3-yr moving average (MA-3) of past, present, and future MGIB benefits. (8) Computations of expected future benefits also assume that enlistees forecast future cost college inflation based on past values. (9)

[FIGURE 5 OMITTED]

The annual percentage change in the MGIB, along with our forecast, is graphed in Figure 5. The MA-3 scheme forecasts positive nominal MGIB growth early on, when in reality no growth occurred. The MA-3 scheme assumes that recruits who entered in 1991 would have forecast positive MGIB growth, but less than the 17% growth that actually occurred. Recruits entering between 1992 and 1996 would have forecast declining MGIB growth and recruits entering after 1996 would have anticipated higher rates of nominal MGIB growth.

VII. FIRST-TERM RETENTION AND 2-YR MGIB USAGE

We begin our analysis by estimating simultaneously the decision to separate from the military after completing the first term of enlistment and the decision to use educational benefits within the first 2 yr of separation using Heckman's two-step probit model with sample selection. This analysis uses individuals in the first data set described in Section V above. The key explanatory variables are the anticipated and unanticipated components of education benefits. Because high-ability individuals may respond differently to benefit changes than low-ability individuals, the estimated models include interactions of AFQT with anticipated and unanticipated benefits. Our models also control for a wide range of individual-level characteristics.

The separation equation contains a measure of the net value of continuing a military career called the annualized cost of leaving (ACOL). (10) This measure varies by reenlistment decision year and is intended to control for trends in military compensation and civilian earnings during our data period that are above and beyond what is accounted for by the inclusion of individual-year effects in the separation equation. Because our ACOL measure does not vary with military rank, and the payoff to a future military career also depends on rank, we also include dummies for military rank in the separation equation. ACOL and the rank effects are excluded from the MGIB usage equations because they affect the value of separation (negatively), but do not affect the payoff to MGIB usage once separation has occurred.

We focus here on the estimated effects of education benefits; full estimates of the marginal effects and standard errors are contained in Tables B1-B4. For purposes of comparison, we also estimate the usage models using a simple probit specification that neglects the correlation between the unobservables that determine usage and separation from the military.

The estimated marginal effects of interest, evaluated at the means of the data, appear in Table 2. We first describe the results in detail for the Army and then briefly summarize the results for the other services. The estimated marginal effect of a $10,000 increase in the expected value of education benefits at entry is about 0.07 in both the simple and bivariate probit models, with a standard error of about 0.0036. The estimated marginal effect of a $10,000 benefit shock is slightly larger, 0.085 with a standard error of 0.012 in the bivariate probit model and a nearly identical 0.083 with a standard error of 0.011 in the simple probit model.

Army separatees with higher AFQT scores are more likely to have used their educational benefits. Evaluated at the means, each 10-point increase in AFQT is estimated to increase benefit usage by about 0.035 (SE = 0.0012). In addition, higher AFQT scores are estimated to increase slightly the sensitivity of usage to entry-level benefits, but reduce slightly (bivariate probit) estimates of the sensitivity to benefit shocks.

Each $10,000 increase in expected educational benefits at entry is estimated to increase separation from the Army by about 0.02 (SE = 0.003). This places an upper limit on the magnitude of selection effects of benefits that operate through changes in the composition of enlistees. Importantly, benefit shocks also exert a positive effect on separation that is slightly larger in magnitude, 0.035 (SE = 0.01), than that of the entry benefit. Thus, changes in education benefits have important incentive effects above and beyond any impact that they may have on selection into the military.

We now briefly summarize the estimates for the other services. The estimated effects of entry benefit changes and benefit changes on benefit usage were somewhat smaller for the Marine Corps (6% and 4%) and smaller still for the Air Force (3.3% and 2.6%) than for the Army. The estimated effects of entry benefits in the Navy were just 2.0% and the estimated effects of benefit shocks in the Navy were negative and insignificant. The estimated effects of benefit changes on separation were very large--almost unbelievably so--in the Air Force, small and statistically insignificant in the Marine Corps, and of mixed sign and significance for the Navy. Taken as a whole, the results are consistent with the predictions of theory, with higher benefits leading to higher separation rates and higher benefit usage rates, even--with the possible exception of the Navy--holding constant the composition of recruits.

The models we estimated allow for the effects of education benefits to interact with AFQT. In addition to evaluating the marginal effects of education benefits at the mean, Table 2 shows their impact at an AFQT of 80, about 20 percentage points higher than the mean for Army, Navy, and Marine Corps veterans, and about 12 percentage points higher than the mean for Air Force veterans. The estimated marginal effect of a $10,000 increase in benefits at entry on Army veteran usage is about 0.09 evaluated at AFQT = 80, roughly 2 percentage points higher than at the mean AFQT. The estimated effect of a $10,000 benefit shock at AFQT = 80, by contrast, is about a percentage point lower than the estimated effect at the mean (0.073 vs. 0.085). The estimated effects of AFQT on the marginal usage effects were much smaller in the other services, with two exceptions: the effects of entry benefits on Air Force veterans (0.043 at AFQT = 80) and benefit shocks on Marine Corps veterans (0.024 at AFQT = 80). Turning to separations, the estimated effects of benefit shocks were higher at AFQT = 80 than at the mean AFQT in the Army (0.052 vs. 0.035), Navy (0.053 vs. 0.038), and Air Force (0.14 vs. 0.11). The same is true of entry benefits in the Army (0.033 vs. 0.022).

Turning to the effects of other covariates, reported in Appendix B, the probability of separation and education benefit usage were strongly decreasing in age at entry into the military, and were lower for males, married individuals, and individuals with dependents. Benefit usage was higher for high school graduates but lower for individuals who had some college at entry. Blacks were between 6 and 12 percentage points less likely to leave military service than otherwise comparable individuals of other races, and about 0.8-1.6 percentage points more likely to use education benefits in the Navy and the Army. Hispanics were between 2 and 6.7 percentage points more likely to use education benefits. Notably, higher current unemployment is associated with lower likelihoods of separation (about 1-2 percentage points per 2 percentage point unemployment rate increase) and a higher likelihood of education benefit usage (again, 1-2 percentage points per 2 percentage point increase).

Finally, we consider the effects of controlling for selection. In addition to the bivariate probit estimates, Table 2 contains simple probit estimates of the effects of education benefits on benefit usage. The estimated effects of entry benefits were within 0.5% in the Marine Corps and 1.2% in the Army. (11) The simple probit estimated effects of entry benefits were 13% and 17% higher than the bivariate estimates for the Air Force and Navy. Put differently, the estimated effects of entry benefits differed by an average of 0.08 standard errors, and none by more than 0.36 standard errors. (12)

Relative to simple probit, the bivariate probit estimated effects of benefit shocks were only 2.4% (not percentage points) higher in the Army, 7% higher in the Air Force, and 4% higher in the Marine Corps. The simple and bivariate probit estimates, including the Navy's inexplicably negative estimate, were an average of 0.18 standard errors apart. The simple and bivariate probit estimates of the effect of AFQT are within 1% in the Army, 2% in the Navy, 7% in the Marine Corps, and 7% in the Air Force. Despite the fairly similar estimated effects, the difference in terms of standard errors is somewhat larger than in the case of entry benefits or benefit shocks--0.3, 0.5, 2.7, and 1.5--for the simple reason that the estimated effects of AFQT are estimated with relatively greater precision.

The main conclusion to take away is that failing to control for the simultaneity of the decision to separate from the military and use education benefits does not appear to lead to serious bias in the estimated effects of education benefits, particularly the unanticipated component.

VIII. BENEFIT USAGE BEYOND THE FIRST 2 YR: A HAZARD APPROACH

The bivariate probit approach has the advantage of jointly modeling decisions to separate and use benefits, but has the disadvantage of ignoring the potential usage effects of benefit shocks that occur after separation. Veterans are eligible to use their education benefits up to 10 yr after separation, during which period the real value of benefits can change. Estimation for longer periods of time requires accounting for the time-varying nature of the real value of education benefits, which are affected by college tuition inflation and congressional legislation.

The discrete time hazard framework lends itself to precisely this analysis. The hazard rate is defined as the probability that a recruit begins to use the benefit in a given time period after service given that he or she has not used the benefit before that period. We divide the 10-yr window of benefit eligibility into 10 discrete time periods, and specify the hazard rate for each period to be a function of time period and a vector of explanatory variables, the most important of which is the level of education benefits. We specified a flexible semiparametric form in which the baseline hazard can take on any of 10 discrete values for each of the 10 yr after separation, thus allowing the model to fit a wide variety of time patterns of usage, including the concave profiles observed in Figure 4. In addition to the level of education benefits, the explanatory variables include controls for individual-level attributes and economic conditions both at the time of enlistment and the time of potential usage.

While the hazard models do not account for the potential simultaneity and correlation between the decision to separate from the military and use education benefits, the results in Section VII suggest that failing to control for selection had only a minor impact on the estimated effects of benefits and AFQT. We therefore estimated the hazard models of education benefit usage using data on all separatees (the second data set described in Section V above), not just those who separated after their first term.

We focus here on the estimated effects of education benefits and AFQT; full model estimates are contained in Appendix C. (13) Table 3 contains point estimates and bootstrapped (100 replications) 95% confidence intervals of the effects of education benefits and AFQT on MGIB usage 2, 5, and 10 yr after leaving military service. We show four cases for each military service: a base case evaluated at the means of all of the covariates, a $10,000 increase in entry benefits, a $10,000 benefit shock, and a 10-point increase in AFQT.

For purposes of comparison, we first consider hazard estimates for 2-yr usage. Each $10,000 increase in entry benefits is estimated to increase MGIB usage by about 5 percentage points in the Army, Air Force, and Marine Corps. The estimated effect of 1.8 percentage points in the Navy, although lower than for the other three services, is close to the probit estimate of 2 percentage points. The probit estimates for the Army and Marine Corps, about 7 and 6 percentage points, respectively, are higher than the hazard estimates of 5.5 and 4.7 percentage points. The difference is most likely due to the fact that the hazard estimates are based on usage within 10 yr of separation, whereas the probit estimates are based on the first 2 yr. The probit estimates for the Air Force were somewhat lower than the hazard estimate--3 versus 5.2 percentage points.

The estimated effects of a $10,000 benefit shock were similar across the services, ranging from 3.6 percentage points in the Navy to 5.5 percentage points in the Army. The 95% confidence intervals are less than 5 percentage points in all four services. The narrowest confidence interval is for the Army, with a $10,000 benefit shock estimated to raise mean benefit usage from a base of 32%-37.5%, the latter confidence interval being 36.1%-38.9%, or 2.8 percentage points wide. The widest confidence interval is obtained for the Marine Corps, with the same benefit shock estimated to raise usage from a base of 31.1%-35%, with a 95% confidence interval for the latter ranging between 32.5 and 37.5%.

The most convenient way to compare our estimates with those obtained by other researchers is to compute elasticities. For the Army, a $10,000 educational benefit increase is about a one-third increase based on the Army sample-period average of $30,000. The 5.5 percentage-point increase in 2-yr usage implied by the Army hazard estimates amounts to a percentage increase of about 17%, for an estimated elasticity of about 0.5. The 5-and 10-yr usage elasticities are 0.44 and 0.38, respectively. The estimated elasticities for the other services are somewhat smaller, the smallest being benefits at entry in the Navy, at just 0.17. Most range between 0.3 and 0.5, and all are of the same order of magnitude as estimates obtained from studies of the effects of federal aid to education.

IX. DURATION OF BENEFIT USAGE

Thus far, we have shown that education benefits have a significant effect on the likelihood of their usage. However, most research on the effects of veterans' benefits examines their impact on the eventual level of education attained and, ultimately, earnings. Ideally, a study of the duration of usage would relate the probability that each veteran uses the benefit at each point in time (month) after separating from the military, for a period of 10 yr or until those benefits are exhausted after 36 mo of usage. Unfortunately, data limitations prevent such an analysis. Rather, we have information for each veteran on whether and when they started using their benefit and total months of benefit usage as of June 2005. Since we do not observe the actual time pattern of usage, we cannot study month-by-month likelihood of usage as a function of benefits available at that particular point in time. Nevertheless, it is useful to study months of benefit usage as a function of benefits available at the start of usage.

Table 4 shows the percentage distribution of months of MGIB usage for all veterans in our data across usage categories defined as 6-mo intervals, as well as the mean number of months of benefits used and the percentage still enrolled for each of three separation-period cohorts: 1991-1994, 1995-1999, and 2000-2004. The table breaks down months of usage by eligibility for CF benefits. The average duration of benefit usage was about 17 mo for veterans who separated between 1991 and 1994, with an average duration of about 21 mo for CF-eligible veterans and 16 mo for those eligible only for MGIB.

Notice that only 1.7% of veterans who separated between 1991 and 1994 were still enrolled in the program in June 2005. The percentage still enrolled rises dramatically for those separating after 1994, with 54% of veterans who separated in the 2000-2004 period still enrolled in the program. This censoring accounts for the decline in average months of usage in more recent separation cohorts, and likely also accounts for the declining usage differential between CF-eligible and ineligible veterans.

Considering the limitations imposed by our data, the Cox duration model provides a suitable empirical framework in which to analyze the duration of usage. The Cox model deals naturally with censoring while providing the added advantage of not having to estimate baseline hazards. We estimated models for duration of MGIB usage as a function of the same covariates as those included in our probit model. Again, we simulated the effects of a $10,000 increase in expected entry benefits, unanticipated benefit shocks, and a 10 percentage-point increase in AFQT. The results are contained in Table 5.

The benefit and benefit interaction variables were statistically significant in these models and in the directions expected. However, in contrast to usage probabilities, the mean duration of usage is estimated to be quite inelastic with respect to even substantial changes in benefits. The Army estimates suggest that a $10,000 increase in expected benefits at entry will increase the mean duration of usage by about 1 mo while a $10,000 benefit shock increases it by 1.5 mo. The estimated effects are somewhat smaller for the Marine Corps and Navy and somewhat larger for the Air Force. The estimated effects of a 10 percentage-point increase in AFQT were of the same order of magnitude.

X. CONCLUSIONS

We have conducted the most comprehensive analysis on GI Bill usage among AVF-era veterans to date. We find that a $10,000 increase in veterans' education benefits increases the probability of MGIB usage by about 5 percentage points. Interestingly, though, the average duration of usage is estimated to be relatively insensitive to the dollar value of benefits. The effects of education benefits on separation from the military were mixed, with small positive effects for the Army on the order of 2-3 percentage points, large positive effects for the Air Force on the order of 7-11 percentage points, but insignificant effects of mixed sign for the Navy and Marine Corps.

The estimates in this paper can be used to evaluate the impact of the Post-9/11 Veterans Education Assistance Act of 2008. Although a complete benefit-cost analysis of this new law is beyond the scope of this paper, our estimates offer some insight regarding the likely magnitude of the effect of the proposal on benefit usage. The legislation roughly doubles the generosity of the education benefit from its current $38,000 potential level, and expands the window of eligibility from 10 to 15 yr. Taken at face value, our estimates suggest that usage would rise by about 20 percentage points from a current rate of about 50% to around 70%.

During the congressional debate over this legislation, the Department of Defense expressed concern about the likely retention effects of the new benefit levels scheduled to go into effect in August of 2009. Our estimates suggest that higher education benefits at entry will increase separation after the first term in all services except the Navy. In the Army, the service that has used education benefits most heavily as a recruiting tool and the service whose retention is most likely to be affected by higher benefits, first-term retention could decline by as much as 8-12 percentage points--from its current rate of about 40% to between 28 and 32%--as a result of the new law. (14) To the extent that such retention declines materialize with the new legislation, the Army would have to offset the declines with the use of higher reenlistment bonuses and other force management tools it has at its disposal. On the other hand, to the extent that higher educational benefits enhance recruiting, as past research suggests they will do, the Army and other services will be able to reduce the number of military recruiters and other resources devoted to recruiting.

Finally, it is useful to observe that concern about the potential adverse retention effects of enhanced educational benefits caused the Department of Defense to seek and Congress to adopt a provision that allows military personnel who serve ten or more years to transfer unused GI Bill entitlements to their spouses and children. Because a very high percentage of military personnel with ten or more years of service are married and have children, this transferability provision is potentially very valuable to them and could serve to reduce the adverse separation effects of the new legislation. The provision may lead to future MGIB usage rates well above those forecast above based on own-veteran usage. Clearly, analysis of the various provisions of the new legislation will be an important area of future research.

ABBREVIATIONS

2-yr obligors: 2-Year Obligators

ACOL: Annualized Cost of Leaving

AFQT: Armed Force Qualification Test

AVF: All-Volunteer Military Force

CPI: Consumer Price Index

CF: College Fund

ETS: Expiration of Term of Service

FY: Fiscal Year

MA-3: 3-Year Moving Average

MGIB: Montgomery GI Bill

VEAP: Veterans Educational Assistance Program

doi: 10.1111/j.1465-7295.2009.00233.x

APPENDIX A SAMPLE MEANS
TABLE A1

Sample Means of Variables in Probit Models for Separation and
MGIB Usage

 Army

 All First- Separated
 Termers

Observations 395,307 238,419
Education benefit ($1K)
 Expected value at entry 29.85 30.76
 Unexpected change 2.47 2.18
 AFQT score 58.83 60.20
 EVE x AFQT 1803.30 1907.31
 UC x AFQT 144.51 129.21
 Entry bonus ($1K) 1.05 0.99
Race-ethnic group (white omitted)
 Black 0.23 0.19
 Hispanic 0.08 0.08
 Other 0.04 0.04
Personal characteristics
 Entry age 20-22 0.29 0.29
 Entry age 23-25 0.10 0.09
 Entry age 26 plus 0.06 0.05
 Male 0.86 0.87
 Married at ETS 0.37 0.27
 No. of dependents at ETS 0.98 0.86
 Some college or better 0.03 0.03
 High school graduate 0.89 0.89
Economic conditions
 Current unemployment rate 5.58 5.64
 Entry unemployment rate 5.80 5.85
 Military/civil pay ratio 1.04 1.03
 CG-HSG differential 1.62 1.62
State socioeconomic characteristics
 Percent college graduates 33.72 33.79
 Percent veterans 40.86 41.27
 Family income ($1K) 40.80 41.29
Enlistment term (omitted = 4-yr term)
 Two-yr term 0.09 0.11
 Three-yr term 0.35 0.35
 Five-yr term 0.06 0.05
 Six-yr term 0.03 0.03
Military occupation group (admin omitted)
 Combat arms 0.32 0.35
 Electronic repair 0.06 0.05
 Communications 0.12 0.13
 Medical 0.06 0.06
 Other technical 0.03 0.02
 Mechanical equipment 0.16 0.15
 Craftsmen 0.02 0.02
 Service and supply 0.12 0.11
Separation year (1992 omitted)
 1993 0.12 0.14
 1994 0.10 0.10
 1995 0.08 0.09
 1996 0.09 0.09
 1997 0.08 0.08
 1998 0.07 0.07
 1999 0.08 0.07
 2000 0.07 0.07
 2001 0.06 0.05
 2002 0.07 0.06
 2003 0.08 0.07

 Navy

 All First- Separated
 Termers

Observations 323,654 183,965
Education benefit ($1K)
 Expected value at entry 28.42 28.01
 Unexpected change 3.32 2.83
 AFQT score 58.25 58.15
 EVE x AFQT 1673.58 1649.91
 UC x AFQT 199.84 170.67
 Entry bonus ($1K) 0.70 0.66
Race-ethnic group (white omitted)
 Black 0.20 0.16
 Hispanic 0.09 0.09
 Other 0.07 0.06
Personal characteristics
 Entry age 20-22 0.27 0.27
 Entry age 23-25 0.07 0.07
 Entry age 26 plus 0.04 0.03
 Male 0.86 0.86
 Married at ETS 0.36 0.29
 No. of dependents at ETS 0.97 0.89
 Some college or better 0.02 0.02
 High school graduate 0.91 0.91
Economic conditions
 Current unemployment rate 5.52 5.55
 Entry unemployment rate 5.95 6.00
 Military/civil pay ratio 1.03 1.03
 CG-HSG differential 1.63 1.62
State socioeconomic characteristics
 Percent college graduates 33.90 33.75
 Percent veterans 41.20 41.90
 Family income ($1K) 41.61 41.80
Enlistment term (omitted = 4-yr term)
 Two-yr term 0.05 0.06
 Three-yr term 0.11 0.12
 Five-yr term 0.09 0.07
 Six-yr term 0.12 0.11
Military occupation group (admin omitted)
 Combat arms 0.13 0.16
 Electronic repair 0.14 0.14
 Communications 0.11 0.10
 Medical 0.08 0.06
 Other technical 0.01 0.01
 Mechanical equipment 0.33 0.34
 Craftsmen 0.06 0.06
 Service and supply 0.05 0.04
Separation year (1992 omitted)
 1993 0.11 0.13
 1994 0.10 0.12
 1995 0.10 0.11
 1996 0.08 0.08
 1997 0.11 0.12
 1998 0.08 0.08
 1999 0.07 0.07
 2000 0.07 0.06
 2001 0.07 0.06
 2002 0.07 0.05
 2003 0.08 0.05

 Air Force

 All First- Separated
 Termers

Observations 215,562 95,387
Education benefit ($1K)
 Expected value at entry 24.77 24.78
 Unexpected change 3.34 2.88
 AFQT score 66.24 67.36
 EVE x AFQT 1639.52 1668.42
 UC x AFQT 219.36 192.83
 Entry bonus ($1K)
Race-ethnic group (white omitted)
 Black 0.14 0.10
 Hispanic 0.05 0.05
 Other 0.07 0.06
Personal characteristics
 Entry age 20-22 0.05 0.03
 Entry age 23-25 0.02 0.01
 Entry age 26 plus 0.02 0.02
 Male 0.78 0.78
 Married at ETS 0.52 0.44
 No. of dependents at ETS 1.06 0.91
 Some college or better 0.03 0.04
 High school graduate 0.94 0.94
Economic conditions
 Current unemployment rate 5.30 5.30
 Entry unemployment rate 5.88 5.95
 Military/civil pay ratio 1.04 1.03
 CG-HSG differential 1.63 1.63
State socioeconomic characteristics
 Percent college graduates 33.73 33.75
 Percent veterans 41.26 41.71
 Family income ($1K) 41.10 41.58
Enlistment term (omitted = 4-yr term)
 Two-yr term
 Three-yr term
 Five-yr term
 Six-yr term 0.05 0.04
Military occupation group (admin omitted)
 Combat arms 0.09 0.10
 Electronic repair 0.11 0.11
 Communications 0.07 0.07
 Medical 0.10 0.10
 Other technical 0.05 0.05
 Mechanical equipment 0.25 0.23
 Craftsmen 0.06 0.06
 Service and supply 0.10 0.11
Separation year (1992 omitted)
 1993 0.11 0.12
 1994 0.09 0.09
 1995 0.08 0.08
 1996 0.11 0.11
 1997 0.10 0.10
 1998 0.09 0.10
 1999 0.09 0.10
 2000 0.08 0.10
 2001 0.06 0.04
 2002 0.09 0.06
 2003 0.06 0.05

 Marine Corps

 All First- Separated
 Termers

Observations 204,302 146,539
Education benefit ($1K)
 Expected value at entry 25.62 25.56
 Unexpected change 3.41 3.01
 AFQT score 58.80 58.16
 EVE x AFQT 1513.46 1492.77
 UC x AFQT 199.62 173.47
 Entry bonus ($1K)
Race-ethnic group (white omitted)
 Black 0.13 0.11
 Hispanic 0.11 0.11
 Other 0.17 0.16
Personal characteristics
 Entry age 20-22 0.23 0.23
 Entry age 23-25 0.05 0.04
 Entry age 26 plus 0.01 0.01
 Male 0.95 0.95
 Married at ETS 0.45 0.40
 No. of dependents at ETS 1.05 0.96
 Some college or better 0.01 0.01
 High school graduate 0.95 0.95
Economic conditions
 Current unemployment rate 5.32 5.35
 Entry unemployment rate 5.96 6.01
 Military/civil pay ratio 1.03 1.03
 CG-HSG differential 1.63 1.63
State socioeconomic characteristics
 Percent college graduates 34.21 34.21
 Percent veterans 40.89 41.28
 Family income ($1K) 42.12 42.36
Enlistment term (omitted = 4-yr term)
 Two-yr term
 Three-yr term
 Five-yr term
 Six-yr term 0.05 0.05
Military occupation group (admin omitted)
 Combat arms 0.34 0.38
 Electronic repair 0.05 0.03
 Communications 0.08 0.08
 Medical 0.00 0.00
 Other technical 0.02 0.02
 Mechanical equipment 0.16 0.14
 Craftsmen 0.03 0.03
 Service and supply 0.16 0.17
Separation year (1992 omitted)
 1993 0.07 0.08
 1994 0.08 0.09
 1995 0.08 0.09
 1996 0.09 0.10
 1997 0.10 0.10
 1998 0.10 0.10
 1999 0.10 0.10
 2000 0.08 0.09
 2001 0.06 0.04
 2002 0.10 0.08
 2003 0.11 0.09

EVE = expected value at entry; CG-HSC = college-graduate <en>
high-school graduate differential


APPENDIX B
TABLE B1
Marginal Effects (ME), 2-Yr Use and Separation: Army

 2-Yr Use

 Simple Probit

 ME SE (ME)
Education benefit ($1K)
 Expected value at entry 0.00262 0.00082
 Unexpected change 0.01319 0.00263
 AFQT score 0.00136 0.00038
 EVE x AFQT 0.00009 0.00001
 UC x AFQT -0.00007 0.00004
 Entry bonus ($1K) -0.00108 0.00058
Race-ethnic group (white omitted)
 Black 0.0163 0.0053
 Hispanic 0.0674 0.0059
 Other 0.1057 0.0065
Personal characteristics
 Entry age 20-22 -0.0182 0.0031
 Entry age 23-25 -0.0762 0.0056
 Entry age 26 plus -0.0908 0.0060
 Male -0.0289 0.0061
 Married at ETS -0.0640 0.0031
 No. of dependents at ETS -0.0271 0.0024
 Some college or better -0.1243 0.0116
 High school graduate 0.0531 0.0064
Economic conditions
 Current unemployment rate 0.0104 0.0015
 Entry unemployment rate -0.0029 0.0015
 Military/civil pay ratio 0.1839 0.0184
 CG-HSG differential -0.0604 0.0388
State socioeconomic characteristics
 Percent college graduates 0.0052 0.0004
 Percent veterans -0.0016 0.0004
 Family income ($1K) 0.0010 0.0001
Enlistment term (omitted = 4-yr term)
 Two-yr term 0.0787 0.0116
 Three-yr term 0.0115 0.0055
 Five-yr term 0.0077 0.0078
 Six-yr term -0.0307 0.0108
Military occupation group (admin omitted)
 Combat arms -0.0254 0.0057
 Electronic repair -0.0613 0.0092
 Communications -0.0192 0.0066
 Medical 0.0413 0.0081
 Other technical -0.0055 0.0096
 Mechanical equipment -0.0781 0.0054
 Craftsmen -0.0905 0.0098
 Service and supply -0.0434 0.0067
Separation year (1992 omitted)
 1993 -0.0475 0.0172
 1994 -0.0630 0.0164
 1995 -0.0695 0.0150
 1996 -0.0874 0.0161
 1997 -0.0793 0.0161
 1998 -0.0554 0.0165
 1999 -0.0716 0.0186
 2000 -0.0252 0.0180
 2001 -0.0158 0.0188
 2002 0.0273 0.0169
 2003 -0.0950 0.0194
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho
 Observations 238,419
 Retained
 Separated

 Bivariate Probit

 ME SE (ME)
Education benefit ($1K)
 Expected value at entry 0.00223 0.00076
 Unexpected change 0.01181 0.00236
 AFQT score 0.00112 0.00034
 EVE x AFQT 0.00008 0.00001
 UC x AFQT -0.00006 0.00003
 Entry bonus ($1K) -0.00111 0.00055
Race-ethnic group (white omitted)
 Black -0.0034 0.0050
 Hispanic 0.0585 0.0056
 Other 0.0933 0.0066
Personal characteristics
 Entry age 20-22 -0.0152 0.0029
 Entry age 23-25 -0.0680 0.0050
 Entry age 26 plus -0.0836 0.0053
 Male -0.0308 0.0060
 Married at ETS -0.0896 0.0046
 No. of dependents at ETS -0.0256 0.0031
 Some college or better -0.1003 0.0100
 High school graduate 0.0475 0.0060
Economic conditions
 Current unemployment rate 0.0091 0.0014
 Entry unemployment rate -0.0025 0.0014
 Military/civil pay ratio 0.1580 0.0170
 CG-HSG differential -0.0612 0.0338
State socioeconomic characteristics
 Percent college graduates 0.0046 0.0003
 Percent veterans -0.0016 0.0004
 Family income ($1K) 0.0010 0.0001
Enlistment term (omitted = 4-yr term)
 Two-yr term 0.0865 0.0126
 Three-yr term 0.0163 0.0056
 Five-yr term 0.0034 0.0070
 Six-yr term -0.0321 0.0095
Military occupation group (admin omitted)
 Combat arms -0.0085 0.0053
 Electronic repair -0.0453 0.0078
 Communications -0.0065 0.0062
 Medical 0.0426 0.0079
 Other technical 0.0020 0.0088
 Mechanical equipment -0.0606 0.0047
 Craftsmen -0.0666 0.0086
 Service and supply -0.0301 0.0063
Separation year (1992 omitted)
 1993 -0.0495 0.0171
 1994 -0.0737 0.0152
 1995 -0.0778 0.0140
 1996 -0.0889 0.0150
 1997 -0.0833 0.0151
 1998 -0.0668 0.0160
 1999 -0.0826 0.0171
 2000 -0.0402 0.0169
 2001 -0.0352 0.0182
 2002 -0.0046 0.0165
 2003 -0.1129 0.0177
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho 0.3234 0.0183
 Observations 395,307
 Retained 156,888
 Separated 238,419

 Separation

 Bivariate Probit

 ME SE (ME)
Education benefit ($1K)
 Expected value at entry -0.00108 0.00050
 Unexpected change -0.00125 0.00224
 AFQT score 0.00004 0.00023
 EVE x AFQT 0.00006 0.00001
 UC x AFQT 0.00008 0.00003
 Entry bonus ($1K) -0.00059 0.00071
Race-ethnic group (white omitted)
 Black -0.0598 0.0075
 Hispanic 0.0385 0.0069
 Other 0.0027 0.0074
Personal characteristics
 Entry age 20-22 0.0279 0.0035
 Entry age 23-25 0.0343 0.0065
 Entry age 26 plus 0.0219 0.0096
 Male -0.0273 0.0075
 Married at ETS -0.1894 0.0176
 No. of dependents at ETS -0.0145 0.0112
 Some college or better 0.1019 0.0089
 High school graduate -0.0138 0.0055
Economic conditions
 Current unemployment rate -0.0054 0.0011
 Entry unemployment rate 0.0032 0.0013
 Military/civil pay ratio -0.0610 0.0216
 CG-HSG differential -0.0322 0.0304
State socioeconomic characteristics
 Percent college graduates -0.0001 0.0003
 Percent veterans -0.0003 0.0004
 Family income ($1K) 0.0013 0.0001
Enlistment term (omitted = 4-yr term)
 Two-yr term -0.2211 0.0118
 Three-yr term -0.0949 0.0072
 Five-yr term 0.0738 0.0092
 Six-yr term 0.1251 0.0081
Military occupation group (admin omitted)
 Combat arms 0.1070 0.0058
 Electronic repair 0.0608 0.0057
 Communications 0.0799 0.0066
 Medical 0.0138 0.0085
 Other technical 0.0630 0.0070
 Mechanical equipment 0.0505 0.0059
 Craftsmen 0.0858 0.0079
 Service and supply 0.0474 0.0066
Separation year (1992 omitted)
 1993 -0.0790 0.0087
 1994 -0.1683 0.0052
 1995 -0.1751 0.0056
 1996 -0.1595 0.0128
 1997 -0.1600 0.0115
 1998 -0.1676 0.0156
 1999 -0.1800 0.0143
 2000 -0.1512 0.0164
 2001 -0.1606 0.0184
 2002 -0.1705 0.0190
 2003 -0.0891 0.0218
Rank at ETS (E4 omitted)
 Less than E4 0.3849 0.0059
 E5 -0.4248 0.0180
 More than E5 -0.6196 0.0050
ACOL ($1K) -0.0119 0.0010
 Rho
 Observations
 Retained
 Separated

EVE = expected value at entry; CG-HSC = college-graduate <en>
high-school graduate differential.

TABLE B2
Marginal Effects (ME), 2-Yr Use and Separation: Navy

 2-Yr Use

 Simple Probit

 ME SE(ME)

Education benefit ($1K)
 Expected value at entry 0.00257 0.00278
 Unexpected change 0.00626 0.00327
 AFQT score 0.00385 0.00083
 EVE x AFQT 0.00000 0.00003
 UC x AFQT -0.00015 0.00005
 Entry bonus ($1K) 0.00198 0.00108
Race-ethnic group (white omitted)
 Black 0.0086 0.0043
 Hispanic 0.0461 0.0053
 Other 0.0729 0.0067
Personal characteristics
 Entry age 20-22 -0.0112 0.0028
 Entry age 23-25 -0.0604 0.0051
 Entry age 26 plus -0.0910 0.0083
 Male -0.0035 0.0059
 Married at ETS -0.0704 0.0038
 No. of dependents at ETS -0.0340 0.0029
 Some college or better -0.0220 0.0122
 High school graduate 0.0489 0.0065
Economic conditions
 Current unemployment rate 0.0067 0.0019
 Entry unemployment rate -0.0003 0.0012
 Military/civil pay ratio 0.1669 0.0207
 CG-HSG differential -0.0418 0.0354
State socioeconomic characteristics
 Percent college grads 0.0053 0.0005
 Percent veterans -0.0009 0.0004
 Family income ($1K) 0.0010 0.0001
Enlistment term (omitted = 4-yr term)
 Two-yr term 0.0039 0.0109
 Three-yr term 0.0199 0.0088
 Five-yr term -0.0376 0.0111
 Six-yr term -0.0840 0.0110
Military occupation group (admin omitted)
 Combat arms 0.0038 0.0086
 Electronic repair -0.0349 0.0075
 Communications 0.0276 0.0059
 Medical 0.0379 0.0092
 Other technical 0.0523 0.0134
 Mechanical equipment -0.0296 0.0052
 Craftsmen -0.0883 0.0058
 Service and supply -0.0268 0.0069
Separation year (1992 omitted)
 1993 0.0084 0.0166
 1994 0.0029 0.0135
 1995 0.0053 0.0151
 1996 -0.0268 0.0156
 1997 -0.0369 0.0141
 1998 -0.0340 0.0190
 1999 -0.0013 0.0318
 2000 0.0467 0.0239
 2001 0.1304 0.0311
 2002 0.1667 0.0252
 2003 0.0539 0.0351
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho
 Observations 183,965
 Retained
 Separated

 Bivariate Probit

 ME SE(ME)

Education benefit ($IK)
 Expected value at entry 0.00160 0.00228
 Unexpected change 0.00594 0.00286
 AFQT score 0.00330 0.00068
 EVE x AFQT 0.00001 0.00003
 UC x AFQT -0.00013 0.00004
 Entry bonus ($1K) 0.00201 0.00101
Race-ethnic group (white omitted)
 Black -0.0062 0.0042
 Hispanic 0.0427 0.0051
 Other 0.0588 0.0069
Personal characteristics
 Entry age 20-22 -0.0110 0.0026
 Entry age 23-25 -0.0595 0.0047
 Entry age 26 plus -0.0921 0.0075
 Male -0.0039 0.0055
 Married at ETS -0.0830 0.0046
 No. of dependents at ETS -0.0322 0.0033
 Some college or better -0.0161 0.0110
 High school graduate 0.0453 0.0058
Economic conditions
 Current unemployment rate 0.0056 0.0018
 Entry unemployment rate -0.0003 0.0012
 Military/civil pay ratio 0.1511 0.0199
 CG-HSG differential -0.0352 0.0300
State socioeconomic characteristics
 Percent college grads 0.0048 0.0004
 Percent veterans -0.0009 0.0004
 Family income ($1K) 0.0010 0.0001
Enlistment term (omitted = 4-yr term)
 Two-yr term -0.0021 0.0094
 Three-yr term 0.0153 0.0103
 Five-yr term -0.0419 0.0105
 Six-yr term -0.0906 0.0095
Military occupation group (admin omitted)
 Combat arms 0.0211 0.0095
 Electronic repair -0.0227 0.0064
 Communications 0.0278 0.0051
 Medical 0.0310 0.0089
 Other technical 0.0421 0.0124
 Mechanical equipment -0.0171 0.0057
 Craftsmen -0.0774 0.0059
 Service and supply -0.0242 0.0065
Separation year (1992 omitted)
 1993 -0.0094 0.0185
 1994 -0.0103 0.0160
 1995 -0.0110 0.0171
 1996 -0.0434 0.0174
 1997 -0.0511 0.0165
 1998 -0.0478 0.0196
 1999 -0.0228 0.0294
 2000 0.0163 0.0245
 2001 0.0802 0.0323
 2002 0.1099 0.0273
 2003 0.0023 0.0334
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho 0.2281 0.0219
 Observations 323,654
 Retained 139,689
 Separated 183,965

 Separation

 Bivariate Probit

 ME SE(ME)

Education benefit ($IK)
 Expected value at entry -0.00903 0.00432
 Unexpected change -0.00020 0.00672
 AFQT score 0.00222 0.00145
 EVE x AFQT -0.00072 0.00159
 UC x AFQT 0.00010 0.00005
 Entry bonus ($1K) 0.00007 0.00007
Race-ethnic group (white omitted)
 Black -0.0790 0.0102
 Hispanic 0.0437 0.0054
 Other -0.0630 0.0055
Personal characteristics
 Entry age 20-22 0.0142 0.0036
 Entry age 23-25 0.0028 0.0070
 Entry age 26 plus -0.0447 0.0114
 Male 0.0407 0.0075
 Married at ETS -0.1569 0.0275
 No. of dependents at ETS -0.0102 0.0174
 Some college or better -0.0285 0.0171
 High school graduate -0.0295 0.0069
Economic conditions
 Current unemployment rate -0.0083 0.0021
 Entry unemployment rate 0.0009 0.0016
 Military/civil pay ratio -0.0512 0.0276
 CG-HSG differential 0.0525 0.0296
State socioeconomic characteristics
 Percent college grads -0.0008 0.0004
 Percent veterans 0.0006 0.0004
 Family income ($1K) 0.0008 0.0001
Enlistment term (omitted = 4-yr term)
 Two-yr term -0.3199 0.0214
 Three-yr term -0.2088 0.0361
 Five-yr term 0.0192 0.0162
 Six-yr term 0.1297 0.0153
Military occupation group (admin omitted)
 Combat arms 0.0793 0.0172
 Electronic repair 0.0429 0.0198
 Communications 0.0680 0.0106
 Medical -0.2196 0.0193
 Other technical -0.0344 0.0120
 Mechanical equipment 0.0679 0.0090
 Craftsmen 0.0845 0.0160
 Service and supply -0.0401 0.0104
Separation year (1992 omitted)
 1993 -0.2165 0.0833
 1994 -0.1939 0.0813
 1995 -0.2483 0.0765
 1996 -0.3007 0.0729
 1997 -0.2754 0.0741
 1998 -0.2507 0.0751
 1999 -0.2626 0.0854
 2000 -0.2692 0.0842
 2001 -0.2936 0.0921
 2002 -0.2431 0.0903
 2003 -0.2465 0.1049
Rank at ETS (E4 omitted)
 Less than E4 0.3708 0.0141
 E5 -0.3784 0.0187
 More than E5 -0.5229 0.0163
ACOL ($1K) -0.0144 0.0012
 Rho
 Observations
 Retained
 Separated

EVE = expected value at entry; CG-HSC = college-graduate <en>
high-school graduate differential.

TABLE B3
Marginal Effects (ME), 2-Yr Use and Separation: Air Force

 2-Yr Use

 Simple Probit

 ME SE (ME)
Education benefit ($1K)
 Expected value at entry -0.00193 0.00344
 Unexpected change 0.00621 0.00396
 AFQT score 0.00108 0.00128
 EVE x AFQT 0.00008 0.00005
 UC x AFQT -0.00005 0.00004
Race-ethnic group (white omitted)
 Black 0.0025 0.0060
 Hispanic 0.0237 0.0072
 Other 0.0310 0.0084
Personal characteristics
 Entry age 20-22 -0.3269 0.0023
 Entry age 23-25 -0.3015 0.0023
 Entry age 26 plus -0.0840 0.0102
 Male -0.0311 0.0091
 Married at ETS -0.0686 0.0042
 No. of dependents at ETS -0.0501 0.0027
 Some college or better -0.1057 0.0112
 High school graduate -0.0316 0.0091
Economic conditions
 Current unemployment rate 0.0051 0.0026
 Entry unemployment rate -0.0020 0.0022
 Military/civil pay ratio 0.1188 0.0246
 CG-HSG differential 0.0265 0.0443
State socioeconomic characteristics
 Percent college graduates 0.0041 0.0004
 Percent veterans 0.0000 0.0004
 Family income ($1K) 0.0005 0.0002
Enlistment term (omitted = 4-yr term)
 Six-yr term -0.0603 0.0078
Military occupation group (admin omitted)
 Combat arms -0.0057 0.0150
 Electronic repair -0.0101 0.0062
 Communications -0.0089 0.0088
 Medical 0.0642 0.0100
 Other technical -0.0216 0.0097
 Mechanical equipment -0.0296 0.0047
 Craftsmen -0.0658 0.0067
 Service and supply -0.0116 0.0087
Separation year (1992 omitted)
 1993 -0.0235 0.0102
 1994 -0.0266 0.0137
 1995 -0.0527 0.0066
 1996 -0.0597 0.0097
 1997 -0.0660 0.0104
 1998 -0.0594 0.0113
 1999 -0.0438 0.0180
 2000 -0.0072 0.0179
 2001 0.0112 0.0260
 2002 -0.0245 0.0209
 2003 0.0437 0.0232
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho
 Observations 95,387
 Retained
 Separated

 Bivariate Probit

 ME SE (ME)
Education benefit ($1K)
 Expected value at entry -0.00127 0.00282
 Unexpected change 0.00486 0.00321
 AFQT score 0.00093 0.00104
 EVE x AFQT 0.00007 0.00004
 UC x AFQT -0.00003 0.00004
Race-ethnic group (white omitted)
 Black -0.0077 0.0052
 Hispanic 0.0215 0.0062
 Other 0.0228 0.0067
Personal characteristics
 Entry age 20-22 -0.2627 0.0122
 Entry age 23-25 -0.2248 0.0117
 Entry age 26 plus -0.0709 0.0089
 Male -0.0278 0.0076
 Married at ETS -0.0659 0.0041
 No. of dependents at ETS -0.0440 0.0028
 Some college or better -0.0827 0.0100
 High school graduate -0.0285 0.0078
Economic conditions
 Current unemployment rate 0.0037 0.0022
 Entry unemployment rate -0.0013 0.0019
 Military/civil pay ratio 0.0925 0.0212
 CG-HSG differential 0.0197 0.0377
State socioeconomic characteristics
 Percent college graduates 0.0033 0.0004
 Percent veterans -0.0001 0.0003
 Family income ($1K) 0.0005 0.0002
Enlistment term (omitted = 4-yr term)
 Six-yr term -0.0507 0.0062
Military occupation group (admin omitted)
 Combat arms 0.0049 0.0138
 Electronic repair -0.0053 0.0058
 Communications -0.0042 0.0076
 Medical 0.0602 0.0085
 Other technical -0.0092 0.0092
 Mechanical equipment -0.0230 0.0044
 Craftsmen -0.0472 0.0068
 Service and supply -0.0037 0.0080
Separation year (1992 omitted)
 1993 -0.0423 0.0102
 1994 -0.0477 0.0115
 1995 -0.0678 0.0067
 1996 -0.0753 0.0080
 1997 -0.0774 0.0079
 1998 -0.0715 0.0078
 1999 -0.0597 0.0143
 2000 -0.0273 0.0147
 2001 -0.0279 0.0221
 2002 -0.0548 0.0166
 2003 -0.0016 0.0221
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho 0.1608 0.047
 Observations 215,562
 Retained 120,175
 Separated 95,387

 Separation

 Bivariate Probit

 ME SE (ME)
Education benefit ($1K)
 Expected value at entry 0.00550 0.00393
 Unexpected change -0.00359 0.00326
 AFQT score 0.00042 0.00136
 EVE x AFQT 0.00002 0.00005
 UC x AFQT 0.00022 0.00004
Race-ethnic group (white omitted)
 Black -0.1237 0.0094
 Hispanic 0.0194 0.0066
 Other -0.0337 0.0059
Personal characteristics
 Entry age 20-22 -0.1826 0.0574
 Entry age 23-25 -0.1843 0.0626
 Entry age 26 plus -0.0181 0.0160
 Male -0.0099 0.0041
 Married at ETS -0.0945 0.0136
 No. of dependents at ETS -0.0353 0.0086
 Some college or better 0.0801 0.0134

 High school graduate -0.0244 0.0088
Economic conditions
 Current unemployment rate -0.0097 0.0019
 Entry unemployment rate 0.0056 0.0015
 Military/civil pay ratio -0.0734 0.0228
 CG-HSG differential -0.0082 0.0353
State socioeconomic characteristics
 Percent college graduates -0.0004 0.0003
 Percent veterans -0.0013 0.0004
 Family income ($1K) 0.0013 0.0002
Enlistment term (omitted = 4-yr term)
 Six-yr term 0.0941 0.0195
Military occupation group (admin omitted)
 Combat arms 0.1515 0.0114
 Electronic repair 0.0531 0.0064
 Communications 0.0585 0.0072
 Medical 0.0790 0.0101
 Other technical 0.1410 0.0064
 Mechanical equipment 0.0273 0.0052
 Craftsmen 0.1168 0.0143
 Service and supply 0.0909 0.0109
Separation year (1992 omitted)
 1993 -0.4078 0.0087
 1994 -0.4099 0.0075
 1995 -0.4038 0.0046
 1996 -0.4247 0.0065
 1997 -0.4071 0.0071
 1998 -0.3956 0.0068
 1999 -0.4086 0.0089
 2000 -0.3747 0.0101
 2001 -0.4478 0.0076
 2002 -0.4439 0.0077
 2003 -0.4297 0.0077
Rank at ETS (E4 omitted)
 Less than E4 0.2418 0.0194
 E5 -0.4154 0.0132
 More than E5 -0.3599 0.0134
ACOL ($1K) -0.0016 0.0013
 Rho
 Observations
 Retained
 Separated

EVE = expected value at entry; CG-HSC = college-graduate
<en> high-school graduate differential.

TABLE B4
Marginal Effects (ME), 2-Yr Use and Separation: Marine Corps

 2-Yr Use

 Simple Probit

 ME SE(ME)

Education benefit ($1K)
 Expected value at entry 0.00626 0.00163
 Unexpected change 0.00902 0.00241
 AFQT score 0.00433 0.00066
 EVE x AFQT 0.00000 0.00002
 UC x AFQT -0.00008 0.00003
Race-ethnic group (white omitted)
 Black -0.0023 0.0071
 Hispanic 0.0330 0.0053
 Other 0.0281 0.0068
Personal characteristics
 Entry age 20-22 -0.0113 0.0026
 Entry age 23-25 -0.0543 0.0066
 Entry age 26 plus -0.0741 0.0140
 Male -0.0199 0.0099
 Married at ETS -0.0729 0.0045
 No. of dependents at ETS -0.0349 0.0024
 Some college or better -0.0874 0.0185
 High school graduate 0.0213 0.0071
Economic conditions
 Current unemployment rate 0.0084 0.0025
 Entry unemployment rate 0.0008 0.0021
 Military/civil pay ratio 0.2353 0.0220
 CG-HSG differential 0.0284 0.0423
State socioeconomic characteristics
 Percent college graduates 0.0069 0.0005
 Percent veterans -0.0004 0.0005
 Family income ($1K) 0.0010 0.0002
Enlistment term (omitted = 4-yr term)
 Six-yr term -0.0197 0.0103
Military occupation group (admin omitted)
 Combat arms -0.0016 0.0043
 Electronic repair -0.0400 0.0066
 Communications -0.0137 0.0045
 Medical -0.0604 0.0905
 Other technical -0.0062 0.0081
 Mechanical equipment -0.0590 0.0052
 Craftsmen -0.0830 0.0074
 Service and supply -0.0500 0.0042
Separation year (1992 omitted)
 1993 -0.0229 0.0212
 1994 -0.0438 0.0107
 1995 -0.0503 0.0149
 1996 -0.0636 0.0177
 1997 -0.0938 0.0171
 1998 -0.0908 0.0176
 1999 -0.0669 0.0181
 2000 -0.0437 0.0122
 2001 -0.0049 0.0131
 2002 0.0113 0.0093
 2003 0.0611 0.0145
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho
 Observations 146,539
 Retained
 Separated

 Bivariate Probit

 ME SE(ME)

Education benefit ($1K)
 Expected value at entry 0.00672 0.00164
 Unexpected change 0.00982 0.00230
 AFQT score 0.00425 0.00064
 EVE x AFQT -0.00001 0.00002
 UC x AFQT -0.00009 0.00003
Race-ethnic group (white omitted)
 Black -0.0107 0.0067
 Hispanic 0.0305 0.0050
 Other 0.0238 0.0064
Personal characteristics
 Entry age 20-22 -0.0094 0.0025
 Entry age 23-25 -0.0513 0.0066
 Entry age 26 plus -0.0739 0.0134
 Male -0.0181 0.0100
 Married at ETS -0.0791 0.0060
 No. of dependents at ETS -0.0355 0.0034
 Some college or better -0.0789 0.0168
 High school graduate 0.0181 0.0067
Economic conditions
 Current unemployment rate 0.0075 0.0024
 Entry unemployment rate 0.0009 0.0020
 Military/civil pay ratio 0.2160 0.0216
 CG-HSG differential 0.0355 0.0390
State socioeconomic characteristics
 Percent college graduates 0.0066 0.0005
 Percent veterans -0.0004 0.0004
 Family income ($1K) 0.0010 0.0002
Enlistment term (omitted = 4-yr term)
 Six-yr term -0.0155 0.0103
Military occupation group (admin omitted)
 Combat arms 0.0088 0.0040
 Electronic repair -0.0741 0.0076
 Communications -0.0134 0.0044
 Medical -0.0916 0.0801
 Other technical -0.0029 0.0080
 Mechanical equipment -0.0653 0.0051
 Craftsmen -0.0734 0.0072
 Service and supply -0.0418 0.0042
Separation year (1992 omitted)
 1993 -0.0347 0.0106
 1994 -0.0435 0.0071
 1995 -0.0573 0.0075
 1996 -0.0848 0.0094
 1997 -0.0814 0.0097
 1998 -0.0597 0.0107
 1999 -0.0405 0.0149
 2000 -0.0033 0.0169
 2001 -0.0205 0.0210
 2002 0.0429 0.0162
 2003 -0.0159 0.0207
Rank at ETS (E4 omitted)
 Less than E4
 E5
 More than E5
ACOL ($1K)
 Rho 0.2650 0.0185
 Observations 204,302
 Retained 57,763
 Separated 146,539

 Separation

 Bivariate Probit

 ME SE(ME)

Education benefit ($1K)
 Expected value at entry 0.00283 0.00259
 Unexpected change 0.00370 0.00398
 AFQT score 0.00016 0.00093
 EVE x AFQT -0.00004 0.00004
 UC x AFQT -0.00003 0.00005
Race-ethnic group (white omitted)
 Black -0.1044 0.0113
 Hispanic 0.0017 0.0048
 Other -0.0302 0.0058
Personal characteristics
 Entry age 20-22 0.0354 0.0035
 Entry age 23-25 0.0408 0.0080
 Entry age 26 plus 0.0166 0.0109
 Male -0.0062 0.0105
 Married at ETS -0.0855 0.0252
 No. of dependents at ETS -0.0213 0.0127
 Some college or better 0.1133 0.0099
 High school graduate 0.0018 0.0064
Economic conditions
 Current unemployment rate -0.0074 0.0015
 Entry unemployment rate 0.0042 0.0013
 Military/civil pay ratio -0.0914 0.0201
 CG-HSG differential 0.0704 0.0336
State socioeconomic characteristics
 Percent college graduates 0.0006 0.0004
 Percent veterans 0.0002 0.0004
 Family income ($1K) 0.0007 0.0001
Enlistment term (omitted = 4-yr term)
 Six-yr term 0.2019 0.0053
Military occupation group (admin omitted)
 Combat arms 0.1231 0.0051
 Electronic repair -0.3069 0.0285
 Communications 0.0234 0.0114
 Medical -0.3547 0.0737
 Other technical 0.0727 0.0088
 Mechanical equipment -0.0578 0.0172
 Craftsmen 0.0594 0.0054
 Service and supply 0.0476 0.0040
Separation year (1992 omitted)
 1993 -0.2062 0.0415
 1994 -0.2279 0.0125
 1995 -0.2315 0.0133
 1996 -0.2125 0.0246
 1997 -0.1787 0.0202
 1998 -0.1699 0.0227
 1999 -0.1469 0.0295
 2000 -0.1219 0.0341
 2001 -0.4434 0.0403
 2002 -0.2860 0.0400
 2003 -0.2998 0.0476
Rank at ETS (E4 omitted)
 Less than E4 0.2211 0.0073
 E5 -0.4844 0.0175
 More than E5 -0.7694 0.0059
ACOL ($1K) 0.0000 0.0000
 Rho
 Observations
 Retained
 Separated

EVE = expected value at entry; CG-HSC = college-graduate
<en> high-school graduate differential.


APPENDIX C
TABLE C1
Average Coefficient and Standard Deviation in 100 Bootstraps of
Hazard Model of MGIB Usage

 Army

 Coefficient SD

Education benefit ($1K)
 Expected value at entry 0.0209 0.0018
 Unexpected change 0.0415 0.0039
 AFQT x EVE 0.0000 0.0000
 AFQT x unexpected change -0.0004 0.0001
 AFQT 0.0115 0.0010
 Entry bonus amount ($1K) -0.0069 0.0013
Race-ethnic group (white omitted)
 Black 0.1496 0.0157
 Hispanic 0.1660 0.0149
 Other 0.2315 0.0132
Personal characteristics
 Entry age 20-22 -0.0632 0.0081
 Entry age 23-25 -0.2233 0.0161
 Entry age 26 plus -0.3247 0.0220
 Male -0.2896 0.0199
 Married at separation -0.1171 0.0096
 No. of dependents at separation -0.0586 0.0062
 Some college or better -0.3070 0.0393
 High school graduate 0.1177 0.0158
Economic conditions
 Current unemployment rate 0.0587 0.0055
 Entry unemployment rate -0.0172 0.0046
 Military/civil pay ratio 0.6543 0.0700
 CG-HSG differential 0.2083 0.1101
State socioeconomic characteristics
 Percent college graduates 0.0143 0.0011
 Percent veterans 0.0007 0.0010
 Family income ($1K) 0.0025 0.0003
Military occupation group (admin omitted)
 Combat arms 0.1945 0.0709
 Electronic repair 0.1181 0.0651
 Communications 0.1788 0.0643
 Medical 0.3025 0.0565
 Other technical 0.2042 0.0657
 Mechanical equipment 0.0139 0.0682
 Craftsmen -0.0751 0.0747
 Service and supply 0.1049 0.0667
Length of initial term (4-YO omitted)
 2-YO 0.1882 0.0372
 3-YO 0.1640 0.0298
 5-YO 0.0186 0.0228
 6-YO -0.0708 0.0206
Number of completed years in service
 FT Attriter -1.4063 0.0589
 3 -0.2489 0.0337
 4 -0.1005 0.0417
 5 -0.2003 0.0458
 6 -0.2305 0.0484
 7 -0.2869 0.0491
 8 -0.3626 0.0528
 9 -0.4039 0.0599
 10 -0.4691 0.0549
 11 -0.5255 0.0724
 12+ -0.6691 0.1015
Separation FY
 1992 -0.0591 0.0657
 1993 -0.0967 0.0579
 1994 -0.0728 0.0451
 1995 -0.0732 0.0604
 1996 -0.1000 0.0475
 1997 -0.0846 0.0478
 1998 -0.0895 0.0434
 1999 -0.0563 0.0524
 2000 -0.0454 0.0613
 2001 0.0608 0.0617
 2002 0.0946 0.0587
 2003 0.0562 0.0593
 2004 -0.0214 0.0583
Years since separation
 2 -0.2522 0.0151
 3 -0.6959 0.0243
 4 -0.9718 0.0229
 5 -1.1776 0.0290
 6 -1.3475 0.0396
 7 -1.4793 0.0363
 8 -1.5945 0.0431
 9 -1.8291 0.0426
 10 -2.1742 0.0549
 Intercept -4.4471 0.2303

 Navy

 Coefficient SD

Education benefit ($1K)
 Expected value at entry 0.0112 0.0054
 Unexpected change 0.0285 0.0056
 AFQT x EVE -0.0001 0.0001
 AFQT x unexpected change -0.0003 0.0001
 AFQT 0.0139 0.0016
 Entry bonus amount ($1K) -0.0016 0.0033
Race-ethnic group (white omitted)
 Black 0.1160 0.0131
 Hispanic 0.1268 0.0150
 Other 0.1429 0.0194
Personal characteristics
 Entry age 20-22 -0.0670 0.0080
 Entry age 23-25 -0.2477 0.0175
 Entry age 26 plus -0.3689 0.0222
 Male -0.1819 0.0150
 Married at separation -0.1594 0.0096
 No. of dependents at separation -0.0941 0.0061
 Some college or better -0.0832 0.0348
 High school graduate 0.1539 0.0182
Economic conditions
 Current unemployment rate 0.0607 0.0068
 Entry unemployment rate -0.0050 0.0033
 Military/civil pay ratio 0.6271 0.0579
 CG-HSG differential 0.3722 0.1155
State socioeconomic characteristics
 Percent college graduates 0.0184 0.0010
 Percent veterans 0.0025 0.0010
 Family income ($1K) 0.0212 0.0036
Military occupation group (admin omitted)
 Combat arms 0.0576 0.0239
 Electronic repair -0.0283 0.0174
 Communications 0.1086 0.0144
 Medical 0.2880 0.0184
 Other technical 0.2113 0.0310
 Mechanical equipment -0.0407 0.0147
 Craftsmen -0.2238 0.0242
 Service and supply -0.0362 0.0233
Length of initial term (4-YO omitted)
 2-YO -0.0027 0.0455
 3-YO 0.1710 0.0345
 5-YO -0.1130 0.0196
 6-YO -0.1436 0.0207
Number of completed years in service
 FT Attriter -0.9643 0.0752
 3 -0.1659 0.0655
 4 0.0252 0.0585
 5 -0.0180 0.0614
 6 -0.0616 0.0604
 7 -0.2399 0.0638
 8 -0.3293 0.0683
 9 -0.3767 0.0778
 10 -0.4333 0.0748
 11 -0.5828 0.0895
 12+ -0.8584 0.1346
Separation FY
 1992 0.0301 0.0522
 1993 0.0781 0.0595
 1994 0.1261 0.0542
 1995 0.1361 0.0484
 1996 0.0633 0.0495
 1997 0.1263 0.0540
 1998 0.0463 0.0753
 1999 0.0854 0.0754
 2000 0.1392 0.0621
 2001 0.2083 0.0698
 2002 0.3000 0.0775
 2003 0.4021 0.0757
 2004 0.2967 0.0884
Years since separation
 2 -0.2318 0.0150
 3 -0.6909 0.0216
 4 -0.9702 0.0239
 5 -1.1654 0.0298
 6 -1.3023 0.0272
 7 -1.4396 0.0337
 8 -1.5554 0.0314
 9 -1.8032 0.0541
 10 -2.0634 0.0483
 Intercept -4.9439 0.2029

 Air Force

 Coefficient SD

Education benefit ($1K)
 Expected value at entry 0.0555 0.0093
 Unexpected change 0.0404 0.0064
 AFQT x EVE -0.0005 0.0001
 AFQT x unexpected change -0.0003 0.0001
 AFQT 0.0246 0.0032
 Entry bonus amount ($1K)
Race-ethnic group (white omitted)
 Black 0.1050 0.0180
 Hispanic 0.0691 0.0194
 Other 0.1192 0.0215
Personal characteristics
 Entry age 20-22 -0.1136 0.0110
 Entry age 23-25 -0.3412 0.0166
 Entry age 26 plus -0.4670 0.0287
 Male -0.2706 0.0164
 Married at separation -0.1896 0.0109
 No. of dependents at separation -0.1396 0.0061
 Some college or better -0.3524 0.0445
 High school graduate -0.0358 0.0264
Economic conditions
 Current unemployment rate 0.0602 0.0066
 Entry unemployment rate -0.0108 0.0039
 Military/civil pay ratio 0.5651 0.0590
 CG-HSG differential 0.2450 0.0926
State socioeconomic characteristics
 Percent college graduates 0.0151 0.0010
 Percent veterans 0.0033 0.0009
 Family income ($1K) 0.0020 0.0004
Military occupation group (admin omitted)
 Combat arms 0.0100 0.0220
 Electronic repair 0.0073 0.0154
 Communications 0.0121 0.0154
 Medical 0.1860 0.0120
 Other technical -0.0159 0.0166
 Mechanical equipment -0.0857 0.0126
 Craftsmen -0.2568 0.0266
 Service and supply -0.0303 0.0210
Length of initial term (4-YO omitted)
 2-YO
 3-YO
 5-YO
 6-YO -0.2100 0.0319
Number of completed years in service
 FT Attriter -0.3156 0.1024
 3 -0.2101 0.0315
 4 0.8230 0.0969
 5 0.9886 0.0890
 6 0.9392 0.0928
 7 0.8770 0.0912
 8 0.6583 0.0880
 9 0.6587 0.0930
 10 0.5158 0.0939
 11 0.4857 0.0924
 12+ 0.3903 0.0864
Separation FY
 1992 0.0806 0.1995
 1993 0.1427 0.1437
 1994 0.1407 0.1465
 1995 0.1384 0.1522
 1996 0.0731 0.1528
 1997 0.1228 0.1503
 1998 0.1307 0.1545
 1999 0.1676 0.1562
 2000 0.1811 0.1559
 2001 0.2743 0.1559
 2002 0.2199 0.1637
 2003 0.3868 0.1672
 2004 0.2499 0.1695
Years since separation
 2 -0.3622 0.0264
 3 -0.7383 0.0324
 4 -1.0102 0.0387
 5 -1.2500 0.0415
 6 -1.4324 0.0474
 7 -1.5186 0.0504
 8 -1.7523 0.0467
 9 -2.0622 0.0423
 10 -2.5365 0.0803
 Intercept -6.3097 0.2892

 Marine Corps

 Coefficient SD

Education benefit ($1K)
 Expected value at entry 0.0320 0.0044
 Unexpected change 0.0393 0.0047
 AFQT x EVE -0.0003 0.0001
 AFQT x unexpected change -0.0004 0.0001
 AFQT 0.0204 0.0015
 Entry bonus amount ($1K)
Race-ethnic group (white omitted)
 Black 0.1055 0.0189
 Hispanic 0.0997 0.0189
 Other 0.0777 0.0213
Personal characteristics
 Entry age 20-22 -0.0745 0.0121
 Entry age 23-25 -0.2443 0.0279
 Entry age 26 plus -0.2918 0.0449
 Male -0.2192 0.0207
 Married at separation -0.1733 0.0144
 No. of dependents at separation -0.0904 0.0075
 Some college or better -0.4004 0.0580
 High school graduate 0.0586 0.0213
Economic conditions
 Current unemployment rate 0.0768 0.0079
 Entry unemployment rate -0.0156 0.0053
 Military/civil pay ratio 0.8669 0.0789
 CG-HSG differential 0.4273 0.1467
State socioeconomic characteristics
 Percent college graduates 0.0197 0.0016
 Percent veterans 0.0012 0.0016
 Family income ($1K) 0.0020 0.0004
Military occupation group (admin omitted)
 Combat arms -0.0140 0.0107
 Electronic repair -0.1036 0.0338
 Communications -0.0605 0.0177
 Medical 0.1342 0.1221
 Other technical -0.0603 0.0298
 Mechanical equipment -0.1987 0.0194
 Craftsmen -0.2755 0.0263
 Service and supply -0.1698 0.0128
Length of initial term (4-YO omitted)
 2-YO
 3-YO
 5-YO
 6-YO -0.1025 0.0336
Number of completed years in service
 FT Attriter -0.0867 0.1407
 3 0.3317 0.1324
 4 0.7773 0.1236
 5 0.7483 0.1253
 6 0.7415 0.1285
 7 0.6010 0.1331
 8 0.5412 0.1263
 9 0.4262 0.1402
 10 0.2437 0.1419
 11 0.1101 0.2150
 12+
Separation FY
 1992 -0.1347 0.0547
 1993 -0.1280 0.0630
 1994 -0.0878 0.0579
 1995 -0.1940 0.0671
 1996 -0.2079 0.0555
 1997 -0.0818 0.0604
 1998 -0.1165 0.0572
 1999 -0.0832 0.0641
 2000 -0.0655 0.0699
 2001 -0.0285 0.0834
 2002 0.0314 0.0746
 2003 0.1057 0.0906
 2004 -0.0887 0.0941
Years since separation
 2 -0.3451 0.0333
 3 -0.7966 0.0369
 4 -1.0859 0.0334
 5 -1.2938 0.0321
 6 -1.4684 0.0339
 7 -1.6141 0.0563
 8 -1.7655 0.0571
 9 -1.9279 0.0689
 10 -2.2965 0.0817
 Intercept -6.2356 0.2999

Note: For the Army and Navy, the completed year of service
coefficients are relative to a veteran with two completed years.
For the Air Force and Marine Corps, coefficients are relative to
a veteran with 12 or more completed years. EVE = expected value
at entry; CG-HSC = college-graduate <en> high-school graduate
differential.


REFERENCES

Angrist, J. "The Effect of Veterans Benefits on Education and Earnings." Industrial and Labor Relations Review, 46, 1993, 637-52.

Asch, B., J. Hosek, D. and Clendenning. A Policy Analysis of Reserve Retirement Reform. Unpublished Report, RAND Corporation, Santa Monica, CA, 2006.

Asch, B., J. Hosek, and J. Warner . "New Economics of Manpower in the Cold War Era," in Handbook of Defense Economics, Volume 2, edited by K. Hartley, and T. Sandler. New York Elsevier 2007, 1076-138.

Asch, B., and J. Warner. "A Theory of Compensation and Personnel Policy in Hierarchical Organizations with Application to the United States Military." Journal of Labor Economics, 19, 2001, 523-62.

Avery, C., and C. Hoxby . "Do and Should Financial Aid Packages Affect Students' College Choices?" in College Choices--The Economics of Where to Go, When to Go, and How to Pay for It, edited by C. Hoxby. Chicago University of Chicago Press, 2004, 239-302.

Bound, J., and S. Turner . "Going to War and Going to College: Did World War II and the GI Bill Increase Educational Attainment for Returning Veterans?" Journal of Labor Economics, 20, 2002, 784-815.

Dynarski, S. "The Behavioral and Distributional Implications of Aid College." American Economic Review, 92, 2002, 279-85.

Dynarski, S. "Does Aid Matter? Measuring the Effect of Student Aid on College Attendance and Completion." American Economic Review, 93, 2003, 279-88.

Gotz, G., and J. McCall . A Dynamic Retention Model of Air Force Officer Retention: Theory and Estimation, R-03028-AF. Santa Monica, CA RAND 1984.

Hilmer, M. "Post-Secondary Fees and the Decision to Attend a University or a Community College." Journal of Public Economics, 67, 1998, 329-48.

Hogan, P., D. Smith, and S. Sylwester . "The Army College Fund: Effects on Attrition, Reenlistment, and Cost," in Military Compensation and Personnel Retention-Models and Evidence, edited by C. Gilroy, D. Home, and D. Smith. United States Army Research Institute for the Behavioral and Social Sciences, 1991, 317-54.

Keane, M. P., and K. I. Wolpin . "The Career Decisions of Young Men." Journal of Political Economy, 105, 1997, 473-522.

McPherson, M., and M. Schapiro. "Does Student Aid Affect College Enrollment? New Evidence on a Persistent Controversy." American Economic Review, 81, 1991, 309-18.

Nielsen, H. S., T. Sorensen, and C. Taber. "Estimating the Effect of Student Aid on College Enrollment: Evidence from a Government Grant Policy Reform." Mimeo, 2008.

O'Neill, D. "Voucher Funding of Training Programs: Evidence from the GI Bill." Journal of Human Resources, 12, 1977, 425-45.

Principi, A. Report of the Congressional Commission on Service Members and Veterans Transition Assistance. Washington, DC Congressional Commission on Servicemembers and Veterans Transition Assistance, 1999.

Seftor, N., and S. Turner. "Back to School: Federal Student Aid Policy and Adult College Enrollment." Journal of Human Resources, 37, 2002, 336-52.

Smith, D., S. Sylwester, and C. Villa. "Army Reenlistment Models," in Military Compensation and Personnel Retention--Models and Evidence, edited by C. Gilroy, D. Home, and D. Smith. United States Army Research Institute for the Behavioral and Social Sciences, 1991, 43-173.

Stanley, M . "College Education and the Midcentury GI Bills." Quarterly Journal of Economics, 118, 2003, 671-708.

Van der Klaauw, W. "Estimating the Effect of Financial Aid Offers on College Enrollment: A Regression-Discontinuity Approach." International Economic Review, 43, 2002, 1249-87.

Warner, J., and G. Solon. "First-Term Attrition and Reenlistment in the U.S. Army," in Military Compensation and Personnel Retention--Models and Evidence, edited by C. Gilroy, D. Home, and D. Smith. Washington, DC United States Army Research Institute for the Behavioral and Social Sciences, 1991, 243-77.

(1.) Early implementations of the GI Bill, first introduced after World War II, did not require a contribution. The $1,200 contribution was first required in the Veteran's Educational Assistance Program, a less-generous program that replaced the Vietnam-era GI Bill that ended in December 1976.

(2.) See Asch, Hosek, and Warner (2007) for a comprehensive review of past studies of the effects of educational benefits on high-quality enlistment. The average elasticity estimate is 0.1 (table 1, p. 1083), suggesting that a doubling of educational benefits would expand high-quality enlistment by about 10 percent.

(3.) Van der Klaauw (2002) studied the effects of college aid on enrollment in a single college. His estimates are therefore not comparable to those cited in the text.

(4.) O'Neill (1977), Bound and Turner (2002), and Stanley (2003) estimated substantial effects of GI Bill benefits on educational outcomes of World War II veterans.

(5.) The dynamic programming approach to military retention was pioneered by Gotz and McCall (1984). Recent empirical implementations of the dynamic programming approach are rooted in Keane and Wolpin (1997) and include Asch and Warner (2001) and Asch, Hosek, and Clendenning (2006).

(6.) The recommendations of the Principi Commission have, however, been implemented in the Post-9/11 Veterans Education Assistance Act of 2008. Beginning August 1, 2009, the $1,200 recruit contribution is eliminated and each veteran will receive (1) tuition less than or equal to that charged by the most expensive public university in each veteran's state of residence, (2) a monthly cost-of-living stipend averaging $1,200 per month, and (3) $1,000 annually for books and supplies. The act also eliminates the $1,200 recruit contribution. For more details, visit http://www.GIBILL.va.gov.

(7.) For the purposes of this study, retention is defined as being on active duty at least 1 yr beyond the individual's initial scheduled ETS date.

(8.) Avery and Hoxby (2004) find evidence of "ignorance" or "naivete" among college students making complex financing decisions, suggesting that a static expectations scenario may in fact be the most plausible one. Ultimately, differences in assumptions about how entrants form expectations had little qualitative or quantitative impact.

(9.) Our forecast was based on a regression of annual college tuition CPI on its own three lags using data before 1988 ([R.sup.2] = 0.75). The equation predicts college tuition inflation of around 7% after the third or fourth forecast period, which was in fact the average over our study period.

(10.) See Asch, Hosek, and Warner (2007) for a derivation of this measure and for a review of retention studies that have used it. In the application here, ACOL is calculated for each first-term retention decision maker based on a time horizon that begins with the first-term decision point and ends with the 20-yr career point (and so includes the annualized value of future military retirement pay). It varies over time in our data due to differences in annual military and civilian earnings growth. Civilian earnings were estimated by age for each individual using data from the Current Population Surveys and were permitted to vary by education level, race, gender, and first-term decision year in addition to age.

(11.) The percentage difference is equal to the absolute difference divided by the average of the two coefficients.

(12.) We averaged the simple and bivariate probit standard errors.

(13.) The estimated effects of the control variables (see Appendix C) were generally in line with the 2-yr bivariate and simple probit estimates.

(14.) As discussed in Section II, estimates from earlier studies suggest a retention decline on the order of 6-8%.

CURTIS J. SIMON [1], SEBASTIAN NEGRUSA [2] and JOHN T. WARNER *

* This research was sponsored by the Accession Policy Directorate of the Office of the Undersecretary of Defense for Personnel and Readiness under contract number HQ0034-06-C-1026. We thank the Director of Accession Policy, Curtis Gilroy, for his support and encouragement of our work. We also thank Heidi Golding and other participants at the 2006 Western Economic Association meetings for comments on an early version of this paper and Alex Gelbet, Martin Feldstein, and other participants at the summer 2008 NBER Program on the Economics of National Security for comments on a more recent version. The views and opinions expressed herein are those of the authors and do not represent any official policy position of the Department of Defense. They alone are responsible for any remaining errors.

SIMON: Associate Professor, Department of Economics, Clemson University, Clemson, SC 29634-1309. Phone 864-656-3966, Fax 864-656-4192, E-mail cjsmn@clemson.edu Clemson University.

NEGRUSA: Associate Economist, RAND Corporation. Phone 703-413-1100, Fax 703-414-4725, E-mail snegrusa@rand.org RAND Corporation.

WARNER: Professor, Department of Economics, 222 Sirrine Hall, Clemson University, Clemson, SC 29634-1309. Phone 864-656-3967, Fax 864-656-4192, E-mail jtwarne@clemson.edu
TABLE 1
Benefit Use Among Eligible Veterans, by YOS
Group
 Army Navy AF MC
All YOS
Number 385,421 303,392 144,131 197,110
MGIB users 190,779 154,915 72,531 94,764
Percent use 49 51 50 48
[less than or equal to] 6 YOS
Number 344,903 268,241 118,327 184,883
MGIB users 175,648 141,790 62,397 90,556
Percent use 51 53 53 49
> 6 YOS
Number 40,518 35,151 25,804 12,227
MGIB users 15,131 13,125 10,134 4,208
Percent use 37 37 39 34

TABLE 2
Estimates of Marginal Effects of Education Benefits and AFQT on
Separation and 2-Yr MGIB Usage

 Simple Probit
 Marginal Standard Marginal
 Effect at Error at Effect at
 Mean Mean Mean
 AFQT AFQT AFQT

Army

$10K EVE 0.0709 0.0035 0.0718

$10K benefit shock 0.0830 0.0114 0.0850

10-point AFQT increase 0.0347 0.0012 0.0350

$10K EVE 0.0218

$10K benefit shock 0.0351

10-point AFQT increase 0.0190

Navy

$10K EVE 0.0238 0.0116 0.0202

$10K benefit shock -0.0213 0.0187 -0.0152

10-point AFQT increase 0.0317 0.0015 0.0309

$10K EVE -0.0329

$10K benefit shock 0.0381

10-point AFQT increase 0.0230

Air Force

$10K EVE 0.0296 0.0115 0.0337

$10K benefit shock 0.0247 0.0153 0.0265

10-point AFQT increase 0.0284 0.0023 0.0255

$10K EVE 0.0716

$10K benefit shock 0.1095

10-point AFQT increase 0.0175

Marine Corps

$10K EVE 0.0592 0.0045 0.0595

$10K benefit shock 0.0411 0.0164 0.0427

10-point AFQT increase 0.0388 0.0009 0.0362

$10K entry value 0.0057

$10K benefit shock 0.0204

10-point AFQT increase -0.0089

 Bivariate Probit
 Standard
 Error at Marginal Standard
 Mean Effect at Error at
 AFQT AFQT = 80 AFQT = 80

Army 2-yr use

$10K EVE 0.0037 0.0895 0.0038

$10K benefit shock 0.0122 0.0732 0.0142

10-point AFQT increase 0.0012 0.0350 0.0012
 Separation

$10K EVE 0.0030 0.0335 0.0033

$10K benefit shock 0.0098 0.0521 0.0101

10-point AFQT increase 0.0011 0.0190 0.0011

Navy 2-yr use

$10K EVE 0.0102 0.0217 0.0077

$10K benefit shock 0.0178 -0.0425 0.0218

10-point AFQT increase 0.0015 0.0309 0.0015

 Separation

$10K EVE 0.0195 -0.0119 0.0170

$10K benefit shock 0.0470 0.0528 0.0478

10-point AFQT increase 0.0021 0.0230 0.0021

Air Force 2-yr use

$10K EVE 0.0116 0.0432 0.0133

$10K benefit shock 0.0158 0.0220 0.0151

10-point AFQT increase 0.0014 0.0255 0.0014

 Separation

$10K EVE 0.0138 0.0750 0.0149

$10K benefit shock 0.0261 0.1393 0.0280

10-point AFQT increase 0.0014 0.0175 0.0014

Marine Corps 2-yr use

$10K EVE 0.0048 0.0568 0.0059

$10K benefit shock 0.0165 0.0236 0.0180

10-point AFQT increase 0.0009 0.0362 0.0009

 Separation

$10K entry value 0.0070 -0.0021 0.0068

$10K benefit shock 0.0218 0.0147 0.0227

10-point AFQT increase 0.0015 -0.0089 0.0015

EVE = expected value at entry

TABLE 3
Estimated effects of Benefit Changes and AFQT Changes on Veterans'
MGIB Usage

 2-Yr Use
 Lower Upper
 Mean 95% CI 95% CI

Army

Base case 0.320 0.311 0.328
$10K EVE 0.375 0.360 0.390
$10K shock 0.375 0.361 0.389
10-point AFQT increase 0.377 0.364 0.390

Navy

Base case 0.312 0.305 0.319
$10K EVE 0.330 0.271 0.389
$10K shock 0.348 0.328 0.369
10-point AFQT increase 0.372 0.360 0.384

Air Force

Base Case 0.296 0.289 0.304
$10K EVE 0.348 0.269 0.427
$10K shock 0.343 0.319 0.366
10-point AFQT increase 0.347 0.326 0.368

Marine Corps

Base case 0.311 0.304 0.318
$10K EVE 0.358 0.320 0.397
$10K shock 0.350 0.325 0.375
10-point AFQT increase 0.378 0.364 0.391

 5-Yr Use
 Lower Upper
 Mean 95% CI 95% CI

Army

Base case 0.474 0.464 0.484
$10K EVE 0.543 0.524 0.561
$10K shock 0.543 0.526 0.560
10-point AFQT increase 0.545 0.529 0.561

Navy

Base case 0.464 0.454 0.474
$10K EVE 0.486 0.411 0.561
$10K shock 0.510 0.484 0.535
10-point AFQT increase 0.539 0.525 0.554

Air Force

Base Case 0.443 0.434 0.452
$10K EVE 0.508 0.408 0.608
$10K shock 0.503 0.473 0.532
10-point AFQT increase 0.508 0.482 0.534

Marine Corps

Base case 0.454 0.443 0.464
$10K EVE 0.513 0.465 0.560
$10K shock 0.503 0.472 0.533
10-point AFQT increase 0.537 0.520 0.553

 10-Yr Use
 Lower Upper
 Mean 95% CI 95% CI

Army

Base case 0.573 0.560 0.586
$10K EVE 0.645 0.626 0.665
$10K shock 0.646 0.629 0.663
10-point AFQT increase 0.648 0.632 0.664

Navy

Base case 0.566 0.554 0.578
$10K EVE 0.589 0.510 0.668
$10K shock 0.615 0.588 0.642
10-point AFQT increase 0.646 0.630 0.661

Air Force

Base Case 0.532 0.520 0.543
$10K EVE 0.601 0.495 0.706
$10K shock 0.596 0.565 0.627
10-point AFQT increase 0.601 0.574 0.629

Marine Corps

Base case 0.546 0.534 0.558
$10K EVE 0.609 0.559 0.659
$10K shock 0.598 0.566 0.631
10-point AFQT increase 0.634 0.617 0.651

Notes: Predicted usage rates and confidence intervals (CI) are
calculated from 100 bootstraps of hazard model for usage. Average
values of parameter estimates and their standard errors are reported
in Table C1. EVE = expected value at entry.

TABLE 4
Percentage Distribution of Months of Usage, Mean Months, and
Percentage Still Enrolled in June 2005

Separation Total 0-6 7-12 13-18 19-24 25-30 31-36
Period

All veterans

FY 1991-1994 87,858 22.6 17.3 16.9 16.5 12.2 14.6
FY 1995-1999 127,125 23.7 18.3 18.5 18.3 12.3 8.9
FY 2000-1904 79,475 33.8 26.4 17.8 12.9 6.5 2.6
Total 294,458 26.1 20.2 17.8 16.3 10.7 8.9

Veterans eligible for service CF benefits

FY 1991-1994 13,871 12.1 12.9 15.3 19.1 16.7 23.9
FY 1995-1999 23,990 17.1 15.9 18.6 21.1 15.1 12.2
FY 2000-2004 11,550 25.3 25.6 18.8 16.5 9.5 4.4
Total 49,411 17.6 17.3 17.8 19.5 14.3 13.6

Veterans eligible for MGIB only

FY 1991-1994 73,947 24.6 18.1 17.1 16 11.4 12.6
FY 1995-1999 103,135 25.2 18.9 18.5 17.6 11.6 8.2
FY 2000-2004 67,925 35 26.6 17.6 12.2 6 2.7
Total 245,047 27.8 20.8 17.8 15.6 10 8

Separation Mean Percent
Period Months Enrolled

All veterans

FY 1991-1994 16.9 1.7
FY 1995-1999 15.7 19.6
FY 2000-1904 11.7 53.9
Total 15 23.6

Veterans eligible for service CF benefits

FY 1991-1994 20.9 1.4
FY 1995-1999 17.8 17.9
FY 2000-2004 13.7 59.1
Total 17.7 23

Veterans eligible for MGIB only

FY 1991-1994 16.1 1.8
FY 1995-1999 15.2 20
FY 2000-2004 11.4 53
Total 14.4 23.7

TABLE 5
Cox Model Estimates of Effects of Benefit and
AFQT Changes on Amount of Educational
Benefit Usage

 $1OK $1OK 20-Point
 Base EVE Benefit AFQT
 Case Increase Shock Increase

Army 17.9 18.9 19.5 19.4
Navy 17.4 17.8 18.3 18.9
Air Force 18.3 19.6 19.7 19.3
Marine Corps 16.9 17.8 17.7 18.4

EVE = expected value at entry.
COPYRIGHT 2010 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2010 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Simon, Curtis J.; Negrusa, Sebastian; Warner, John T.
Publication:Economic Inquiry
Date:Oct 1, 2010
Words:17575
Previous Article:Skilled-unskilled wage inequality and urban unemployment.
Next Article:Workforce composition and firm productivity: evidence from Taiwan.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters