Fighting for Education: Financial Aid

and Degree Attainment

Andrew Barr, Texas A&M University

The Post-9/11 GI Bill brought about the largest expansion in vet-

eran education benefits since the end of World War II, increasing an-
nual benefit expenditures from $3 billion to more than $13 billion.
Leveraging variation over time, geography, and type of veterans, I
explore the effect of financial aid on degree attainment. I find that
the aid expansion increased degree attainment by 5–6 percentage
points (25%), roughly 0.4 percentage points per $1,000 of additional
maximum aid. These findings indicate that financial aid can increase
degree attainment, even for individuals with high levels of initial
support.

I. Introduction
Each year billions of dollars are spent on financial aid in the hope of help-
ing millions of students enroll in college and obtain a degree. While there is

I am especially grateful to Sarah Turner, Leora Friedberg, and Bill Johnson for
helpful suggestions. I also am thankful for feedback from participants in the CESifo
Economics of Education meeting and seminars at RAND, Texas A&M, Duke, the
University of Chicago, Notre Dame, the University of Illinois at Urbana-Champaign,
the University of Georgia, the University of Virginia, the Upjohn Institute, the As-
sociation for Public Policy Analysis and Management (APPAM), the Association
for Education Finance and Policy (AEFP), and the Society of Labor Economists
(SOLE). I acknowledge the National Science Foundation, which provided support
through a SBE Doctoral Dissertation Improvement Grant and through an Ameri-
can Educational Research Association dissertation fellowship. I also acknowledge
funding from a National Academy of Education/Spencer dissertation fellowship,
a Bankard fellowship, and the University of Virginia Department of Economics. I
am also grateful to Kit Tong, Peter Cerussi, Kris Hoffman, and a number of other
individuals at the Defense Manpower Data Center and the Pentagon for helping

[ Journal of Labor Economics, 2019, vol. 37, no. 2]

© 2019 by The University of Chicago. All rights reserved. 0734-306X/2019/3702-0006$10.00
Submitted December 11, 2016; Accepted December 11, 2017; Electronically published February 11, 2019

509

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 510 Barr

abundant evidence of the effects of aid on increasing college access, little ev-
idence exists on the effects of aid on degree attainment. Furthermore, the
existing evidence as to whether financial aid improves educational attain-
ment is mixed, limited to very local treatment effects or small experimental
groups, and focuses almost entirely on traditional students (Dynarski 2008;
Scott-Clayton 2011; Goldrick-Rab et al. 2012; Sjoquist and Winters 2015;
Castleman and Long 2016). This study brings new evidence to understand-
ing the effects of financial aid on degree attainment and, unlike most existing
research, how financial aid affects the attainment of older, nontraditional
students.
I leverage variation provided by the Post-9/11 GI Bill, one of the largest
financial aid expansions in decades, to identify the effects of financial aid on
the degree attainment of an important group of nontraditional students, re-
cent military veterans. Veterans are similar to other nontraditional students
along a number of demographic dimensions, including family background,
race, age, and marital status, as well as patterns of college enrollment in terms
of motivations for attending, institution type, and major.1 Each year roughly
200,000 military veterans transition out of the military. As the nation’s
largest employer of young adults and an organization that markets itself
as providing educational opportunities to youth who might not otherwise
have the means to pay for college, the military provides an interesting con-
text in which to study how financial aid affects the decision to invest in ed-
ucation during a period of transition between occupations.2 Easing this
transition was one of the stated goals of the benefit expansion.3
In addition to addressing a fundamental question about the effect of fi-
nancial aid on degree attainment, the economic magnitude of the ongoing
investment—now more than $13 billion annually—makes it worthy of in-
vestigation. To put this in perspective relative to other well-known finan-
cial aid programs, annual expenditures on the GI Bill now exceed the an-
nual amount of grants and scholarships provided by all states combined,
are roughly half the size of federal expenditures on Pell grants, and are more
than three-quarters the size of federal education tax benefits.

to provide access to data. All remaining errors are my own. Contact Andrew Barr at
barrandrewc@[Link]. Information concerning access to the data used in this pa-
per is available as supplemental material online.
1
Source: “Military Service Members and Veterans: A Profile of Those Enrolled
in Undergraduate and Graduate Education in 2007–2008” (NCES 2011-163).
2
Military veterans are substantially more likely to enroll in college during this
transition than several years after joining the civilian labor force. In some ways this
is similar to civilian workers who are substantially more likely to enroll between
occupations.
3
“The majority of the people who go into the military get out on or before the
end of their first enlistment . . . and they deserve the same opportunity to readjust
into civilian life and a first-class shot at the future” (Senator Webb, 2008, https://
[Link]/news/how-the-gi-bill-changed-america/).

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 511

One of the primary roadblocks to research in this area is the lack of data
that track enrollment and degree attainment over time. While prior work
examines the effect of the Post-9/11 GI Bill on enrollment, data limitations
made it impossible to determine the extent to which the observed increases
in enrollment were true increases in educational attainment or merely shifts
in enrollment timing (Barr 2015). As no existing data set contained a suffi-
cient number of veterans to provide even a cursory analysis of veteran de-
gree attainment, I created a new panel combining data on the choices of mil-
itary service members from the Defense Manpower Data Center (DMDC)
with postsecondary data from the National Student Clearinghouse (NSC).
This unique panel data set was constructed specifically for the purpose of
examining veteran educational outcomes, particularly the effects of the ad-
ditional aid allocated by the Post-9/11 GI Bill.
I focus on individuals separating prior to the announcement of the benefit
expansion, circumventing potential concerns related to endogenous acces-
sion to or separation from the military.4 To estimate effects of the benefit
expansion, I use a difference-in-differences strategy that leverages variation
in the postseparation generosity of available benefits over time that differs
between eligible and ineligible veterans. As eligibility depends on an indi-
vidual having an honorable characterization of service, I spend a consid-
erable amount of time demonstrating the relative similarity of eligible and
ineligible veterans in both observables and preperiod enrollment and attain-
ment trends. I also address concerns related to the possibility of heteroge-
neous enrollment and attainment responses of these two groups to the Great
Recession, which overlapped substantially with the early period of the ben-
efit expansion. While it is impossible to conclusively rule out the contribu-
tion of the Great Recession to the break in the enrollment and attainment
trends of the two groups, a variety of empirical evidence suggests that it is
unlikely to have played a major role. This conclusion is supported by my
second empirical strategy, which leverages variation within states over time
that is unrelated to the differences across states in the severity of the reces-
sion.
Unique among the GI Bills and other federal financial aid programs, the
Post-9/11 GI Bill provides different benefit levels depending on the state
and zip code of enrollment. This feature creates variation in the size of
the benefit expansion experienced in different states and zip codes; individ-
uals in some areas received almost no change in maximum benefit levels,
while those in others received tens of thousands of dollars in additional max-
imum benefits per year. I leverage this state-level variation in covered tu-
ition and housing allowance levels as an additional strategy to identify the
effect of aid.

4
This is because benefit levels were based on active-duty service durations after
September 11, 2001, but the benefit was not passed into law until the summer of 2008.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 512 Barr

I find that the benefit expansion increased the likelihood that a veteran
would obtain a degree within 6 years of separation by 5–6 percentage points
(25%), roughly 0.4 percentage points per $1,000 of additional aid. Most of
the increase in degree attainment occurred via an increase in bachelor’s de-
gree attainment. These results indicate that additional financial aid can in-
crease degree attainment, even at high initial levels of support. This suggests
that similar aid expansions for other older, nontraditional students, partic-
ularly those transitioning between jobs, could increase degree attainment
rates substantially.

II. Financial Aid, College Success, and Veterans

A. Financial Aid and Degree Attainment
While research on the effects of financial aid on the enrollment of tradi-
tional students is abundant, relatively little is known about its effects on de-
gree attainment, particularly for older, nontraditional students. Previous stud-
ies focusing on persistence and degree attainment are largely limited to the
evaluations of state merit-aid programs, and the evidence is mixed (Dynarski
2008; Scott-Clayton 2011; Sjoquist and Winters 2015).5 Work on the longer-
term effects of need-based aid is sparse and limited to very local treatment
effects or small experimental groups (Goldrick-Rab et al. 2012; Castleman
and Long 2016).6 While Castleman and Long (2016) suggest an important
role for financial aid in increasing degree attainment in Florida, Goldrick-
Rab et al. (2012) find no positive effect of financial aid on persistence after
the second year in college. So the question remains open as to whether,
how much, or in what respect various kinds and levels of financial aid impact
degree attainment for different types of students.
Individuals who delay enrollment, enroll part time or at community col-
leges, or face additional life responsibilities such as raising a family may re-
spond quite differently to changes in the cost of college. The greater noncol-
lege financial demands faced by such students may mean that financial aid
plays an even more important role in college-going for veterans and other
nontraditional students, particularly if this aid is targeted to the period of
transition between occupations.7 The current paper answers this question.

5
There is also some recent evidence demonstrating positive effects of increased
aid for individuals who are already enrolled in college (e.g., Murphy and Wyness
2016; Denning 2017).
6
In a somewhat related study, Lovenheim and Owens (2014) demonstrate that
cutting off federal financial aid access (loans, grants, or work study) to drug offend-
ers may have an effect on college attendance and degree attainment. However, the
confidence intervals are too large to draw strong conclusions.
7
Two recent papers explore the effects on enrollment of unemployment benefit
availability (Barr and Turner 2015) and information about financial aid (Barr and
Turner 2017) during a similar transition period for civilian unemployment insur-
ance recipients but do not explicitly measure how financial aid affects educational

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 513

B. Effects of GI Bills on Educational Outcomes

Noted historian Sidney Burrell argued that the original GI Bill, passed af-
ter World War II, brought about “what may have been the most important
educational and social transformation in American history” (Burrell 1967).
The bill contributed to a near doubling of college enrollment in less than a
decade, and it is widely credited as a driving force behind the creation of the
middle class. A number of studies have provided support for Burrell’s claim,
finding that the World War II and Korean War GI Bills had positive effects
on educational attainment and earnings (Card and Lemieux 2001; Bound and
Turner 2002; Stanley 2003). These studies identify the combined effect of
military service and higher aid levels primarily by relying on variation in pre-
existing enlistment rates or the likelihood of being drafted.
Like studies of the World War II and Korean GI Bills, studies of the
Vietnam-era GI Bill also estimate the combined effects of compulsory mil-
itary service and higher education benefit levels. Through a clever use of the
Vietnam-era draft lotteries, Angrist (1990) demonstrated that the combined
effects of compulsory service and higher benefit levels resulted in decreased
earnings for young veterans. Twenty years later, Angrist and Chen (2011)
followed up on the same cohorts of individuals and found large increases
in schooling, consistent with those reported for earlier GI Bills. Surprisingly,
though, there was no effect on earnings. The authors reconciled these results
with the earlier earnings estimate by arguing that the earnings gap closed due
to a flattening of the age-earnings profile at later ages as well as modest re-
turns to schooling that were induced by the GI Bill.
However, the combined effect of compulsory military service and higher
aid levels is less relevant in the current era of the volunteer military. The re-
search on post-Vietnam-era GI Bill benefits remains fairly sparse, but the
studies that do exist are distinct from earlier GI Bill analyses in that they
focus on the effects of additional benefits conditional on military service in-
stead of on the joint effect of both service and benefits. Angrist (1993) stud-
ied a small group of veterans who participated in the 1987 Survey of Veter-
ans, finding suggestive evidence of increased educational attainment levels
and earnings in response to higher benefit levels. Similarly, Simon, Negrusa,
and Warner (2010) use plausibly unanticipated changes in the level of edu-
cation benefits provided by the Montgomery GI Bill (MGIB) during the
1990s to explore the impact of those changes on veteran benefit usage. They
find a half percentage point increase in benefit usage per $1,000 of additional
benefits, substantially more modest responses than those found when eval-
uating traditional student financial aid programs (Dynarski and Scott-Clayton

outcomes. Seftor and Turner (2002) present some of the only other evidence for
nontraditional students, estimating the effect of a change in aid using a change in
the definition of independent students in 1986. They find that access to aid increases
enrollment of the affected students by several percentage points.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 514 Barr

2013). However, limited amounts of variation and the absence of a control

group in Simon, Negrusa, and Warners’ study force them to rely on the as-
sumption that an extensive set of explanatory variables adequately controls
for other changes occurring over this period. In short, although interesting
work has been done, it has been limited by the lack of variation in benefit
levels over this period.
Even less is known about the Post-9/11 GI Bill’s effects on educational
outcomes; in fact, there is only one paper of which I am aware that explores
the short-term causal effects of the benefit expansion (Barr 2015). Using pub-
licly available survey data and a difference-in-differences strategy, I found
that the benefit expansion increased by 20% the probability that young vet-
erans would be enrolled at a particular age. However, data limitations made
it impossible to determine the extent to which the observed increases in en-
rollment were true increases in educational attainment or merely shifts in
enrollment timing. As with much of the literature on the effects of financial
aid, data constraints prevented an exploration of the effects of the benefit
expansion on degree attainment.

III. Expansion of Benefits from MGIB to Post-9/11 GI Bill

A brief background on the previous GI Bill helps to clarify how benefits
changed when the Post-9/11 GI Bill was implemented. Prior to the Post-
9/11 GI Bill, the primary education benefit for military veterans was the
MGIB. Implemented in 1985, the MGIB remains an option for enlisting
active-duty personnel as of the writing of this paper. Eligibility for the MGIB
has always depended largely on three factors. First, because the MGIB was
not an entitlement, individuals who chose to participate in the program had
to commit to a pay reduction of $100 for each of the first 12 months on ac-
tive duty. Second, to be eligible an individual had to complete the minimum
active-duty contract agreed to on enlistment (generally at least 3 years) with
an honorable discharge. Third, individuals who had been commissioned
through military academies or who had received Reserve Officer Training
Corps (ROTC) scholarship funds were not eligible for the MGIB benefit.
Under the MGIB, the benefits were sent directly from the Department of
Veterans Affairs (VA) to the veterans who had chosen to enroll. Benefit lev-
els were raised periodically to adjust for inflation, and the flat amounts paid
to individuals monthly were prorated according to half-time, three-quarter-
time, and full-time enrollment.8 In 2007, a full-time student who had com-
pleted three or more years of active-duty service received approximately

8
MGIB benefit levels were essentially flat in real terms between 2003 and 2007,
around $1,140 per month in 2009 dollars, before rising to around $1,300 per month
in 2008 and 2009. Individuals enrolling less than half time receive an amount equiv-
alent to their tuition and fees. However, the partial benefit amounts only subtract
from total benefit eligibility proportionally; i.e., a quarter-time student receiving

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 515

$1,100 per month under the MGIB. Veterans were eligible for 36 months of
benefits; in other words, an individual enrolled full time for 4 years on a tra-
ditional two-semester system would receive benefits throughout the degree
program. The MGIB was not tied to a particular school or location, and
there was little restriction on the types of programs approved. Individuals
could use the benefit for vocational training, apprenticeships, flight classes,
and test fees as well as formal degree-seeking coursework.
Education benefits for veterans were dramatically expanded in the sum-
mer of 2008 when Congress approved the Veterans Educational Assistance
Act of 2008, more commonly known as the Post-9/11 GI Bill.9 The actual
benefit expansion began in August 2009, as additional benefits were provided
to individuals who had served on active duty after September 11, 2001. These
benefits could be applied only to enrollment occurring after August 2009.
Unlike the previous GI Bill, the Post-9/11 GI Bill does not require individ-
uals to opt in to the program at enlistment or prepay for the benefit. The vast
majority of individuals serving on active duty after September 11, 2001, and
receiving an honorable discharge are eligible.10 Benefit levels are tiered on the
basis of active-duty service durations; veterans with at least 90 days of active-
duty service after September 11, 2001, are eligible for 40% of maximum ben-
efit levels, and those with 6 months are eligible for 50%. Each additional
6 months of post-9/11 service results in an additional 10% of eligibility; vet-
erans with greater than 3 years are eligible for 100% of maximum benefit lev-
els. In practice, because the bill is retroactive, nearly all veterans who sepa-
rated honorably after 2005 are eligible for the maximum benefit.11
The Post-9/11 GI Bill benefit has two major components: (1) tuition and
fee coverage and (2) a monthly basic allowance for housing (BAH). For the
tuition and fee benefit, maximum benefit eligibility is based on the highest
tuition level and fee level of any public college in an individual’s state of res-
idence; the VA determines the amounts for each based on information pro-
vided by each state.12 Unlike coverage under the MGIB, the VA pays the
tuition and fee benefit directly to schools, reimbursing the level of tuition

approximately $300 per month uses only approximately one-fourth ($300/$1,100) of

a month of benefits.
9
The delay between announcement and implementation of the benefit expansion
suggests that individuals considering enrollment between these two points might
have had an incentive to delay enrollment in order to capture the higher level of
benefits available in the fall of 2009. However, Barr (2015) demonstrates that this
is not occurring in practice.
10
Officers commissioned at military academies and individuals previously receiv-
ing substantial ROTC scholarships are eligible only after completing an additional
period of service above and beyond their initial requirements.
11
This is because nearly all veterans who separate honorably will have served at
least 3 years.
12
In 2011, coverage for private colleges changed to a maximum of $20,000 across
the country. As very few veterans are likely to be affected by this change (very few

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 516 Barr

and fees up to the in-state maximum of each.13 For an individual attending a

community college or as an in-state student at a public 4-year college, the
school attended would receive reimbursement of, at most, the level of tu-
ition and fees charged to the student. The benefit component paid on behalf
of an out-of-state student at a public 4-year college or any student at a pri-
vate college would be capped at the maximum public in-state tuition and fee
levels set by the VA.14 Veteran students enrolled half time or more were el-
igible to receive a second benefit, a monthly housing allowance. Determined
by the zip code of the institution that the student attends, the 2009 housing
allowance ranged from under $800 in many rural areas to more than $2,700
in New York City. The housing allowance, like the MGIB, is paid directly
to veterans as a conditional cash transfer.15
The greatest difference between the MGIB and the Post-9/11 GI Bill is the
level of benefits provided. In 2008, the MGIB provided roughly $1,300 of
benefits per month for up to 36 months, a maximum of roughly $50,000 in
total benefits. By contrast, the designers of the new GI Bill wanted the ben-
efit to cover the full cost of attendance at a 4-year school and thus set benefit
levels to correspond to each component of this cost. The maximum per-
credit benefit provided by the Post-9/11 GI Bill is more than $1,000 in sev-
eral states, implying a reimbursement of up to $15,000 for a single semester
of tuition. In addition to this, the new GI Bill provides coverage for thou-
sands of dollars in fees per term in nearly all states. Adding in the over
$2,000 monthly housing allowance in areas with a high cost of living, it is
possible for an individual to receive more than $50,000 in benefits in a single
school year. While the actual level of benefits received by most veterans is
much smaller than this, a back-of-the-envelope calculation suggests that
the Post-9/11 GI Bill roughly doubled average maximum benefit levels.16

veterans are enrolled in colleges that would have been affected), it is unlikely to
have a substantial effect on the estimates.
13
That schools are reimbursed directly may be important for at least two reasons.
First, it raises interesting questions about the incidence of the benefits and whether
it is easier for schools to reduce institutional aid to offset veteran aid if they receive
the benefits directly (and thus observe the benefits). These questions are being in-
vestigated in a complementary project. Second, benefits sent to the veteran on a reg-
ular basis may be more salient to the veteran.
14
The VA also initiated a separate yellow ribbon program to cover the tuition
gap for veterans attending out-of-state or private schools. Schools participating
in the yellow ribbon program received matching dollars from the VA for every dol-
lar they contributed to a 100% eligible veteran’s tuition.
15
Students also receive an annual book allowance; the amount is $1,000 for full-
time students.
16
This is based on comparing the veteran population weighted average of state
maximum benefit levels under the Post-9/11 GI Bill with the maximum benefit
available under the MGIB.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 517

Although the Post-9/11 GI Bill greatly expanded education benefits for

veterans, the degree of benefit expansion varies dramatically by geographic
region. To quantify this variation, I estimate the combined annual maxi-
mum tuition and housing allowance for each state: Benefits 5 24

Credit Maximum 1 9
BAH.17 Table A1 and figure 1 (tables A1–A6,
B1, C1–C6 are available online) show that the annual maximum benefit lev-
els under the Post-9/11 GI Bill range from close to MGIB levels in some
states to tens of thousands of dollars more than the MGIB in others; on av-
erage, the increase in a year’s worth of maximum available benefits was
roughly $13,000. I also produce a simulated state benefit under the assump-
tion that the distribution of enrollment across school sectors within a state
will mirror that observed nationally after availability of the Post-9/11 GI
Bill benefits. The realized increase in average benefit levels and the simulated
increase (holding school choice constant) are closer to $5,000 to $6,000. This
variation in nominal benefit levels under the new GI Bill in essence generates
51 microexperiments, in which the level of benefits that veterans received de-
pended on the states to which they likely returned and whether they re-
turned before or after the Post-9/11 GI Bill benefits were available.18

IV. Data
Data constraints are one of the primary factors that have limited rigorous
research on the effect of financial aid on degree attainment. These con-
straints are heightened in assessing the effects on veterans of the expansion
in GI Bill benefits, as no existing data set contains data on enough veterans
to provide even a basic idea of veteran collegiate outcomes, much less a rig-
orous evaluation of the benefit expansion’s effects on educational attain-
ment. As former US Senator Tom Harkin, then chairman of the Committee
on Health, Education, Labor, and Pensions, noted in 2014, “It remains im-
possible for anyone other than the companies and colleges to determine
how veterans are performing” (Harkin 2014). To overcome these data con-
straints, I assembled a unique panel of administrative data that tracks mili-
tary service members while in the military as well as their educational choices
after separation.

17
As an estimate of the annual tuition benefit, I multiply the credit maximum
within a state by 24. As an estimate of a year’s worth of housing allowance benefits
available in a given state, I use the weighted average monthly housing allowance
across zip codes in the state, multiplied by 9. This corresponds to 9 months of en-
rollment in a traditional two-semester system. I exclude the benefits for fees be-
cause the maximum fee levels are generally orders of magnitude larger than the true
fee levels charged by all public institutions in a state.
18
As discussed in the estimation strategy section, I assign veterans to the state
listed as their home of record prior to separation, avoiding concerns related to en-
dogenous location.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 FIG. 1.—Estimate of annual combined maximum benefits. The estimate of annual combined maximum is equivalent to
9
BAH 1 24
Credit Maximum. Tuition credit maximums for 2009–10 were obtained from [Link]
/ch33/tuition_and_fees_2009.htm. Basic allowances for housing are weighted state averages of zip code basic allowance for housing
(BAH) levels for 2010. A color version of this figure is available online.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 519

My primary source of data on those with military service is the active-duty

files of the DMDC. These are administrative data collected by the various
branches and maintained by the Department of Defense. I received data on
the universe of individuals on active duty from 1998 through the end of
2010. The data sets include all individuals on active duty in the Army, Navy,
Marine Corps, and Air Force, for a total of more than 4 million service
members.19 In addition to the standard demographics available in most large
survey data sets, I also observed information on birth, home, and residence
locations; occupation codes; and aptitude (Armed Forces Qualification Test
[AFQT]) scores. I used yearly records for individuals, combined with in-
formation on military separations from the DMDC loss file, to construct
measures of service duration, dates of separation, and likely eligibility for
the Post-9/11 GI Bill (see apps. A–C, available online, for further details).
Likely eligibility for the Post-9/11 GI Bill is determined by both an individ-
ual’s separation date and his or her service characterization.
Having collected data on military service members, I link these data with
information on educational outcomes. Due to the financial cost associated
with the data-linking process, I restricted the population in several ways be-
fore drawing a random sample to be matched. First, I restricted the military
data set to enlisted individuals between ages 22 and 39 who had separated
within 10 years of initial entry into the military.20 I excluded those who had
separated with more than 10 years of service and those aged 40 and over be-
cause (1) they were more likely to view the military as a permanent career
and (2) they were substantially less likely to enroll in college. I then selected
a random sample of individuals who had separated with an honorable dis-
charge between January 1, 2000, and December 31, 2009, as well as a smaller
random sample of those who had separated with less than an honorable dis-
charge. The latter would be used as a control group, as they are ineligible for
GI Bill benefits. I linked the sample of approximately 73,000 veterans with
postsecondary data obtained from the NSC.
Colleges submit enrollment data to the NSC that indicate the beginning
and end dates of student enrollment by term. Participating institutions also
indicate whether a student has earned a degree, the title of the degree, and
the associated college major.21 The linkage between data sets occurred in

19
Most analyses exclude Air Force veterans because codes indicating the home of
record are missing.
20
More than 80% of enlisted individuals will separate within 10 years of initial
entry. Of those who remain in service after 10 years, more than three-quarters will
remain in service at least 20 years, the vesting point for retirement benefits.
21
As Dynarski, Hemelt, and Hyman (2015) note, the NSC’s expansion in enroll-
ment coverage (caused by the addition of new colleges) over time can create a chal-
lenge for researchers who are examining the effect of policies on different cohorts of
students. I follow their suggestion for how to account for the change in enrollment
coverage by restricting my sample to institutions reporting at the beginning of my

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 520 Barr

Table 1
Descriptive Statistics
Mean
Variable (1)
Age at separation 25.05
Black .155
Male .840
Married .419
AFQT score 60.12
High school degree at separation .902
Enroll within 3 years .447
Any degree within 5 years .181
Any degree within 6 years .222
Bachelor’s degree within 5 years .097
Bachelor’s degree within 6 years .127
Observations 29,361
Enrolled fall 2006 through fall 2013:
Share 2-year public .45
Share 4-year public .32
Share for profit .13
Share private nonprofit .10
NOTE.—The table presents descriptive statistics for individuals
in the National Student Clearinghouse sample who separated be-
tween 2003 and 2008. Enrollment statistics present the break-
down of school types for those observed enrolled between the fall
of 2006 and the fall of 2013. AFQT 5 Armed Forces Qualifica-
tion Test.

August 2014, so the sample contains enrollment and degree attainment in-
formation up to that point. The NSC data provide a picture of veteran en-
rollment and degree attainment not previously available, but they contain
very little information about the characteristics of schools themselves. Us-
ing school-specific identifiers in the NSC data, I link all institutions to in-
formation on the universe of Title IV aid-receiving colleges contained in
the Integrated Postsecondary Education Data System. These data contain
information about the control of the institution (i.e., public, private non-
profit, or for profit), whether the college is a 2- or 4-year school, measures
of selectivity, tuition and fee levels, and graduation rates.
Table 1 provides summary statistics for the sample of veterans linked
with NSC data. Veterans are substantially more likely to be male and have
been married than the general student population. Nearly half of the veter-
ans in my sample enroll in college within 3 years of separating from the mil-
itary. Most veteran enrollment occurs at public 2-year (45%) or 4-year
(32%) institutions. Enrollment at for-profit institutions (13%) is likely un-

sample; in most specifications, this was 2002. Enrollment results are robust to vary-
ing the cutoff year for inclusion. This has no effect on the degree outcomes. See
apps. A–C for further details on the NSC data and the data-linking process.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 521

derestimated given the coverage of the NSC data (see Dynarski, Hemelt,
and Hyman 2015).
Overall, the summary statistics indicate that nearly half of the eligible vet-
erans in the sample enrolled in college, and around 20% obtained an asso-
ciate’s degree or higher within 6 years. Despite the negative press surround-
ing veteran educational outcomes, these statistics imply that around 40% of
veterans who go to school within 3 years of separation obtain a degree of
some kind within 6 years; this is comparable to the 6-year degree receipt
of individuals starting college between ages 19 and 2422 and substantially
higher than statistics about veterans that have been cited in the media.23

V. The Timing of Benefit Availability across Cohorts

Table 2 illustrates the differential treatment of separating cohorts, with the
earliest cohorts (2000 and 2001) having no eligibility for the Post-9/11 GI
Bill, the middle cohorts (2002–4) having partial eligibility, and the most re-
cent cohorts (2005–9) having full eligibility.24 As most veterans begin to use
their benefits within a few years of separation, columns 3 and 4 indicate the
number of years that the Post-9/11 GI Bill was actually available to individ-
uals in each cohort during the 5–6 years after separation. In other words,
although individuals in the 2003 separating cohort were eventually given
partial Post-9/11 GI Bill eligibility in 2009, their education decisions during
the period 5–6 years after separation were made when only the MGIB was
available.25 In contrast, individuals in the 2008 separating cohort were able
to use the Post-9/11 GI Bill during nearly all of the 5–6 years after separa-
tion.
The main idea of the paper is to leverage this variation in postseparation
benefit availability to identify the effect of the generosity of benefits avail-
able during the years immediately following separation on degree attain-
ment. I focus on this particular margin (1) because of the policy import of
understanding how increased benefit generosity facilitates investments in
human capital during the transition into the civilian labor force (a stated mo-
tivation of the benefit expansion), (2) because most veterans begin using

22
Author’s calculations using Beginning Postsecondary Students.
23
See, e.g., [Link]
-out_n_2016926.html, which cites a dropout figure of 88%.
24
Here, a cohort is defined to include individuals separating prior to August 1 of
the year indicated and after July 31 of the prior year. Thus, individuals in the 2008
separation cohort have separation dates between August 1, 2007, and July 31, 2008.
I use this definition for two reasons: (1) it follows the traditional college calendar
and (2) it allows me to focus on veterans who separated at least 1 year prior to the
implementation of the Post-9/11 GI Bill, whose separation was thus unlikely to have
been influenced by the benefit expansion.
25
Recall that the extent of eligibility depends on the duration of active-duty ser-
vice after September 11, 2001.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 522 Barr

Table 2
Timeline of Eligibility and Availability of Benefits
Years of Post-9/11 Availability
5 Years after 6 Years after
MGIB Eligibility Post-9/11 Eligibility Separation Separation
Separating Cohort (1) (2) (3) (4)
2000 Yes No 0 0
2001 Yes No 0 0
2002 Yes Partial 0 0
2003 Yes Partial 0 0
2004 Yes Partial 0 1
2005 Yes Full 1 2
2006 Yes Full 2 3
2007 Yes Full 3 4
2008 Yes Full 4 5
2009 Yes Full 5 6
NOTE.—Separating cohorts are defined as indicated in the text. All cohorts are eligible for the Montgom-
ery GI Bill (MGIB; col. 1). Column 2 indicates eligibility for Post-9/11 GI Bill benefits, which is based on
the duration of active-duty service after September 11, 2001 (see the text for details). While all cohorts after
2001 had at least partial eligibility for Post-9/11 GI Bill benefits, these benefits were not available until the
fall of 2009. Columns 3 and 4 indicate the number of years that Post-9/11 GI Bill benefits were available
during the 5 or 6 years after separation. Column 3 indicates the number of years out of the 5 years after
separation that Post-9/11 GI Bill benefits were available to a cohort. Similarly, col. 4 indicates the number
of years out of the 6 years after separation that Post-9/11 GI Bill benefits were available to a cohort. Co-
horts used to produce main estimates are in bold.

their benefits immediately after separation, making it the natural period to

study,26 and (3) to ease comparison with the existing literature, which tends
to focus on degree attainment within 5–6 years of high school graduation.
Figure 2 illustrates how the main estimation strategy will leverage this var-
iation in postseparation benefit availability over time to identify the effect of
additional aid. Figure 2 plots the proportion of veterans in a cohort who
have obtained an associate’s degree or higher in the years after separation.
For consistency with my main estimation strategy, I focus on the 2003,
2004, and 2008 cohorts; a figure with all cohorts is included in appendix A
(fig. A1; figs. A1–A4, C1–C3 are available online).27 Degree attainment
rates for the 2003 and 2004 cohorts (blue and red circles), which had highly
delayed access to Post-9/11 GI Bill benefits, are nearly identical to each
other.28 Only about 17% of these individuals received a college degree within

26
Fifty percent of those eligible begin using their benefits within 3 years after
separation. By 9 years after separation, only 55%–60% (i.e., an additional 5%–
10%) have begun using their benefits.
27
Figure A1 shows the attainment trajectories for all separating cohorts, illustrat-
ing (as expected) that while the partially treated cohorts had somewhat higher at-
tainment the levels, the cohorts separating around the time of the actual benefit ex-
pansion experienced the greatest increases in degree attainment.
28
The 2003 cohort would not receive Post-9/11 GI Bill funds until the sev-
enth year after separation, while the 2004 cohort would be eligible for benefits in
their sixth year after separation.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 523

FIG. 2.—Attainment of any degree by separation year. Each marker represents

the share of eligible veterans who obtained any degree by years after separation.
Statistics are presented separately for the 2003, 2004, and 2008 cohorts. See table 3
or the text for details on the sample restrictions. A color version of this figure is
available online.

6 years of separation. By contrast, the 2008 cohort (green triangles), which

had access to Post-9/11 GI Bill benefits soon after separation, were between
5–7 percentage points more likely to obtain a degree over the 6 years follow-
ing separation. Figure 2 thus indicates that individuals who were able to ac-
cess the higher levels of aid from the Post-9/11 GI Bill (those separating in
2008) were more likely to obtain a degree than those who were not (those
separating several years earlier).
Figure 3 combines the variation in postseparation benefit availability over
time with cross-sectional variation in eligibility to illustrate the difference-
in-differences strategy. The figure plots the percentage of likely GI Bill–
eligible and GI Bill–ineligible veterans who enrolled within 3 years of sep-
aration, by year of separation.29 Here, likely eligibility is determined by
having an honorable service characterization at separation.30 As expected,
29
The set of schools is restricted to those that began reporting to the NSC by the
beginning of 2002.
30
Recall that having an honorable service characterization is required for benefit
eligibility. As I use the ineligible group as part of my estimation strategy, I discuss
its suitability as a control group at length in that section. Although veterans can
have their service characterization altered after separation, the characterization at

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 524 Barr

FIG. 3.—Three-year enrollment rates by year of separation. The figure plots the
share of veterans enrolling within 3 years of separation by year of separation. Cir-
cles represent the enrollment rates of eligible veterans, and squares represent the en-
rollment rates of ineligible veterans. See the note to table 3 or the text for details on
the sample restrictions. A color version of this figure is available online.

the enrollment rate of veterans unlikely to be eligible for the MGIB or the
Post-9/11 GI Bill is substantially lower than that of eligible veterans. How-
ever, the trend in enrollment of both groups is relatively flat between 2002
and 2006. As benefits for the Post-9/11 GI Bill first became available in Au-
gust 2009, many veterans in the cohort separating in 2007 would have had
access to the higher benefits during their third year after separation; this can
explain the small uptick in the enrollment of eligible veterans for the 2007
cohort. All eligible veterans separating in 2008 had access to the higher ben-
efit levels within 1–2 years of separation, which explains the substantially
higher enrollment levels for these individuals. The 3-year enrollment rate
of eligible veterans separating in 2008 is roughly 8 percentage points higher
than those of cohorts separating between 2002 and 2006. The enrollment of
veterans unlikely to be eligible exhibits no such pattern. For completeness, I
include the 2009 separating cohort. This cohort was potentially affected by

separation serves as a good proxy for benefit eligibility because only 10% of en-
rolled veterans with a non-honorable-service characterization are observed using
MGIB benefits. The degree that individuals classified as ineligible are actually eli-
gible biases my estimates toward zero.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 525

endogenous separation from the military in response to higher benefit levels

and/or endogenous changes in service characterization. While I do not in-
clude this cohort in the formal analyses, there is evidence of even higher en-
rollment rates for these individuals. Combined, the evidence suggests that a
portion of the increase in degree attainment came via increased enrollment.

VI. Estimating the Effects of Aid Expansion

To identify the effects of financial aid on veterans’ degree attainment, I
begin with a simple difference-in-differences approach, identifying the ef-
fect of the benefit expansion as the difference between degree attainment
rates of eligible veterans separating in the preperiod (2003 or 2004) and those
separating in the postperiod (2008), using ineligible veterans as a control
group.31 I then turn to a strategy that leverages the geographic variation in
the size of the benefit expansion.

A. Difference in Differences: Estimation Strategy

The difference-in-differences specification takes advantage of the differ-
ence in the availability of benefits over time across veteran types. Eligible
veterans (those with an honorable service characterization at separation)
form the treatment group, and ineligible veterans (those with a less than
honorable service characterization) are the controls. The basic specification
is
Eist 5 b1 Xist 1 b2 Zst 1 as 1 lt 1 g0 Eligiblei
(1)
1 g1 Post911t
Eligiblei 1 eist :
Here, Eist indicates whether individual i who lists home of record of state s
and separated in year t has obtained a degree within 5 or 6 years of separa-
tion. The vector Xist refers to characteristics at the time of separation, includ-
ing age, sex, AFQT (aptitude score), marital status, whether a veteran is
black, separation month, and education level at separation. In Zst, I control
for changing state-level characteristics, including the average unemployment
rate faced by an individual in state s during the year following separation.
Fixed effects as for each home-of-record state control for time-invariant dif-
ferences in the likelihood of degree attainment for individuals returning to
different states, and fixed effects lt for each year of separation control for
changes over time that affect the likelihood of degree attainment for all vet-
erans within a cohort. The sample is restricted to individuals separating in
the 2003, 2004, and 2008 cohorts. As mentioned above, I use the 2003 and
2004 cohorts as the preperiod because their 5–6-year degree receipt was un-

31
Recall that while both of these cohorts separated prior to the implementation
of the Post-9/11 GI Bill, there is substantial variation in the postseparation benefit
availability between those separating in the preperiod and the postperiod (table 2).

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 526 Barr

likely to have been affected by a benefit expansion implemented in the fall of

2009. Those separating in 2008 are in the postperiod; Post911t is an indicator
variable set to 1 if an individual separated in 2008. As mentioned previously,
my approach is to compare cohorts that lack postseparation availability of
Post-9/11 GI Bill benefits (2003 and 2004) with a cohort that has the ben-
efits available soon after separation (2008). The treatment group is the group
of eligible veterans, given by Eligiblei. The parameter of interest, g1, is the
difference-in-differences estimate of the effect of additional benefits. If ad-
ditional aid has an effect on degree receipt, the difference in attainment rates
of eligible and ineligible veterans should be greater for those separating in
2008 than those separating in 2003 or 2004.
The key identifying assumption is that, conditional on a large and de-
tailed set of controls, the change in the degree attainment of ineligible vet-
erans serves as a reasonable proxy for the counterfactual change in eligible
veteran degree attainment if the benefit expansion did not occur—the stan-
dard parallel trends assumption. I devote considerable attention to discussing
whether ineligible veterans are a reasonable control group. Here, I merely
note two pieces of evidence that support the notion: first, the two groups
are observationally similar; and second, degree attainment (and enrollment)
levels trend together prior to the benefit expansion, suggesting that the two
groups respond similarly to changing conditions. To further address this
concern over the control group selection, I also implement a matching strat-
egy to select a control group that is observationally equivalent to the treated
group (Smith and Todd 2005; Ferraro and Miranda 2014).32

1. Difference in Differences: Estimates

The main difference-in-differences estimates in column 2 of table 3, which
condition on soldier characteristics, indicate that eligible veterans separat-
ing a year prior to implementation of the Post-9/11 GI Bill are 4.5 percent-
age points more likely to obtain a degree within 5 years than those separat-
ing 5 or 6 years prior to implementation, an increase of more than 25%.
Dividing by an estimate of the average increase in maximum benefit levels
of $13,000 suggests that every $1,000 of aid increases degree attainment
within 5 years by 0.35 percentage points.33 Most of this increase in attain-

32
See app. A for more details on how this approach is implemented in the current
paper. Recent work by Ferraro and Miranda (2014) suggests that estimates from
observational studies may be improved if researchers “preprocess” the data to care-
fully select an appropriate control group. The basic idea is to use propensity score
matching based on fixed observables to generate a better control group. Once this
control group is determined, standard difference-in-differences estimation is imple-
mented.
33
The effect of actual aid levels is likely larger because most veterans do not ac-
cess maximum benefit levels. Dividing by an estimate of the realized average in-

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 527

Table 3
Degree Effects 5 and 6 Years after Separation: Difference-in-Differences
Regressions
Associate’s Associate’s Associate’s Bachelor’s Bachelor’s Bachelor’s
or Higher or Higher or Higher or Higher or Higher or Higher
Variable (1) (2) (3) (4) (5) (6)
Five years after
separation .056*** .045*** .051** .041*** .033*** .047***
(.015) (.016) (.024) (.011) (.012) (.016)
Covariates X X X X
Preprocessed X X
Observations 15,457 15,457 15,136 15,457 15,457 15,136
Mean .164 .164 .164 .0994 .0994 .0994
Six years after
separation .072*** .060*** .064** .054*** .045** .062***
(.018) (.021) (.027) (.017) (.018) (.020)
Covariates X X X X
Preprocessed X X
Observations 15,457 15,457 15,136 15,457 15,457 15,136
Mean .201 .201 .201 .131 .131 .131
NOTE.—The table shows coefficients from difference-in-differences regressions that compare degree at-
tainment levels for eligible and ineligible veterans from before to after the implementation of the Post-
9/11 GI Bill. Each column presents the estimate from a single regression for a different dependent variable.
All specifications also include year and state fixed effects and the average unemployment calculated for an
individual’s home of record during the 12 months after separation. Specifications with covariates also in-
clude indicator variables for sex, whether the person is black, marital status, high school degree at separa-
tion, and month of year of separation as well as Armed Forces Qualification Test scores, controls for age of
separation, and baseline degree level. Preprocessed data use a matching procedure to form control groups as
described in the text and apps. A–C. Robust standard errors clustered at the state level are shown in paren-
theses. The sample is restricted to veterans aged 22–39 with less than a bachelor’s degree who separated
between August 1, 2002, and July 31, 2004 (the preperiod), or between August 1, 2007, and July 31,
2008 (the postperiod), and listed a US state as a home of record.
** p < .05.
*** p < .01.

ment occurs through an increase in the receipt of degrees at the bachelor’s

level or higher. Extending the horizon to 6 years, the estimate grows to
6 percentage points, or 0.45 percentage points for every $1,000 of aid, for
any degree attainment and 4.5 percentage points for bachelor’s degree at-
tainment, more than 25% higher than the base.34

crease in benefit levels, $5,000 suggests a larger 0.9 percentage point effect for each
$1,000 of aid.
34
Standard errors are clustered at the state level. These standard errors may be
underestimated if the residuals are autocorrelated within cross sections (eligibility
groups). For completeness, I note here that p-values are quite similar and are even
smaller when clustering on eligibility (which would account for this correlation)
as well as eligibility by state, separation year, or separation year by eligibility. While
in some cases this results in few clusters, p-values remain well below 0.05 even after
making the appropriate adjustment to the sample test statistics (i.e., using TðG 2 1Þ
p-values). The standard errors and p-values presented are the most conservative
produced across any of these strategies.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 528 Barr

2. Difference in Differences: Addressing Threats to Internal Validity

The primary assumption underlying the difference-in-differences analy-
ses is that the degree attainment of ineligible veterans is a good proxy for the
change that would have occurred for eligible veterans, absent the benefit ex-
pansion. Figure 4 addresses this parallel trends assumption. The event study
plots the coefficients and confidence intervals from a similar specification as
in equation (1) but with the full set of Sep Yeari, Eligiblei, and Sep Yeari

Eligiblei indicator variables. The coefficients for each separation year inter-
acted with benefit eligibility are plotted. The larger black circles represent
the years used in estimation. The plot suggests that the attainment levels
of eligible and ineligible veterans roughly move together for veterans sepa-
rating between 2001 and 2006 but diverge sharply in 2007, 2008, and 2009.
The lack of divergence for the 2005–6 cohorts may be somewhat surprising
because these individuals were eligible for the higher level of benefits within
4 or 5 years of separation and suggests that aid is substantially more effective
if it comes during the first few years after separation. In contrast, a similar
event study for bachelor’s degree receipt shows an upward trend in the at-

FIG. 4.—Event study: effect on any degree 5 years after separation. The depen-
dent variable is receipt of any degree within 5 years of separation (results are similar
using 6 years). Circles represent the coefficient estimates for each year of separation
interacted with eligibility. Large black circles represent years used in main estimates
presented in tables. Smaller and lighter plus signs indicate the 95% confidence in-
terval (CI) for each point estimate. The 2004 coefficient is set to zero because it is
the omitted year. See the note to table 3 for details on the sample restrictions.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 529

FIG. 5.—Event study: effect on bachelor’s degree 5 years after separation. The
dependent variable is receipt of a bachelor’s degree or higher within 5 years of sep-
aration (results are similar using 6 years). Circles represent the coefficient estimates
for each year of separation interacted with eligibility. Large black circles represent
years used in main estimates presented in tables. Smaller and lighter plus signs in-
dicate the 95% confidence interval (CI) for each point estimate. The 2004 coeffi-
cient is set to zero because it is the omitted year. See the note to table 3 for details
on the sample restrictions.

tainment rate beginning for the 2005 and 2006 separating cohorts (fig. 5).
This is, perhaps, expected because bachelor’s degree recipients generally en-
roll for at least 4 years, and thus individuals who separated in 2005 and 2006
and attempted a bachelor’s degree were likely still enrolled when the bene-
fit expansion was implemented. It also suggests that some individuals may
have foregone attainment of an associate’s degree in favor of bachelor’s de-
gree attainment.
I include the 2001, 2002, 2005–7, and 2009 separation cohorts in figure 4
for illustrative purposes, although they are not in the sample underlying ta-
ble 3.35 My approach is to compare cohorts (the dark black circles in figs. 4,

35
The 2001 and 2002 cohorts are excluded because MGIB benefit levels were in-
creased between 2001 and 2003. The 2005–7 cohorts are excluded because they are
partially treated (i.e., eligible for the higher level of benefits within a few years of
separation), and the 2009 cohort is excluded because of concerns related to endog-
enous separation from the military in response to the benefit expansion.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 530 Barr

5) who lack postseparation of Post-9/11 GI Bill benefits (2003, 2004) with a

cohort (2008) that has the higher level of benefits available soon after sepa-
ration. While I think it is natural to focus on the stark variation in post-
separation availability of benefits between these cohorts, I also present
estimates that leverage the intervening cohorts. This is addressed more for-
mally in table A2, but the event studies and associated discussion suggest the
answers I will find; the attainment of ineligible and eligible veterans appears
to trend together until we get to cohorts that experienced exposure to the
higher level of benefits. The partially treated cohorts (2005–7) appear to
be affected to some extent (particularly in terms of bachelor’s degree attain-
ment), but less so than fully treated cohorts (2008, 2009). Interpretation of
the effect on the 2009 separating cohort is muddied by concerns related to
the endogenous separation and service characterization of individuals in re-
sponse to the benefit expansion.
The similar pretreatment trends suggest that ineligible veterans are an ap-
propriate control group for eligible veterans, but the groups differ in other
ways. In table A3, I provide information on the covariate balance of the two
groups. While the eligible veterans are slightly older, are less likely to be
black, are slightly more educated, and have higher aptitude scores, the dif-
ferences are fairly small. Furthermore, there is very little change in the ob-
servables of the two groups from the preperiod to the postperiod (table A3)
or over the sample period more generally (fig. A2). A comparison of col-
umns 1 and 2 in table 3 indicates how the inclusion of covariates affects the
estimates. While the inclusion of covariates reduces the magnitudes slightly,
the estimates are statistically distinguishable. I also use a matching approach
to select ineligible veterans who are similar to eligible veterans.36 In column 3
of table 3, I present the difference-in-differences estimates using this matched
control group. The point estimates are somewhat larger but statistically indis-
tinguishable from those using the other difference-in-differences approaches.
In addition to general concerns about the appropriateness of the control
group, another potential concern with the difference-in-differences strategy
is that the Great Recession may be responsible for the relative increase in the
attainment levels of eligible veterans. For this to be the case, eligible veteran
educational attainment would have to be more responsive to downturns
than noneligible veteran educational attainment. This may be reasonable
if noneligible enrollees have reduced resources to buffer economic down-
turns. However, the inclusion of state-year unemployment rates Unempst
and their interaction with eligibility status helps control for this concern;

36
I implement this approach separately in the pre- and postperiods (see app. A
for details on the process). Table A3 illustrates how the preprocessing improved
the covariate balance between the two groups, making them observationally equiv-
alent across a range of covariates.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 531

this strategy leaves the results unchanged.37 This point is buttressed by re-
cent evidence that the college enrollment of veterans is not affected by state
unemployment rates (Borgschulte and Martorell 2016). Furthermore, fig-
ures 4 and 5 suggest that the two groups were trending together in the
preperiod, including the period of the 2001 recession. While a shallower
and shorter recession, this suggests that there is likely not a strong difference
in how the two groups respond to business cycles. Although it is unlikely
that the Great Recession is driving the results, in the next section I turn
to estimates that use geographic variation in the size of benefit expansion
over time; this variation is unrelated to differences across states in the sever-
ity of the recession (as measured by the unemployment rate).
A related concern is that changes to the generosity of Pell grants, which
occurred at roughly the same time as the benefit expansion and Great Re-
cession, may have affected GI Bill–eligible and GI Bill–ineligible veterans
differently. This is unlikely to be the case because both groups of veterans
face similar requirements to access Pell grants. Furthermore, the Post-9/11 GI
Bill benefits have no effect on Pell eligibility. Pell eligibility is based on an
individual’s expected family contribution. By statute, veteran education ben-
efits are excluded from income for the purposes of this calculation. Further-
more, eligibility for a Pell grant does not take into account other scholarships
or awards, so there was no crowd-out from changes in policy.
Another related concern with the difference-in-differences strategy is that
the composition of separating veterans has changed over time in ways that
are not controlled for or observed, perhaps as a result of the Great Recession.
However, given the administrative process of separations and the fact that
labor market conditions did not begin to deteriorate until the middle of
2008, it is unlikely that endogenous separation as a result of the Great Reces-
sion is responsible for the observed results. Furthermore, there is no evi-
dence of a change in reenlistment rates of soldiers over this period. While
my regressions all control for a rich set of individual characteristics, fig-
ure A2 further illustrates relatively small changes in a variety of demographic
characteristics over the time period of the analysis.38 We might still be con-

37
The unemployment interaction term is insignificant and the point estimates
(with standard errors in parentheses) for my preferred specifications in the top panel
of table 3 (cols. 2, 5) are 0.048 (0.016) vs. 0.045 (0.016) for associate’s degree or higher
and 0.035 (0.011) vs. 0.033 (0.012) for bachelor’s degree or higher. The analogous
comparisons in the bottom panel are 0.064 (0.020) vs. 0.060 (0.021) for associate’s de-
gree or higher and 0.047 (0.017) vs. 0.045 (0.018) for bachelor’s degree or higher.
38
Barr (2015) also explores a variety of factors that may have affected the com-
position of military veterans, including changes in the composition of recruits, sui-
cide, and mortality. Overall, the evidence suggests that, if anything, changes in
composition likely would have led to lower enrollment and attainment levels as
the military lowered standards slightly for enlistment during the mid-2000s.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 532 Barr

cerned if veterans with unobservable preferences for education were more

likely to separate in 2008. However, the sample is restricted to veterans sep-
arating prior to the end of the summer during which the Post-9/11 GI Bill
was passed. Given the administrative process of separation, it is extremely
unlikely that individuals in the sample separated in response to a benefit ex-
pansion that had not yet been passed.

B. Geographic Variation: Estimation Strategy

I now turn to another strategy for estimating the effects of veteran finan-
cial aid expansion, one that does not make use of ineligible veterans as a con-
trol group. This complementary strategy instead leverages geographic var-
iation in the size of the benefit expansion generated by the transition from
the MGIB to the Post-9/11 GI Bill (fig. 1 or table A1). Assigning eligible
individuals to states based on their home of record at the time of separation,
I link in average maximum expected benefit levels:
Eist 5 b1 Xist 1 b2 Zst 1 as 1 lt
(2)
1 gPost911t
Benefits 1 eist :
The effect of additional aid is identified by the covariation between degree
attainment levels and the size of the benefit expansion experienced by vet-
erans returning to a particular state (Benefits). For example, in 2005 an MGIB-
eligible veteran returning to any state was eligible for roughly $11,000 (2009
dollars) in 1 year of benefits. In the fall of 2009, a veteran returning to
Oklahoma was eligible for about $3,500 in tuition benefits and $8,280 in
housing allowance benefits, for a total of just under $12,000. In contrast,
a veteran returning to New Jersey was eligible for about $10,800 in tuition
benefits and $17,000 in housing allowance benefits, almost $28,000 in total.
If additional aid causes higher educational attainment, we would expect to
see a larger increase in the educational attainment levels of veterans return-
ing to New Jersey versus those returning to Oklahoma, for example, when
comparing cohorts separating several years before Post-9/11 GI Bill imple-
mentation with those separating within a year of the benefit expansion.39
My key assumption is that, conditional on the set of observables, the un-
observed factors that affect attainment are uncorrelated with the size of the
benefit expansion in a state. For example, if veterans interested in college en-
rollment migrate to states with higher benefit levels following the benefit
expansion and I used their eventual state of residence to assign benefit levels,
this assumption would be violated. In other words, the actual residence of

39
As the new GI Bill itself illustrates, a dollar may not be worth the same amount
everywhere in the country. In apps. A–C, I discuss this point and also present es-
timates that adjust the differences in the benefit expansion for cost-of-living differ-
ences.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 533

the veteran once the details on the benefit expansion are available is endog-
enous. It is for this reason specifically that I limit the sample to individuals
separating prior to the fall of 2008 and use the veteran’s home-of-record
state at separation to assign benefit levels.40 The benefit maximum in a vet-
eran’s home-of-record state at separation will be strongly correlated with
the actual benefit maximum faced by the veteran, but it will be uncorrelated
with preferences for education after conditioning on state fixed effects. To
the degree that veterans do not reside in their home of record at separation
or move to take advantage of the higher benefit levels in another state, my
estimates will be biased toward zero. In this sense, the geographic estimates
provide an intent-to-treat estimate and should be scaled up when they are
being compared with the average treatment effect estimates provided by
the other estimation approaches.

1. Geographic Variation: Estimates

In table 4, I present estimates of the effect of an additional $1,000 in max-
imum annual combined tuition and housing allowance benefit levels in-
stead of the effect of the Post-9/11 GI Bill as a whole. The coefficient on
Post911t
Combined Max indicates how 6-year degree attainment levels
vary with the size of the benefit expansion in a state. The estimate in column 1
indicates that $1,000 of additional annual benefits increases the probability
that a veteran obtains any degree within 6 years by 0.2 percentage points;
the estimate is 0.4 percentage points when using the simulated benefit size.41
Both sets of estimates are highly statistically significant.42 As discussed in
the previous section, not all veterans actually return to or remain in their
home-of-record state; this “crossover” of veterans from one state to another
is similar to the crossover of participants from one treatment arm to another
in a randomized controlled trial. As such, the results in table 4 are intent-to-
treat estimates that should be scaled up before being compared with the
difference-in-differences estimates. Roughly 65% of enrolled veterans are
enrolled in their home-of-record state. Under the assumption that veterans
will be more likely to migrate toward higher-benefit states, dividing the
point estimate by .65 provides a lower-bound estimate for the treatment ef-

40
Results are nearly identical using the home-of-record state from a year or two
prior to separation (which varies little from that listed at separation).
41
Recall that the simulated benefit size is generated using the nationwide
postbenefit expansion enrollment distribution across sectors to generate state-level
simulated average benefit levels.
42
In figs. A3 and A4, I demonstrate the robustness of the p-values to a conser-
vative form of randomization inference. I randomly reassign benefit levels to states
and estimate the coefficients in col. 1 of table 4. I repeat this process 1,000 times and
plot the distribution of estimates. For both specifications, just over 1% of estimates
are larger than the estimate obtained using the true assignment of benefit levels.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 534 Barr

Table 4
Degree Effects Determined Using Geographic Variation (6 Years)
Associate’s or Higher Bachelor’s or Higher
Variable (1) (2)
Post911t
Combined Max .0020*** .0016***
(.0005) (.0005)
Post911t
Combined Simulated .0040** .0037**
(.0016) (.0015)
Observations 14,222 14,222
Mean .201 .131
NOTE.—The table shows coefficients from regressions of degree attainment levels on the size of the an-
nual estimate of the combined Post-9/11 GI Bill maximum or the simulated size of the annual benefit (in
thousands) interacted with whether the individual separated between August 1, 2007, and July 31, 2008 (the
postperiod). Each cell presents the estimate from a single regression. The specification also includes indi-
cator variables for sex, whether the person is black, marital status, high school degree at separation, and
month of year of separation as well as Armed Forces Qualification Test scores, controls for age of separa-
tion, baseline degree level, year and state fixed effects, and the average unemployment calculated for an in-
dividual’s home of record during the 12 months after separation. Robust standard errors clustered at the
state level are shown in parentheses. The sample is restricted to veterans aged 22–39 with less than a bach-
elor’s degree who separated between August 1, 2002, and July 31, 2004 (the preperiod), or between Au-
gust 1, 2007, and July 31, 2008 (the postperiod), and listed a US state as a home of record.
** p < .05.
*** p < .01.

fect.43 This calculation indicates that each additional $1,000 of aid increases
the likelihood of degree attainment by 0.31 percentage points, slightly
smaller than, but statistically indistinguishable from, the effect sizes implied
by the difference-in-differences estimates.44 The size of the benefit may not
have been salient to all veterans, explaining the somewhat smaller estimates
obtained using geographic variation in the size of the benefit expansion.
First, a portion of the higher benefit level is accessible only if one attends
a more expensive school. Second, it is unclear to what degree all veterans
were aware of the exact nature of the provided benefit (see apps. A–C for
further discussion). Finally, if veterans moved to take advantage of benefit
levels, estimates will be biased toward zero.
An alternative approach to scaling the point estimates is to calculate the
combined benefit level for veterans with a particular home of record under
the assumption that veterans separating in 2008 will return to geographic
locations in the same way that individuals separating earlier did.45 In other

43
This also assumes that the migration decisions of all veterans are roughly sim-
ilar to those of enrolled veterans.
44
The effect of the simulated benefit size is scaled to 0.62 percentage points.
45
I follow the literature in estimating the effect of additional available benefits,
but another interesting suggestion that has been made is to instrument for the actual
benefit level received by each veteran. While data constraints prevent a full explo-
ration of this approach, there are also concerns related to the interpretation of the
effect of actual benefits received in a structural equation where positive benefit re-
ceipt is conditional on enrollment (i.e., enrollment must equal 1 if the level of actual
benefits received is positive). For example, if the benefit expansion has no effect on
the level of benefits conditional on enrollment but does increase enrollment, the av-

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 535

words, if 50% of veterans with a Kansas home of record end up in Colorado

and 50% end up in Kansas, the maximum benefit for an individual with a
Kansas home of record will be set to equal the average of the Colorado
and Kansas benefit levels.46 Estimates from this strategy indicate that each
additional $1,000 of maximum aid increases the likelihood of degree attain-
ment by 0.30 percentage points, statistically indistinguishable from the ef-
fect sizes implied by the scaled geographic estimates and the difference-
in-differences estimates.

2. Geographic Variation: Addressing Threats to Internal Validity

A major concern associated with these geography-based estimates is that
the size of the Post-9/11 GI Bill benefits in a state—and thus the magnitude
of the benefit increase—is somehow related to other factors that influence
educational choices. For example, the Post-9/11 GI Bill benefit maximum
in a state may be correlated with trends in resources available for higher ed-
ucation or labor market opportunities. If this is the case, it may be that these
factors, not the changing benefit availability, are causing veteran attainment
rates to change differentially across states. Table A6 indicates that there is no
relationship between trends in appropriations or labor market conditions
and the size of the benefit increase; in fact, the point estimate for log appro-
priations to higher education is negative, suggesting that states with larger
veteran benefit increases have actually been spending less on higher educa-
tion over the relevant period. The estimate for the unemployment rate is a
tightly estimated zero.
A separate approach to addressing this threat to validity is to test for
trends in attainment or enrollment among the civilian population that cor-
relate with the Post-9/11 GI Bill benefit maximum. Table A6 presents these
estimates for a number of age groups and education restrictions using the
2006–12 samples of the American Community Survey. Individuals in higher-
benefit states become less (not more) likely to have a degree or be enrolled in
college over this time period.
Finally, I present an event study that illustrates that the relative increase in
the degree attainment of veterans returning to states with higher benefit lev-
els under the Post-9/11 GI Bill began to occur only for the cohorts separat-

erage benefit level received will go up. However, interpreting this as an effect of ad-
ditional benefits is misleading. A related approach, instrumenting for the maximum
benefits in the state an individual ends up in, would do a better job addressing the
scaling issues discussed in this section, but I do not observe where nonenrolled vet-
erans reside.
46
In practice, I use data on veterans separating in 2003 and 2004 to calculate the
share of individuals with a particular home of record who are enrolled in each state.
I use these shares to create a weighted average benefit level for each home of record.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 536 Barr

FIG. 6.—Event study (geographic variation): effect of state benefit level on any
degree. The dependent variable is receipt of any degree within 5 years of separation
(results are similar using 6 years). Circles represent the coefficient estimates for each
year of separation interacted with the simulated state benefit size. Large black cir-
cles represent years used in main estimates presented in tables. Smaller and lighter
plus signs indicate the 95% confidence interval (CI) for each point estimate. The
2004 coefficient is set to zero because it is the omitted year. See the note to table 3
for details on the sample restrictions.

ing in 2008 and 2009 (fig. 6). Prior to those separating cohorts, there was a
flat to slightly downward trend.47

VII. Heterogeneity and Mechanisms

The evidence presented above strongly indicates that higher levels of fi-
nancial aid cause individuals to be more likely to obtain a degree, but it does
not explain why. Degree attainment is a product of the decision to enroll
and the decision to persist until the degree is earned. In this context, the
Post-9/11 GI Bill has primarily three effects. First, it reduces the cost of en-
rolling at most institutions. Second, the Post-9/11 GI Bill reduces the rela-
tive cost of institutions with high tuition or, in areas with high costs of liv-
ing, makes those types of institutions more appealing. This may affect degree
attainment if veterans shift their enrollment toward higher-quality institu-
47
Combining this result with the small negative coefficients for the civilian pop-
ulation and the appropriation results suggests another reason that the geographic
variation approach may produce underestimates for the effect on veterans.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 537

tions. Third, the Post-9/11 GI Bill reduces the direct cost of persisting in col-
lege.48

A. Enrollment and Heterogeneity

Having laid out the channels through which the benefit expansion may
increase degree attainment, I first estimate specifications similar to equa-
tion (1) for enrollment. The bottom panel of table 5 presents the standard
difference-in-differences estimate of whether an individual has enrolled in
any program within 3 years of separation. Post-9/11 GI Bill eligibility re-
sults in a 7.7 percentage point increase in the likelihood that a veteran enrolls
within 3 years of separation, an increase of more than 15%.49 The difference-
in-differences estimate implies a substantially larger effect than the enroll-
ment estimates obtained using geographic variation in the size of the benefit
expansion (0.001; table 6), suggesting that veterans may have been unaware
of the variation across geography. Reconciling this with the effects on degree
attainment suggests that veteran enrollment was responsive to the awareness
of the general increase in benefit levels, while degree attainment was respon-
sive to the size of the increase. This implies that veterans were potentially
unaware of the size of the benefit expansion when making their enrollment
decisions but were affected by the additional dollars available once enrolled.
This lack of awareness of the net cost of attending different institutions is
consistent with recent work by Hoxby and Turner (2015) that explores
the effect of information among low-income high-achieving students.
Table 5 also illustrates how the enrollment and degree attainment esti-
mates vary across individuals. While all of the point estimates are statistically
indistinguishable from each other, there is suggestive evidence that the ben-
efit expansion increased the college attainment of high-aptitude individuals
more than that of low-aptitude individuals, with the degree attainment of
high-aptitude individuals increasing by nearly 9 percentage points, almost
double the point estimate for those with AFQT scores below the median.
Interestingly, the enrollment estimates for low and high aptitude are oppo-
site in magnitude from the degree attainment estimates, suggesting that the
higher benefit levels may have pulled some lower-aptitude students into
school, who then failed to graduate.
Does the overall increase in enrollment account for the higher degree at-
tainment rates? Assuming that the rate of degree attainment among marginal
enrollees is similar to other veterans, we can get a back-of-the-envelope es-

48
See apps. A–C for a discussion of how changes in the cost of persisting may
have contributed directly to the increase in degree attainment rates.
49
While Barr (2015) focuses on a somewhat different population and set of out-
comes, the finding of a 15%–20% increase in the likelihood of veterans being en-
rolled at a particular age is consistent with the estimates in table 5.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 538 Barr

Table 5
Difference-in-Differences Demographic Heterogeneity
Full Not Low High
Sample Black Black Male Female Young Old Aptitude Aptitude
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9)
Associate’s
or higher
(6 years) .060*** .058 .064** .062** .019 .061** .001 .046 .087*
(.018) (.050) (.029) (.026) (.103) (.026) (.112) (.028) (.045)
Group
difference
(p-value) .868 .395 .437 .119
Mean .201 .169 .207 .185 .279 .201 .191 .136 .268
Bachelor’s
or higher
(6 years) .045** .038 .050** .045** .006 .046** 2.015 .022 .083**
(.018) (.043) (.024) (.022) (.088) (.022) (.098) (.022) (.040)
Group
difference
(p-value) .724 .304 .330 .220
Mean .131 .115 .134 .121 .179 .131 .137 .0771 .187
Enrollment
(3 years) .077** 2.014 .109*** .090*** 2.084 .074** .02 .097** .055
(.029) (.069) (.036) (.033) (.117) (.033) (.142) (.041) (.052)
Group
difference
(p-value) .014 .153 .470 .345
Mean .460 .450 .462 .446 .534 .465 .377 .406 .517
Observations 15,456 2,529 12,927 13,025 2,431 14,652 804 8,147 7,309
NOTE.—The table shows coefficients from difference-in-differences regressions that compare degree at-
tainment and enrollment levels for eligible and ineligible veterans from before to after the implementation
of the Post-9/11 GI Bill. Each cell presents the estimate from a single regression. Young is defined as being
under the age of 30. The specification also includes indicator variables for sex, whether the person is black,
marital status, high school degree at separation, and month of year of separation as well as Armed Forces
Qualification Test scores, controls for age of separation, baseline degree level, year and state fixed effects,
and the average unemployment calculated for an individual’s home of record during the 12 months after
separation. Robust standard errors clustered at the state level are in parentheses. The sample is restricted
to veterans aged 22–39 with less than a bachelor’s degree who separated between August 1, 2002, and
July 31, 2004 (the preperiod), or between August 1, 2007, and July 31, 2008 (the postperiod), and listed
a US state as a home of record.
* p < .10.
** p < .05.
*** p < .01.

timate of the effect on degree attainment. Multiplying the estimate of mar-

ginal enrollment (.077) by the share of enrolled veterans obtaining a degree
within 6 years of separation suggests that marginal enrollees are responsible
for at most 3 percentage points, or one-half, of the observed increase in at-
tainment.50 This finding suggests that the other half of the degree attainment

50
While it is possible that the marginal enrollees were more likely to finish a de-
gree, this runs counter to most human capital investment models and is inconsistent
with the evidence that the marginal enrollees had lower aptitudes on average.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Table 6
Enrollment within 3 Years: School Choice
For Private School Graduation Graduation Open
Enroll 4-Year 2-Year Public Profit Nonprofit Cost Rate ≥25 Rate ≥40 Admission
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Post911t .065*** .048*** .013 .027*** .025*** .010*** .433*** .015** .008 .008
(.009) (.007) (.008) (.009) (.004) (.003) (.114) (.007) (.005) (.012)
Post911t
Combined
Max .0013* .0008 .0003 .0008 2.0001 .0004** .0195*** .0001 .0004 2.0021
(.0007) (.0009) (.0008) (.0005) (.0003) (.0002) (.0072) (.0005) (.0005) (.0013)
Observations 14,219 14,219 14,219 14,219 14,219 6,324 14,219 14,219 14,219 6,562
Mean .460 .186 .261 .377 .0451 .0247 15.46 .196 .091 .684
NOTE.—The table shows coefficients from regressions that compare the enrollment of veterans from before to after the implementation of the Post-9/11 GI Bill. Each cell presents
the estimate from a single regression. The specification also includes indicator variables for sex, whether the person is black, marital status, high school degree at separation, and month
of year of separation as well as Armed Forces Qualification Test scores, controls for age of separation, baseline degree level, year and state fixed effects, and the average unemployment
calculated for an individual’s home of record during the 12 months after separation. Robust standard errors clustered at the state level are shown in parentheses. The sample is re-
stricted to veterans aged 22–39 with less than a bachelor’s degree who separated between August 1, 2002, and July 31, 2004 (the preperiod), or between August 1, 2007, and July 31,
2008 (the postperiod), and listed a US state as a home of record.
* p < .10.
** p < .05.
*** p < .01.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 540 Barr

effect is coming through the channel of increased persistence, which may

result from increased school quality or the increased level of resources avail-
able to veterans (holding school choice constant). While I cannot pin down
these elements completely, I can provide some evidence as to the contribu-
tions of each.

B. School Choice and Persistence

Recent evidence points to the importance of school quality in influencing
completion rates and adult outcomes (Bound, Lovenheim, and Turner 2010;
Cohodes and Goodman 2014; Cellini and Turner 2016). Because the Post-
9/11 GI Bill reduces the relative cost of more expensive institutions, we
might expect individuals to gravitate toward these types of schools. Follow-
ing this logic, if college quality is positively correlated with cost, then the ex-
pansion in aid may result in quality upgrading that translates into higher de-
gree attainment. For several reasons, though, it is difficult to determine whether
this hypothesis is borne out in the data. First, college quality is very difficult
to measure, particularly outside of the most selective institutions. Second,
disentangling the effect of the benefit expansion on school quality is com-
plicated by the fact that the expansion also affected veteran enrollment. So
if, for example, the marginal group of veteran enrollees has an unobservable
preference for inexpensive schools, it may appear that the average school
cost has decreased when in fact inframarginal enrollees are enrolling in more
expensive schools. The third reason why it is difficult to assess whether the
aid expansion results in higher school quality is that, due to limitations in
the coverage of the NSC data, I am likely to underestimate effects on for-
profit enrollment, which may affect measured school quality. With these ca-
veats in mind, I present estimates of how school choice has changed since
the passage of the Post-9/11 GI Bill (table 6).
The first row presents the simple change in enrollment for eligible veter-
ans from the preperiod to the postperiod. Most of the enrollment increase
occurs at 4-year schools. The increase in for-profit enrollment, while small
in percentage terms (2.5 percentage points), is more than 50%. Column 7
indicates that individuals are enrolling at schools with higher Post-9/11 GI
Bill benefit levels; veterans separating in the postperiod are enrolling in col-
leges with an additional $433 per year in benefits under the Post-9/11 GI Bill
(col. 7). While it appears that veterans are enrolling in more expensive insti-
tutions, columns 8 and 9 indicate that they are not necessarily enrolling in
institutions with higher graduation rates. I interpret these results cautiously,
as the types of veterans enrolled have changed at the same time as the incen-
tives to enroll in different school types.
The second set of results uses the geographic variation in the benefit ex-
pansion. These results are largely consistent with the overall changes in the
first row. A $1,000 increase in maximum benefits in the state results in en-
rollment at a school with, on average, roughly $20 more in annual benefits

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 541

under the Post-9/11 GI Bill (col. 7). The last three columns contain some
suggestive evidence that individuals experiencing increases in maximum
benefit levels are more likely to enroll in schools with higher graduation
rates and less likely to enroll in nonselective institutions. In summary, while
it appears that veterans are shifting toward schools with higher Post-9/11 GI
Bill benefit levels, there is only limited evidence that this has resulted in an
upgrade in school quality. This indicates that something else is likely re-
sponsible for the increase in veteran persistence. Additional analyses sug-
gest that the direct effect of additional aid may play an important role.51

VIII. Discussion and Conclusion

Overall, my results suggest that the Post-9/11 GI Bill increased degree at-
tainment by more than 25%, roughly 0.4 percentage points per $1,000 of
additional maximum aid. Scaling these by an estimate of the average realized
increase in benefits ($5,000) suggests an effect of roughly 1 percentage point
per $1,000 of additional aid. This effect is the result of getting additional vet-
erans into school as well as making inframarginal students (in terms of en-
rollment) more likely to obtain a degree. Although the data show that vet-
erans are attending more expensive schools, there is little evidence that
increased school quality has contributed to higher persistence levels. In-
stead, it appears that the direct effect of the aid has contributed to increased
rates of year-to-year persistence.
Veterans belong to a larger category of older, nontraditional students
who frequently enroll as a way to transition between occupations. I have
shown here that even when these types of individuals have high initial levels
of support (as under the MGIB), offering additional financial aid makes
them more likely to obtain a college degree. This initially high level of aid
may explain why the effects estimated in this paper are substantially smaller
than some estimates on the effects of financial aid on the degree attainment
of traditional students (e.g., Castleman and Long [2016], who find a 4 per-
centage point effect per $1,000) as well as the more substantial set of esti-
mates of the effect of aid on enrollment (Dynarski and Scott-Clayton
[2013], who summarize a 3–4 percentage point effect per $1,000). However,
it is worth noting that prior estimates of the effect of financial aid on the
enrollment of older, nontraditional students indicate a smaller effect (1 per-
centage point per $1,000) for this group (Seftor and Turner 2002). The ef-
fects of additional aid on the college success of the broader group of nontra-
ditional civilian students may be larger than that observed among veterans,
as the latter group already received relatively generous levels of financial aid
prior to the benefit expansion.

51
See apps. A–C for further discussion.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 542 Barr

Given the US government’s enormous expenditures on the Post-9/11 GI

Bill, a natural question is whether this significant aid intervention on behalf
of veterans is a sound investment. Prior estimates of the effect of degree
attainment on earnings and employment suggest that the increase in educa-
tion levels among veterans will help hundreds of thousands of them transi-
tion into the workforce. Whether the benefit expansion is a beneficial in-
vestment for society, however, is another question. We can produce a
back-of-the-envelope estimate of the direct cost of each additional degree
by estimating the increase in benefits paid out and dividing by the number
of additional degrees obtained. These calculations suggest costs between
$75,000 and $150,000.52 Higher than analogous estimates from some finan-
cial aid interventions, this cost per degree is substantially smaller than net
present value estimates of the return to a college degree.53 Future work will
explore the degree to which prior research on the return to education gen-
eralizes to post-9/11 veterans.

References
Angrist, Joshua D. 1990. Lifetime earnings and the Vietnam era draft lot-
tery: Evidence from Social Security administrative records. American
Economic Review 80, no. 3:313–36.
———. 1993. The effect of veterans benefits on education and earnings. In-
dustrial and Labor Relations Review 10:637–52.
Angrist, Joshua D., and Stacey H. Chen. 2011. Schooling and the Vietnam-
era GI Bill: Evidence from the draft lottery. American Economic Journal:
Applied Economics 3, no. 2:96–118.
Avery, Christopher, and Sarah Turner. 2012. Student loans: Do college stu-
dents borrow too much or not enough? Journal of Economic Perspectives
26, no. 1:165–92.
Barr, Andrew. 2015. From the battlefield to the schoolyard: The short-
term impact of the Post-9/11 GI Bill. Journal of Human Resources 50,
no. 3:580–613.
Barr, Andrew, and Sarah Turner. 2015. Out of work and into school: Labor
market policies and college enrollment during the Great Recession. Jour-
nal of Public Economics 124:63–73.
———. 2017. A letter and encouragement: Does information increase post-
secondary enrollment of UI recipients? Technical report, National Bu-
reau of Economic Research.

52
See app. B for details on these calculations. These calculations ignore any ben-
efits or costs that might accrue to society through the use of the GI Bill as a recruit-
ing tool.
53
See Oreopoulos and Petronijevic (2013) and Avery and Turner (2012) for
overviews of these net present value calculations.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 Fighting for Education 543

Borgschulte, Mark, and Paco Martorell. 2016. Paying to avoid recession:

Using reenlistment to estimate the cost of unemployment. Technical re-
port, Institute for the Study of Labor.
Bound, John, Michael F. Lovenheim, and Sarah Turner. 2010. Why have
college completion rates declined? An analysis of changing student prep-
aration and collegiate resources. American Economic Journal: Applied
Economics 2, no. 3:129–57.
Bound, John, and Sarah Turner. 2002. Going to war and going to college:
Did World War II and the GI Bill increase educational attainment for re-
turning veterans? Journal of Labor Economics 20, no. 4:784–815.
Burrell, Sidney A. 1967. The GI Bill and the great American transformation.
Boston University Graduate Journal 15, no. 3.
Castleman, Benjamin L., and Bridget Terry Long. 2016. Looking beyond
enrollment: The causal effect of need-based grants on college access, per-
sistence, and graduation. Journal of Labor Economics 34, no. 3:1023–73.
Cellini, Stephanie Riegg, and Nicholas Turner. 2016. Gainfully employed?
Assessing the employment and earnings of for-profit college students us-
ing administrative data. Technical report, National Bureau of Economic
Research.
Cohodes, Sarah R., and Joshua S. Goodman. 2014. Merit aid, college qual-
ity, and college completion: Massachusetts’ Adams scholarship as an in-
kind subsidy. American Economic Journal: Applied Economics 6, no. 4:251–
85.
Denning, Jeffrey T. 2017. Born under a lucky star: Financial aid, college
completion, labor supply, and credit constraints. Working Paper no. 17-
267, Upjohn Institute, Kalamazoo, MI.
Dynarski, Susan. 2008. Building the stock of college-educated labor. Jour-
nal of Human Resources 43, no. 3:576–610.
Dynarski, Susan, and Judith Scott-Clayton. 2013. Financial aid policy: Les-
sons from research. Future of Children 23, no. 1:67–92.
Dynarski, Susan, Steven Hemelt, and Joshua Hyman. 2015. The missing
manual: Using National Student Clearinghouse data to track postsecond-
ary outcomes. Educational Evaluation and Policy Analysis 37, suppl. 1:
53S–79S.
Ferraro, Paul J., and Juan José Miranda. 2014. The performance of non-
experimental designs in the evaluation of environmental programs: A
design-replication study using a large-scale randomized experiment as a
benchmark. Journal of Economic Behavior and Organization 107:344–65.
Goldrick-Rab, Sara, Douglas N. Harris, Robert Kelchen, and James Ben-
son. 2012. Need-based financial aid and college persistence: Experimental
evidence from Wisconsin. Institute for Education Sciences.
Harkin, Tom. 2014. “Is the new G.I. Bill working? For-profit colleges in-
creasing veteran enrollment and federal funds. Technical report, Health,
Education, Labor, and Pension Committee of the United States Senate.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]
 544 Barr

Hoxby, Caroline M., and Sarah Turner. 2015. What high-achieving low-
income students know about college. American Economic Review 105,
no. 5:514.
Lemieux, Thomas, and David Card. 2001. Education, earnings, and the
“Canadian G.I. Bill.” Canadian Journal of Economics 34, no. 2:313–44.
Lovenheim, Michael F., and Emily G. Owens. 2014. Does federal financial
aid affect college enrollment? Evidence from drug offenders and the
Higher Education Act of 1998. Journal of Urban Economics 81:1–13.
Murphy, Richard, and Gill Wyness. 2016. Testing means-tested aid. Tech-
nical report.
Oreopoulos, Philip, and Uros Petronijevic. 2013. Making college worth
it: A review of the returns to higher education. Future of Children 23,
no. 1:41–65.
Scott-Clayton, Judith. 2011. On money and motivation a quasi-experimental
analysis of financial incentives for college achievement. Journal of Hu-
man Resources 46, no. 3:614–46.
Seftor, Neil S., and Sarah E. Turner. 2002. Back to school: Federal student
aid policy and adult college enrollment. Journal of Human Resources 37,
no. 2:336–52.
Simon, Curtis J., Sebastian Negrusa, and John T. Warner. 2010. Educational
benefits and military service: An analysis of enlistment, reenlistment, and
veterans’ benefit usage 1991–2005. Economic Inquiry 48, no. 4:1008–31.
Sjoquist, David L., and John V. Winters. 2015. State merit-based finan-
cial aid programs and college attainment. Journal of Regional Science 55,
no. 3:364–90.
Smith, Jeffrey A., and Petra E. Todd. 2005. Does matching overcome
LaLonde’s critique of nonexperimental estimators? Journal of Econo-
metrics 125, no. 1:305–53.
Stanley, Marcus. 2003. College education and the midcentury GI Bills.
Quarterly Journal of Economics 118, no. 2:671–708.

This content downloaded from [Link] on May 15, 2019 17:09:24 PM

All use subject to University of Chicago Press Terms and Conditions ([Link]