Content uploaded by Christopher M. Cornwell
Author content
All content in this area was uploaded by Christopher M. Cornwell
Content may be subject to copyright.
MERIT-BASED COLLEGE SCHOLARSHIPS
AND CAR SALES
Christopher Cornwell∗
cornwl@terry.uga.edu
and
David B. Mustard
mustard@terry.uga.edu
Department of Economics
University of Georgia
Athens, GA 30602
December 11, 2006
∗Cornwell and Mustard gratefully acknowledge the support of the NSF under grant SES-9986469, and
the Terry College of Business through its Terry-Sanford Research Grant program. An earlier version of
this paper was presented at the 2004 AEFA meetings. We are grateful to Quentin Mostoller and Kyung
Hee Lee for their helpful research assistance.
Abstract
Since the early 1990s, state governments have distributed billions of dollars in fi-
nancial aid through merit-based college scholarships, most of which have no means tests.
The model for most of these programs is Georgia’s “Helping Outstanding Pupils Educa-
tionally” (HOPE) Scholarship. Given the high correlation between pre-college academic
achievement and family income, the program characteristics raise the question: to what
extent are HOPE disbursements simply rent payments to households otherwise inclined
to send their children to college? This paper addresses the rent question by examining
the effect of HOPE on automobile consumption. The relatively swift passage of the lot-
tery law and establishment of the program created an unanticipated windfall large enough
to encourage the financing of consumer durables purchases, such as automobiles, out of
household savings targeted for college. First, we compare car registrations in Georgia with
those in sets of control group states before and after HOPE. We do not find a statistically
significant overall HOPE effect, but allowing the HOPE coefficient to vary by year reveals
statistically significant percentage increases in registered vehicles in 1994 and 1995, when
the program’s income cap was raised and then removed. Next, we examine the relationship
between car registrations and HOPE recipients by county. Our results indicate that the
number of HOPE recipients attending degree-granting institutions increases car registra-
tions in counties above the 75th percentile in per-capita income; there is no evidence of a
relationship in counties below the 25th per-capita income percentile.
1 Introduction
College financial aid programs have traditionally focused on need to increase access to
higher education for low-income students and expand college choice by enlarging the set
of affordable institutions. Until the late 1980s, a relatively small fraction of total student
aid was allocated on the basis of merit, and most of it related to individual institutions’
attempts to attract academically proficient students. However, since the early 1990s, state
governments have distributed billions of dollars in financial aid through newly established,
merit-based college scholarships, most of which have no means tests. A common justifica-
tion for these actions is to induce the state’s best high-school students to remain home for
their college educations. The model for these programs is Georgia’s “Helping Outstanding
Pupils Educationally” (HOPE) Scholarship.1
Initiated in 1993, Georgia’s HOPE Scholarship covers tuition, fees and book expenses
for all eligible high-school graduates attending Georgia public, post-secondary institutions.
The award value has accounted for more than 40 percent of the total cost of attendance
at the state’s top public universities, amounting to $5264 in the 2005-06 academic year.
Eligible students who attend in-state private institutions receive a fixed payment of $3000.
To qualify for the scholarship, a high-school student must graduate with a “B” average
and be a Georgia resident. In the first year of the program, a household income cap
of $66,000 was imposed, which was raised to $100,000 the following year and eliminated
entirely thereafter. Since 1995, there have been no income restrictions. From September
1993 through November 2006, almost $3.6 billion in scholarship funds have been disbursed
1The list of states that have adopted HOPE-style scholarships has grown to 15, including Georgia’s
neighbors, Florida, South Carolina and Tennessee. See Cornwell, Leidner, and Mustard (2006) for details.
1
to over 950,000 students. HOPE is financed through a state lottery established under the
Georgia Lottery for Education Act in 1992.
Given the strong correlation between academic merit and household income, we ex-
pect that the vast majority of scholarship dollars go to households with students who
would attend college even in the absence of HOPE. Support for this contention is pro-
vided by Cornwell, Mustard, and Sridhar (2006) (hereafter CMS), who contrast first-time
freshmen enrollment rates in Georgia with the other member states of the Southern Re-
gional Education Board (SREB) over the 1988–97 period. Using data from the Integrated
Postsecondary Education Data System (IPEDS) administered by the National Center for
Education Statistics, CMS find that HOPE increased the overall enrollment in Georgia
colleges and universities by almost about 6 percent relative to the rest of the SREB, with
the gains concentrated in 4-year schools. Based on the IPEDS residency and migration
data on freshmen recently graduated from high school, they estimate that about two-thirds
of the HOPE effect on enrollments at 4-year colleges can be attributed to a decrease in
residents leaving the state.2The 6-percent overall enrollment effect represents only 15
percent of all freshmen scholarship recipients. Thus, the HOPE program has operated
largely as a transfer to students who would have enrolled in college anyway, although its
relative price effects have influenced where students attend.3
The HOPE rent and the consumption it facilitates motivates this paper. Where do
the HOPE Scholarship dollars go? Infra-marginal HOPE payments are obviously available
for consumption in general. However, it is certainly plausible that the scholarship leads
2Recent-graduate freshmen—students who graduated high school in the previous twelve months ac-
counted for 78 percent of all first-time freshmen at Georgia’s 4-year institutions.
3Dynarski (2000) reports qualitatively similar findings based on a similar empirical strategy applied
to college attendance data on Current Population Survey respondents. Although she does not examine
HOPE’s effect by institution type, she concludes that at least 80 percent of award dollars are allocated to
students who would attend college anyway.
2
to automobile purchases using household savings otherwise targeted for college expenses,
because the scholarship accounts for such a large fraction of college costs. This is probably
more true in households that benefited from the program in the early years, for whom
the swift passage of the lottery law and establishment of the HOPE program was an
exogenous, unanticipated event. For those households, HOPE was like a transitory income
shock, which often show up in durables consumption, and especially in automobile sales
as they account for the largest share of durables spending (Lam (1991)).
For our purposes, an ideal empirical setting would provide individual-level consump-
tion data on households with potential college enrollees. The Consumer Expenditure
Survey and the Panel Study of Income Dynamics provide such data, but their coverage
is too thin to permit inference about the effect of HOPE. As alternatives, we pursue two
empirical strategies: one analyzes state-level differences in differences in car registrations,
and another relates county-level automobile registrations to the the number of HOPE re-
cipients in each county. Both sets of results suggest a link between HOPE and automobile
consumption.
First, we treat HOPE as a natural experiment, comparing the log of car registrations in
Georgia before and after the introduction of the program with those in the member states
of the Southern Regional Education Board (SREB), between 1988 and 1997 (the same
time period examined by CMS). The estimated overall HOPE effect is .006 or .6 percent,
but it is very imprecise. Whether we exclude all but the states that border Georgia or
expand the control group to include the entire US, the results are never precise enough to
infer statistical significance. There is a more evidence of HOPE’s influence on registrations
when we allow the scholarship’s effect to vary by period. We find that registrations are
3
higher in 1994 and 1995 (the years coinciding with the elimination of the income cap) than
in any pre-HOPE period.
Second, we relate county-level panel data on car registrations to HOPE awards by
institution type, covering the 1993–2001 period. Knowing the type of institution award
recipients attend allows us to distinguish merit scholarship winners from students receiving
the non-merit-based HOPE Grant. Further, we examine the registration-award relation-
ship separately for groups of counties that are above the 75th and below the 25th percentiles
in per-capita income. For students from upper-income counties attending state-system and
private colleges, we estimate awards elasticities of about .05, although neither is significant
at the 5 percent level.
2 Georgia’s HOPE Program
Georgia’s HOPE Scholarship is the largest state-financed, merit-based aid program
in the US. By 1997, total non-need aid awarded by Georgia was greater than that of the
other fourteen SREB states combined.4By 1999, the size and scope of the HOPE program
exceeded that of the federal Pell Grant in Georgia by about a factor of two. Since 2002,
Georgia has distributed more financial aid per full-time equivalent student that any other
state.
HOPE awards can be used at 103 institutions in Georgia, each of which can be cate-
gorized as either a state-system, private or technical school. There are 34 degree-granting,
state-system schools (20 four-year and 14 two-year); 25 degree-granting, private colleges
4See the National Association of State Scholarship and Grant Aid Programs 19th Annual Survey Report,
Academic Year 1987-88 and 29th Annual Survey Report, Academic Year 1997-98 . Georgia’s total 1998
aid is 55 percent higher than that of the second-ranked state, Florida.
4
and universities (20 four-year and 5 two-year); and 34 technical schools (all of which are
2- or less-than-two-year) specializing in certificates and diplomas.5
2.1 Program Rules and Awards
There are two separate components of the HOPE program, the merit-based scholarship
and the HOPE Grant. Eligibility for the former depends on a student’s high-school grade-
point average, while the latter applies only to non-degree programs at two-year and less-
than-two-year schools and has no merit requirements. Thus, the incentives related to merit
aid are limited to students with degree objectives. Neither award is restricted by family
income.
To qualify for the scholarship, an entering freshman must have graduated from an
eligible Georgia high school since 1993 with at least a “B” average and be a Georgia
resident.6To retain a HOPE Scholarship a student must maintain a 3.0 GPA at each
of three checkpoints.7In contrast, the HOPE Grant is an entitlement and there are no
restrictions based on when a student graduated from high school. The grant covers tuition
and HOPE-approved mandatory fees for all coursework leading to a certificate or diploma.
Because technical school tuition is $25 per credit hour—just a fraction of that charged
by state-system colleges—the value of the grant is considerably smaller than that of the
5“Less-than-two-year” schools only grant certificates and diplomas requiring less than two academic
years of coursework. In general, they are “technical” schools, but a few with the “technical” label offer
two-year degrees. In addition, 4 two-year, state-system schools offer diplomas and certificates.
6To reduce the number of HOPE Scholars and avoid future funding shortages, the eligibility rules were
tightened for students who graduated high school in 2000 to demand a “B” average in “core-curriculum”
subjects. Interestingly, the predicted 35 percent drop in HOPE qualifiers did not materialize. The number
of HOPE recipients declined only 4.3 percent from the previous year, raising the question of whether
grades were inflated in reaction to the stiffer requirements (“Hope Suffers Funding Shortage,” Athens
Banner Herald, 30 Sep 2000).
7The GPA checkpoints occur at the end of one’s freshmen, sophomore, and junior years, which cor-
respond to 30, 60, and 90 hours under a semester system. Those who do not qualify for HOPE in high
school can become eligible at each checkpoint if their GPAs are at least 3.0.
5
scholarship. Thus, we do not expect to find that car registrations are related to the number
of grant recipients.
2.2 Award Distribution
The Georgia Student Finance Commission reports the number of HOPE awards by the
school type HOPE qualifiers attend: state system (two- and four-year public colleges and
universities), private (two- and four-year private colleges and universities) and technical
(which specialize in diploma and certificate programs). HOPE Scholars populate institu-
tions in the state-system and private categories, while grant recipients enroll in technical
schools. The distribution of HOPE recipients and disbursements by award and institution
type through 2002 are summarized in Table 1. Over the sample period, state system insti-
tutions claimed 41 percent of awards and almost 70 percent of aid dollars. Together, public
and private degree-granting colleges and universities accounted for 55 percent of awards
and just under 80 percent of disbursements. Given the differences in award value between
the scholarship and grant, the overwhelming majority of HOPE aid goes to scholarship
winners.
Obscured in the cumulative statistics presented in Table 1 are the trends in scholarship
and grant allocations. In terms of both the number of recipients and dollars disbursed, the
scholarship grew much more rapidly than the grant through 2002. The primary factor in
this growth has been the rise in the fraction of high-school graduates and the percentage
who satisfy the merit requirements.
If, as calculated by CMS, upwards of 85 percent of scholarship expenditures is rent, the
data in Table 1 indicate a large source of funds for expanding the consumption possibilities
of households with HOPE Scholars. In addition, students whose behavior is affected by
6
the scholarship may share household resources that would have been used to finance out-
of-state college costs.
2.3 Award Capitalization
For high-school students entering the 1992-93 academic year, HOPE was almost en-
tirely unanticipated. When then-Governor Zell Miller proposed the scholarship, its lottery
funding scheme required an amendment to the state’s constitution, which had been voted
down in the past. The amendment did pass, but not until November 1992, and the fi-
nal vote was quite close. Initially, there was an income cap of $66,000, but it was raised
to $100,000 in 1994 and eliminated in 1995. The removal of the cap caused the number
of HOPE Scholars to rise dramatically. Overall, freshmen scholarship recipients almost
tripled from 1993 to 1995. Furthermore, the expansion in the program targeted high-
income households, which would be much more likely to use HOPE to divert planned
educational expenditures into automobile purchases. Thus, if HOPE is capitalized in au-
tomobile purchases, the evidence should be relatively strong in 1994 and 1995. The effect
of the program should diminish over time, as households come to expect the benefits and
factor them into household allocation decisions earlier and in different ways. One obvious
possibility is that HOPE may reduce household savings for college. We test this hypothesis
with the state-level data by allowing the HOPE effect to vary by year.
HOPE reduces the relative price of remaining in state for college. The influence of
this relative price change comes up often in our conversations with Georgia residents who
are students at the University of Georgia (UGA), virtually all of whom have entered with
the scholarship since the income cap was dropped in 1995. A remarkably common theme is
“I chose UGA over my best out-of-state alternative because my parents agreed to buy me
7
a car.” For parents, the cost of college is reduced substantially, from potentially in excess
of $100,000 over four years to an annual room and board expenditure of roughly $4000.
The intra-family bargain over this difference is likely to occur more frequently in higher
income households that have the resources to send their children out of state. In lower
income households, HOPE’s effect on the relative price of a 4-year school is probably more
important. By lowering the relative price of a 4-year college, HOPE affords a student the
opportunity to attend a higher quality 4-year school away from home instead of a 2-year
school close by. Such a decision may not entail a diversion of educational expenditures into
other consumption, and the diversion may even go the other way. For example, a household
may forgo a car purchase (among other things) to facilitate a student’s matriculation at
a higher quality institution or a school that is farther from home. Therefore, we expect
the relationship between car purchases and HOPE receipt to vary by income. To test this
hypothesis, we separately examine the relationship between car registrations and HOPE
awards in counties above the 75th and below the 25th percentiles in the per-capita income.
3 Evidence from Cross-State Car Registrations
3.1 Empirical Model
Our first attempt to connect the HOPE rent to automobile consumption exploits the
natural experiment aspect of the program’s introduction in 1993. Specifically, we contrast
car registrations in Georgia before and after the HOPE “treatment” with those in sets
of control-group states. We implement this strategy in a regression framework with the
following form:
ln Rit =α+βtYt+γi1Si+γi2(Si×t) + δ(SGA ×Ht) + X0
itξ+²it ,(1)
8
where Rit is the number of privately registered automobiles in state iin year t(t=
1988, . . . , 1997), Ytis a dummy variable for year t,Siis a dummy variable for state i,SGA
is a dummy variable for Georgia, Htis a HOPE indicator, equal to 1 when t≥1993 and 0
otherwise, Xit is a vector of covariates, and ²it is a random error. The interaction of Siwith
taccounts for state-specific trends in registrations. Included in Xit are measures of state
income, population and employment in the automobile industry (each in logs) to account
for time-varying state differences that could be correlated with the establishment of the
program.8The motivation for the latter comes from the discounts available to employees
of car manufacturers and the opening of several large plants in the southeast during the
early and mid-1990s.
Our focus is on the coefficient of the interaction between Htand SGA,δ, which captures
the difference in differences between ln Rit in Georgia and the control-group states over
the pre- and post-HOPE periods. We are also interested in the timing of the HOPE effect,
expecting its influence to be greater in the years immediately following the removal of the
income cap. So, we also estimate a version of (1) that allows the HOPE effect to depend
on t:
ln Rit =α+βtYt+γi1Si+γi2(Si×t) + δGA,t(SGA ×Yt) + X0
itξ+²it .(2)
The pattern exhibited in the estimated δGA,t will provide some evidence on whether the
overall difference-in-differences estimate can be temporally disaggregated in a manner con-
sistent with the start of the program. We estimate both (1) and (2) by ordinary least
squares and report t-ratios that are heteroscedasticity and autocorrelation consistent (fol-
lowing Arellano (1987) and Bertrand, et al. (2004)).
8Preferable to state population would be the number of registered drivers. However, a consistent
registered-driver series is not available until 1990. In any event, using registered drivers instead of popu-
lation has virtually no impact on the HOPE effects estimated over the 1990–97 period.
9
3.2 Data
Data on motor-vehicle registrations by state are available from the Office of Highway
Policy Information of the Federal Highway Administration (FHWA).9As described by
the FHWA, these data are organized by major vehicle class (automobiles, buses, trucks,
and motorcycles) and ownership (private/commerical or public). We measure Rit as to-
tal private/commerical, automobile registrations, ignoring the truck category since it is
dominated by commercial registrations that are not likely influenced by a HOPE-style
intervention.10 Sources of duplicate registrations, such as transfers, have been eliminated
from the vehicle counts to the extent possible. While the annual vehicle registration date
varies by state, the information is reported by the FHWA on a calendar-year basis. The
Bureau of Economic Analysis (BEA) provided the measures of personal income for state-
level auto industry employment. State population data were obtained from the US Census
Bureau.
We limit the sample period to 1988-97 for three reasons. One is so we can relate
our findings more easily to the enrollment results of CMS. The second is that extending
the analysis beyond 1997 severely limits the number of states that can serve as members
of a control group, as other HOPE-style programs appeared in the late 1990s.11 Finally,
our ability to observe a HOPE effect will likely diminish as the program ages, because
households will adjust to factor expected scholarship benefits in consumption decisions
9The data for each year since 1992 can be obtained from http://www.fhwa.dot.gov/ohim/ohimstat.htm.
10 Only since 1993 has the truck category distinguished pickups, vans and sport utility vehicles (SUVs)
from truck tractors, farm vehicles and other light trucks, so it is not possible to extract registrations of
these vehicle types for the pre-HOPE period.
11 The only such program is Florida’s Bright Futures Scholarship, which was modeled directly after HOPE
and initiated in last year of our sample. Two years prior to HOPE, Arkansas introduced its Academic
Challenge Scholarship. However, its scope is much smaller: the benefits are limited to $2500 per year and
to households with incomes less than $50,000, while maintaining similar eligibility requirements. Excluding
these states from the analysis has virtually no affect on our findings.
10
earlier and in different ways. Similarly to CMS, we focus our attention on a control group
made up of the other member states of the SREB.12 However, we also consider contrasts
with all other US states and the states that border Georgia.
Georgia’s automobile registration series is compared to that of each SREB state in
Figures 1a–n. Each figure plots log automobile registrations purged of state and time
effects; that is, residuals from regressions of log registrations on state and time dummies.
In all but a few cases (Virginia and West Virginia), marked differences (especially in 1994
and 1995) appear between Georgia and the comparison state in the post-HOPE period,
While the figures are suggestive, they do not account for state-specific trends which could
be correlated with the timing of HOPE.
3.3 Results
Table 2 presents the summary statistics for the variables used in our state-level anal-
ysis, over pre- and post-HOPE periods for Georgia and the SREB control group. As a
preliminary observation, we note that Georgia has outpaced the rest of the SREB states in
registration, population and personal income growth over the sample period. We begin the
formal analysis by observing the simple percentage difference in differences in automobile
registrations between Georgia and the other SREB states taken as a whole is about 9.5.
Taken at face value, this suggests that car registrations were almost 10 percent higher in
Georgia between 1993 and 1997 because of HOPE. An effect of this magnitude is clearly
implausible, as it implies an annual increase of well almost 400,000 registered vehicles, a
figure that is difficult to reconcile with the recipient data in Table 1.
12 There are 16 members of the SREB: Alabama, Arkansas, Delaware, Florida, Georgia, Kentucky,
Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia
and West Virginia. Delaware only recently joined the SREB, and therefore was not included in our sample.
11
Table 3 reports the estimated HOPE effects from equations (1) and (2), which control
for state and year fixed effects, state-specific trends, and the income, population and
automobile employment covariates, for three sets of control-group states. The first row
provides the overall estimates and the rows beneath the time-varying results (with 1993
set as the base year). Focusing on the SREB case, we find an overall estimated HOPE
effect of .006 or about .6 percent. An effect of this magnitude suggests a somewhat more
plausible annual registrations increase in the range of 25,000. However, the estimate is
very imprecise, with at tratio of only .22. The results based on the other two control
groups are similarly imprecise.
In contrast, the time-varying HOPE effect estimates point to statistically significant
percentage increases in registered vehicles in 1994 and 1995, the years when the income cap
on the scholarship was raised and then removed. As we pointed out earlier, the elimination
of the cap greatly increased the number of HOPE Scholars, with the new recipients coming
from high-income households, which would be much more likely to capitalize their awards
into car purchases. Still the effects are large enough to invite skepticism. In the SREB case,
1994 and 1995 coefficient estimates are at least 5 percentage points larger than the pre-
HOPE ˆ
δGA,ts, all but one of which is statistically insignificant. After 1995, the estimated
HOPE effects diminish and are considerably less precise. This pattern is illustrated in
Figure 2, which plots the ˆ
δGA,ts and their corresponding 95 percent confidence band. The
US and border-state cases repeat the pattern with similar point estimates in each period.
4 Evidence from Georgia Counties
Now we consider the relationship between HOPE incidence and car consumption
across Georgia counties. An advantage of the county-level data is the ability to distin-
guish awards by the type of institution in which a recipient enrolls.
12
4.1 Empirical Model
Here the empirical strategy does not involve a natural experiment; instead, we examine
the county-level relationship between HOPE incidence and registered vehicles since the
program’s introduction. This is possible, because in Georgia the county of registration
is the county of permanent residence. HOPE awards at the county level are recorded in
the same way. Thus, a student’s car will be registered, and her receipt of a HOPE award
recorded, in her home county. Her car will not be registered in the county where she
attends college, if it differs from her home county.
The empirical framework for this analysis is a regression model of the form:
ln Rit =α0+αtYt+γi1Ci+γi2(Ci×t) + β1ln Ait +X0
itβ2+²it ,(3)
where Rit is the total number of privately registered automobiles in county iin year t
(t= 1993, . . . , 2001), Ytis defined as in (1) and (2), Ciis a dummy variable for county
i,Ait is the number of HOPE awards (recipients) in a county in a given year, Xit is
a vector of control variables, and ²it is a random error. County trends are accounted
for in the interaction of Ciwith t. Included in Xit are county income and population
variables. We estimate equation (3) for each institution class, first for all counties and
then for groups of counties that are relatively resource rich and resource poor. Isolating
rich and poor counties allows us to roughly distinguish upper-income from lower-income
award recipients. Again, we report t-ratios based on standard errors that are robust to
heteroscedasticity and autocorrelation. OLS applied to (3) will yield a consistent estimate
of the reduced-form effect of award incidence on car registrations as long as Ait and the
variables in Xit are strictly exogenous with respect to ²it.13
13 We actually estimate (3) by applying the fixed-effects estimator to data that have been first-differenced,
which allows us to account for county trends without directly estimating 158 extra parameters.
13
4.2 Data
The Georgia Department of Revenue publishes its annual Statistical Report data on
motor-vehicle registrations by county, separating vehicles into several categories: cars,
trucks, motorcycles, trailers and buses. As in the state-level analysis, we focus on car
registrations, recording the number of registered cars in each county from 1993 to 2001.14
To match the registrations panel, we obtained HOPE award data by county for the aca-
demic years 1993–4 to 2001–02 from the Georgia Student Finance Commissions (GSFC).
The GSFC breaks down the total number of awards by the school type HOPE qualifiers
attend. As described in section 2, state-system institutions include all public, 2- and 4-
year, degree-granting schools; the private-school category includes all 2- and 4-year private
colleges and universities in the state; and technical schools are 2-year institutions that
specialize in diploma and certificate programs. Measures of county-level personal income
and population come from the Regional Economic Information System (REIS) published
by the BEA. Table 4 presents summary statistics for the registration, award and control-
variable data, for all Georgia counties and those above the 75th percentile and below the
25th percentile in per-capita income (PCI).
There are three important facts about the award data. First, over the sample period,
the awards to students attending state-system schools rose by a factor of eight, while the
numbers flowing to the other two school categories increased much less less rapidly. This
is consistent with the increasing relative importance of the HOPE Scholarship discussed
in section 2. Second, the number of private-college awards fell in the 1996 academic year,
when the merit requirements for eligibility were imposed, but its value was raised to $3000.
Prior to 1996, the merit rules did not apply to HOPE awards designated for private schools.
14 In contrast to the national registration data, the definition of a “car” is uniform across counties and
encompasses all passenger vehicles (see http://www.dmvs.ga.gov/motor/stats/stats.asp).
14
Third, students from counties above the 75th percentile in per-capita income received 12.5
times more scholarships to state-system institutions and 13.3 times more scholarships to
private colleges than their counterparts in counties below the 25th percentile, but only
4.3 times more awards to technical schools. By comparison, the average population of the
richer counties is 9.8 times greater and these counties have 11.5 times more registered cars.
4.3 Results
Table 5 presents the results from estimating equation (3) for each institution class
using all Georgia counties. Each column reports the estimated award elasticities with
those of the population and income controls. Only in the case of private-schools is the
elasticity estimate positive (.008), but its tratio is well below 1. The state-system and
technical school estimates are less than .01 in absolute value and even less precise. Thus,
there is no evidence of a relationship between HOPE awards car registrations in the sample
of all counties.
However, as we have argued, the capitalization of the scholarship into car purchases is
more likely in upper-income than lower-income households. We distinguish between these
two groups by estimating our empirical model separately for resource-rich and poor coun-
ties. We do this by isolating the counties above and below the 75th and 25th percentiles
in per-capita income. We define the percentiles to refer to per-capita income in 1993.
First, consider the results for counties with per-capita income above the 75th percentile
reported in Table 6. The estimated state-system award elasticity is .045 with a t-ratio of
1.4, indicating statistical significance at the .17 level. The estimate in the private school
case is about the same size and has a tratio of almost 2. In fact, these are the only two
coefficient estimates that approach statistical significance in the entire table. The effects
15
of population and income are partialed out with the county trends. The technical-school
estimate is smaller and negative, and not statistically significant at even the .4 level.
The magnitudes of the state-system and private school imply a greater than one-
to-one correspondence between awards and registrations, which seems implausible. One
possible explanation is feedback from registrations to awards in high per-capita income
counties, violating strict exogeneity. However, following Wooldridge (2002), we check the
stict exogeneity of log registrations by adding lnAi,t+1 to (3) and testing its significance. In
every case, the corresponding tratio is less than 1. Still this does rule out contemporaneous
county factors beyond population and income, and not controlled for through state trends,
that explain both awards and registrations.15
In any event, the story is very different for lower-income counties. Table 7 shows that
registrations do not significantly vary with awards of any type in counties below the 25th
percentile in per-capita income. No estimated elasticity has a tratio bigger than 1.1.
At least in a qualitative sense, the county-level findings support the existence of a
HOPE rent, accruing to HOPE Scholarship winners from resource-rich counties matricu-
lating at state-system and private colleges and universities. The source of the HOPE effect
on car registrations is likely not restricted to car purchases for (or by) HOPE Scholars,
since the receipt of a HOPE award may present an opportunity in some households to
redistribute savings targeted for college to other household members.
15 Controlling for the age distribution and racial composition of the population, poverty levels, and
unemployment rates has no effect on the estimates.
16
5 Conclusion
Since the introduction of Georgia’s HOPE program in 1993, state-sponsored merit-
based college scholarships have proliferated. This has occurred against the backdrop of
a general trend toward merit-based financial aid at the individual institution level. Since
household income is an important determinant of a high-school student’s academic achieve-
ment, scholarship funding generally flows to those who would have attended college any-
way. Analyzing Georgia’s program over the 1993–97 period, Cornwell, Mustard and Srid-
har (2006) find enrollment rates in Georgia institutions did rise, mostly due to HOPE’s
incentive to remain in state, but at least 85 percent of scholarship expenditures were rent.
In this paper, we investigated where the HOPE rent goes—in particular, whether it
is capitalized in car purchases. We focused on automobile demand because it is sensitive
to transitory income changes, like HOPE generated, especially in the early years of the
program. The value of the scholarship makes it plausible that its receipt could induce a
reallocation of college savings to the purchase of a car. This is particularly true for higher
income households.
Our findings provide some qualified support for the capitalization story. First, using
state-level panel data on car registrations between 1988 and 1997, we treated HOPE as a
natural experiment, contrasting registrations in Georgia before and after the introduction
of the program with those in sets of control-group states. Although we do not find a
statistically significant overall HOPE effect, allowing the HOPE coefficient to vary by year
reveals statistically significant percentage increases in registered vehicles in 1994 and 1995,
when the income cap was raised and then removed. After 1995, the estimated coefficients
decrease in magnitude and become statistically insignificant, as we might expect.
17
Second, we related county-level car ownership to HOPE incidence by institution type,
exploiting panel data on county registrations and HOPE awards over the 1993–2001 pe-
riod. Distinguishing HOPE recipients by the type of institution they attend allowed us
to separately examine the influence of the scholarship from that of the grant. We show
that registrations increase with the number of HOPE recipients in counties above the 75th
percentile in per-capita income attending state-system and private schools. The estimated
awards elasticity is about .05 in both cases. We never find a statistically significant effect
for HOPE recipients attending technical school, consistent with the low value of the HOPE
Grant and the characteristics of non-degree students.
Finally, the evidence we present should be interpreted cautiously. Where the data
point to a scholarship effect, the implied increase in car registrations is too large to be
accounted for simply by car purchases for college students.
18
References
Arellano, M. (1987), “Computing Robust Standard Errors for Within-Groups Estimators”,
Oxford Bulletin of Economics and Statistics 49(4), 431-434.
Bertrand, M., E. Duflo, and S. Mullainathan, “How Much Should We Trust Differences-
in-Differences Estimates?”, Quarterly Journal of Economics 119 (2004) 249-275.
Cornwell, C., M. Leidner and D. Mustard, “Rules, Incentives and Policy Implications of
Large-Scale Merit-Aid Programs”, University of Georgia Economics Department
working paper (2006).
Cornwell, C., D. Mustard and D. Sridhar, “The Enrollment Effects of Merit-Based Finan-
cial Aid: Evidence from Georgia’s HOPE Scholarship,” Journal of Labor Eco-
nomics 24 (2006) 761-786.
Dynarski, S., “HOPE for Whom? Financial Aid for the Middle Class and Its Impact on
College Attendance,” National Tax Journal (2000) 629-661.
Lam, P. (1991), “Permanent Income, Liquidity, and Adjustments of Automobile Stocks:
Evidence from Panel Data,” Quarterly Journal of Economics (1991) 203–230.
Wooldridge, J. (2002), Econometric Analysis of Cross-Section and Panel Data, Cambridge:
MIT Press.
19
Table 1
HOPE Awards
Recipients and Aid Amounts by Institution Type, 1993–2002
Recipients Aid in $millions
Institution Category (% of total) (% of total)
Overall HOPE Program 1,217,172 1564.3
State System 498,814 1077.57
(41.0) (68.9)
Four-year 389,452 840.09
(32.0) (53.7)
Two-year 109,362 237.48
(9.0) (15.2)
Private 171,280 143.86
(14.0) (9.2)
Four-year 136,581 101.91
(11.2) (6.5)
Two-year 34,699 41.95
(2.8) (2.7)
Technical Schools 547,078 342.86
(44.9) (21.9)
Note: The listing of technical schools in the breakdown of the HOPE
Scholarship totals reflects the fact that 13 of the 34 technical schools
offer associate’s degrees. In addition, a few, state-system, 2- and 4-
year schools award diplomas and certificates.
20
Table 2
Means and Standard Deviations of State Variables
Georgia vs SREB States, Pre- and Post-HOPE
State(s) Pre-HOPE Post-HOPE
Georgia
Registrations 38.51 39.54
( 1.82) ( 2.25)
Population 65.23 71.89
(1.74) ( 2.33)
Personal Income 115.04 160.89
( 11.76) ( 18.00)
Auto Employment 14.86 15.32
( 2.53) ( 1.52)
SREB
Registrations 31.38 29.26
( 23.70) ( 22.00)
Population 55.76 59.50
( 41.82) ( 46.28)
Personal Income 97.68 129.71
( 79.73) ( 108.48)
Auto Employment 10.40 13.62
( 7.32) ( 10.78)
Note: Registrations and population are reported in 100,000s;
auto employment in thousands. Personal income is measured
in billions.
21
Table 3
Estimated HOPE Effects on State
Car Registrations, 1988–97
HOPE Effect US SREB Border States
ˆ
δGA –0.007 0.006 0.009
(–0.607) (0.217) (0.200)
se 0.050 0.047 0.048
ˆ
δGA,89 0.001 –0.008 0.000
(0.150) (–0.583) (0.005)
ˆ
δGA,90 0.016 0.010 0.002
(2.222) (0.470) (0.049)
ˆ
δGA,91 0.066 0.035 0.025
(7.463) (1.491) (0.552)
ˆ
δGA,92 0.076 0.056 0.070
(12.320) (3.581) (2.272)
ˆ
δGA,94 0.138 0.113 0.109
(10.159) (3.574) (1.850)
ˆ
δGA,95 0.113 0.107 0.085
(6.267) (2.588) (0.835)
ˆ
δGA,96 0.060 0.062 0.041
(2.699) (1.124) (0.365)
ˆ
δGA,97 0.009 0.019 –0.012
(0.352) (0.301) (–0.091)
se 0.049 0.046 0.045
NT 510 150 60
Note: Estimated HOPE effects are conditional on state-level
differences in personal income, population, automobile in-
dustry employment, state and year fixed effects, and state-
specific trends. Robust t-ratios are given in parentheses. Note
that 1993 is the base year for the time-varying HOPE effect
estimates and “se” denotes the standard error of the regres-
sion.
22
Table 4
Variable Means and Standard Deviations
Georgia Counties, 1993–2001
Above 75th Below 25th
Variable Overall PCI Percentile PCI Percentile
Car 27.88 77.91 6.71
Registrations ( 101.58) ( 57.07) ( 5.51)
State System 331.98 930.06 74.44
Awards ( 775.75) ( 1361.71) ( 87.01 )
Private College 114.30 333.98 25.44
Awards ( 277.24) ( 482.32) ( 31.93)
Tech College 374.59 762.29 176.45
Awards ( 537.45) ( 848.61) ( 204.22)
Population 48.36 131.39 13.42
( 101.58) ( 176.13) ( 11.49)
Personal 1180.44 3597.34 213.84
Income ( 3369.16) ( 6088.67) ( 187.36)
Number of counties 159 40 40
Note: Car registrations and population are in thousands and personal
income in millions. The HOPE awards are reported in their actual
levels.
23
Table 5
Car Registrations and HOPE Awards
Fixed-effects Results, 1993–2001
All Counties
Variable State System Private Technical
HOPE Awards –0.012 0.008 –0.003
(–0.695) (0.785) (–0.197)
Population 0.527 0.524 0.529
(1.208) (1.190) (1.205)
Income –0.264 –0.270 –0.262
(–1.699) (–1.733) (–1.695)
se 0.112 0.112 0.112
N1272 1272 1272
Notes: Variables measured in logs. Results are conditional on county
and year fixed effects and county-specific trends. Robust asymptotic
t-ratios in parentheses and “se” denotes the standard error of the
regression
24
Table 6
Car Registrations and HOPE Awards
Fixed-effects Results, 1993–2001
Counties Above 75th Percentile in Per Capita Income
Variable State System Private Technical
HOPE Awards 0.045 0.055 –0.021
(1.390) (1.966) (–0.827)
Population 0.473 0.596 0.662
(0.808) (1.092) (1.161)
Income –0.336 –0.332 –0.319
(–0.822) (–0.791) (–0.783)
se 0.089 0.089 0.089
N320 320 320
Notes: Variables measured in logs. Results are conditional on county
and year fixed effects and county-specific trends. Robust asymptotic
t-ratios in parentheses and “se” denotes the standard error of the
regression
25
Table 7
Car Registrations and HOPE Awards
Fixed-effects Results, 1993–2001
Counties Below 25th Percentile in Per Capita Income
Variable State System Private Technical
HOPE Awards –0.030 –0.002 0.020
(–1.083) (–0.135) (0.841)
Population –0.224 –0.189 –0.183
(–0.444) (–0.353) (–0.341)
Income –0.084 –0.090 –0.096
(–0.252) (–0.274) (–0.292)
se 0.126 0.126 0.126
N320 320 320
Notes: Variables measured in logs. Results are conditional on county
and year fixed effects and county-specific trends. Robust asymptotic
t-ratios in parentheses and “se” denotes the standard error of the
regression
26
Figure 1
Car Registrations of Georgia compared to SREB States
1a: GA vs. AL
-6
-4
-2
0
2
4
6
8
1990 1992 1994 1996
Year
Cars-Residuals
AL
GA
1c: GA vs. FL
-10
-5
0
5
10
1990 1992 1994 1996
Year
Cars-Residuals
FL
GA
1e: GA vs. LA
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Residuals
GA
LA
1g: GA vs. MS
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
GA
MS
1b: GA vs. AR
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Residuals
AR
GA
1d: GA vs. KY
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Resdiuals
GA
KY
1f: GA vs. MD
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Residuals
GA
MD
1h: GA vs. NC
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Ye a r
GA
NC
Figure 1 (continued)
Car Registrations of Georgia compared to SREB States
1i: GA vs. OK
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Residuals
GA
OK
1k: GA vs. TN
-6
-4
-2
0
2
4
6
1990 1992 1994 1996
Year
Cars-Residuals
GA
TN
1m: GA vs. VA
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Residual
s
GA
VA
1j: GA vs. SC
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Residuals
G
A
SC
1l: GA vs. TX
-15
-10
-5
0
5
1990 1992 1994 1996
Year
Cars-Residuals
GA
TX
1n: GA vs. WV
-4
-3
-2
-1
0
1
2
3
4
1990 1992 1994 1996
Year
Cars-Residuals
GA
WV
Note: Each figure plots log automobile registrations purged of state and time effects (the
residuals from regressions of log registrations on state and time dummies).
Figure 2
Estimated Yearly HOPE Effects on Log Car Registrations
With 95% Confidence Intervals
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
0.25
1989 1990 1991 1992 1993 1994 1995 1996 1997
Year
Estimated Effec
t
Estimate
95% CI Min
95% CI Max
Note: 1993 is base year and thus is omitted.
