Hope for Whom? Financial Aid for the Middle Class and Its ...users.nber.org/~dynarski/2000 Hope for Whom.pdf · and high-income families. I estimate the impact of aid on the col-lege

Hope for Whom?

629

National Tax JournalVol. LIII, No. 3, Part 2

Abstract - The federal government and the states have recentlyenacted a slew of aid policies aimed at college students from middle-and high-income families. I estimate the impact of aid on the col-lege attendance of middle- and upper-income youth by evaluatingGeorgia’s HOPE Scholarship, the inspiration of the new federalHope Scholarship. The results suggest that Georgia’s program hashad a surprisingly large impact on the college attendance rate ofmiddle- and high-income youth. Using a set of nearby states as acontrol group, I find that Georgia’s program has likely increasedthe college attendance rate of all 18- to 19-year-olds by 7.0 to 7.9percentage points. The results suggest that each $1,000 in aid($1998) increased the college attendance rate in Georgia by 3.7 to4.2 percentage points. Due to key differences between the federaland Georgia programs, these estimates should be treated as a gen-erous upper bound on the predicted effect of the federal Hope Schol-arship. Further, the evidence suggests that Georgia’s program haswidened the gap in college attendance between blacks and whitesand between those from low- and high-income families. The federalHope Scholarship, should it have its intended effect on middle- andupper-income attendance, will also widen already large racial andincome gaps in college attendance in the U.S.

INTRODUCTION

Federal and state governments have recently ushered in anew generation of student aid policies. The largest of these

new programs are the federal tax incentives known as theHope Scholarship and Lifetime Learning Credit, which al-low families of college students to offset their educationalcosts with tax benefits of up to $1,500 a year. For the 1998 taxyear, a total of $3.5 billion was delivered to 4.8 million fami-lies through these two programs, making them one of thelargest sources of federal subsidies for college students.1 Asecond federal initiative, the Education IRA, allows familiesto put after–tax dollars into college savings and accumulateinterest tax–free. The federal programs join a wide array of

Hope for Whom? Financial Aid for theMiddle Class and Its Impact on

College Attendance

Susan Dynarski Kennedy School ofGovernment, HarvardUniversity, Cambridge,MA 02138 andNBER, Cambridge,MA 02138

1 Estimate based on initial analysis of 1998 tax returns provided by Julie–Anne Cronin of the Office of Tax Analysis, U.S. Department of Treasury. Byway of comparison, during the academic year 1997–8 students received$6.3 billion in Pell Grants (College Board, 1998).

National Tax JournalVol. 53 no. 3 Part 2 (September 2000) pp. 629-662

NATIONAL TAX JOURNAL

630

aid programs introduced by the states.Many states now have tax–free collegesavings plans and several have introducedtuition tax credits. State–funded merit–based scholarships are the latest studentaid fashion to sweep across the states, withmore than a dozen legislatures consider-ing such programs. Georgia’s HOPE(Helping Outstanding Students Educa-tionally) Scholarship, the namesake andinspiration of the federal program, is thelargest and best known of the state meritscholarships.

All of these new state and federal pro-grams differ from traditional student aidin one crucial aspect: they are not need–based. Historically, government aid forcollege has been strongly focused on low–income students. Eligibility for the twolargest federal aid programs, the PellGrant and Stafford Loan, is determinedby a complex formula that defines finan-cial need on the basis of income, assets andfamily size. The formula is quite progres-sive: 90 percent of dependent studentswho receive federal grants grew up infamilies with incomes less than $40,000.2

By contrast, the new aid programs areaimed squarely at middle– and high–in-come families. Tax–deferred savings plansmost benefit upper–income families, whoboth face the highest marginal tax ratesand have the highest saving rate. The fed-eral Hope Scholarship and LifetimeLearning Credit have three key character-istics that limit their benefit to low–incomefamilies. First, the income cutoffs for eli-gibility for the subsidies are set highenough that less than ten percent of filinghouseholds exceeds them.3 Second, allow-able educational expenses are offset byany need–based aid received. As a result,a student who attends the typical two–

year college and is poor enough to receivethe maximum Pell Grant receives no HopeScholarship. Third, the subsidy takes theform of a non–refundable tax credit, sothat a family too poor to pay taxes receivesno Hope Scholarship.

How will this new breed of student aidaffect college attendance rates? Will aid tomiddle– and high–income families actu-ally increase college attendance, or will thenew programs simply transfer funds toinfra–marginal students? We have littleevidence with which to answer thesequestions. There is scant research concern-ing the impact of tuition subsidies onmiddle– and upper–income youth, for thesimple reason that most existing aid pro-grams focus on needy students. Historyhas therefore provided few experimentsthat would allow us to measure the re-sponsiveness to aid of middle– and up–per–income youth. There are reasons tosuspect that low– and upper–incomeyouth respond differently to aid: wealth,parental education and academic pre-paredness are all tightly correlated withincome and each have their own impacton the decision to attend college. And asI will show later in the paper, a fairlysimple model of human capital invest-ment suggests that the effect of aid onschooling decisions will vary by parentalincome.

In this paper, I estimate the impact ofaid on the college attendance of middle–and upper–income students by evaluat-ing the program that is the namesake andinspiration of the federal Hope Scholar-ship: the Georgia HOPE Scholarship. In1993, Georgia initiated HOPE, which isfunded by a state lottery. The program al-lows free attendance at Georgia’s publiccolleges for state residents with at least a

2 Calculated from data in Table 314 in U.S. Department of Education (1998a).3 The income cap is set at an adjusted gross income of $100,000 for married–couple families and $50,000 for

single filers. Early analysis of 1997 tax returns indicates that only 7.4 percent of household tax returns fellabove these income cutoffs; this estimate is expected to rise somewhat as late returns are tabulated (Hollenbeckand Kahn, 1998).


Hope for Whom?

631

B average in high school.4 More than adozen states are weighing the introduc-tion of merit scholarships like Georgia’sand governors in Alabama and SouthCarolina were elected in 1998 on the basisof pledges that they would initiate lotter-ies to fund education in their own states(Selingo, 1999). Despite the widespreadattention paid Georgia’s HOPE Scholar-ship, there has been no rigorous evalua-tion of its impact upon college atten-dance.5 Do programs such as Georgia’sHOPE actually increase college enroll-ment? Or do they simply transfer fundsto families who would have sent their chil-dren to college anyway?

I use data from the Current PopulationSurvey to evaluate the impact of Georgia’sHOPE Scholarship on college attendance.Using a set of nearby states as a controlgroup, I find that Georgia’s program haslikely increased college attendance ratesamong all 18– to 19–year-olds by 7.0 to7.9 percentage points. I obtain similar re-sults using a within–state control groupto estimate the program’s effect.

I further find that the increase is con-centrated among Georgia’s white stu-dents, who have experienced a 12.3 per-centage point rise in their enrollment raterelative to whites in nearby states. Theblack enrollment rate in Georgia appearsunaffected by HOPE. The differential im-pact of HOPE on blacks and whites islikely due to the focus of HOPE onmiddle– and upper–income students whoperform well in high school.

Among the subset of youth that is mostlikely eligible for Georgia HOPE—thosefrom upper–income families—I find theattendance rate has risen 11.4 percentage

points relative to that of a similar popula-tion in nearby states. There has been norelative rise in attendance among youthfrom lower–income families. However,these last estimates should be treated withcaution, as the analysis indicates that thesub–sample of youth for whom family–income data is available is non–randomlyselected. This particular estimate, unlikethe others of the paper, may therefore bebiased by sample selection.

Overall, the results suggest that for each$1,000 of subsidy the college attendancerate of middle– and upper–income youthrises by four to six percentage points. Thisis a surprisingly large response: the esti-mate is of the same order of magnitude asthose reported by studies that examine theeffect of aid on low–income students. Wecan use these estimates to calculate theshare of HOPE funds that are going tomarginal and infra–marginal students.The college attendance rate in Georgiabefore HOPE’s introduction was 29.9 per-cent. The estimates of the paper indicatethat HOPE increased this rate by seven toeight percentage points, suggesting thatabout 20 percent of post–HOPE college at-tendance by 18– to 19–year-olds was in-duced by HOPE. Roughly, then, about 80percent of HOPE funds flow to those whowould have gone to college in the absenceof the subsidy. While the average HOPEScholarship paid for those at Georgia’scolleges and universities is $1,900, four in-framarginal students must be subsidizedin order to induce one into college.

The results should be extrapolated toother states and programs with caution.Georgia had attendance rates well belowthe national average before HOPE was

4 The federal Hope Scholarship in its proposed form also required minimum grades in high school, but thisprovision was dropped over concerns about the cost and propriety of having the Internal Revenue Servicegather high school transcripts.

5 The Council for School Performance of Georgia State University has conducted a number of studies of HOPE.These studies contrast the academic performance of HOPE recipients with that of college students who don’tget HOPE. Since HOPE scholars are selected on the basis of academic merit, these studies do not provide avalid test of HOPE’s impact.



632

introduced, and it is possible that a simi-lar program in a high–attendance statesuch as Massachusetts would not havea similar impact. Further, Georgia’sprogram is unusual in its simplicity, scale,and publicity. A less transparent formof subsidy—such as a tax credit or tax–free interest on college savings—maynot produce responses of similar magni-tude.

The HOPE Scholarship has also affectedanother margin of educational decision–making: the choice of college. Using in-stitutional data from the University Sys-tem of Georgia and the federal Depart-ment of Education, I find that HOPE hasincreased the likelihood that Georgia stu-dents will attend college in their homestate. As a result, the University Systemof Georgia has seen an increase in theshare of its students who are from thestate; this effect is most pronounced in thestate’s most selective four–year public col-leges. In particular, fewer Georgia stu-dents now attend college in the states thatborder Georgia. At the ten schools inGeorgia’s border states that draw the mostGeorgia freshmen, enrollment of Geor-gians dropped from 17 percent of fresh-men enrollment in 1992 to 9 percent in1998.

The paper is organized as follows. Thenext section provides a short overview ofthe literature on the effect of subsidies oncollege enrollment and offers a theoreti-cal motivation for why we might suspectthat the effect of a subsidy varies withfamily income. The third section discussesGeorgia’s program in detail. The fourthsection explains the empirical methodol-ogy and data used in the analysis. The fifthsection presents results. The sixth sectionexplores the policy implications of thepaper’s findings; the following sectionconcludes.

BACKGROUND AND LITERATUREREVIEW

A long literature attempts to identify theeffect of aid on college attendance. Leslieand Brinkman (1988) review studieswhose estimates indicate that $1,000 in aidincreases college attendance rates by threeto five percentage points.6 The effect of aidon college attendance is generally poorlyidentified in these studies. Identificationis an important consideration in this con-text, because aid is correlated with manyobservable and unobservable characteris-tics that have their own influence on edu-cation. Dynarski (1999) reviews a hand-ful of well-identified studies that use adiscrete shift in aid policy as a source ofexogenous variation in aid. Kane (1994)finds that a $1,000 drop in public tuitionincreases college attendance by about fourpercentage points. Reyes (1995) finds that$1,000 in loan subsidy increases collegeattendance by 1.5 percentage points; thisrelatively low response may indicate thatyouths do not fully recognize the subsidyvalue of a loan. Dynarski (1999) exploitsvariation produced by the elimination inthe early 1980s of the Social Security Stu-dent Benefit Program, which annuallyprovided aid to nearly one million collegestudents. She finds that $1,000 in aid in-creases college attendance by four per-centage points.

None of these studies, however, focus-es on the effect of a price subsidy onmiddle– and upper–income students.7 Arecent study by Kane (1999) focuses mostclosely on the question posed by this pa-per. Kane uses longitudinal data to exam-ine how the effect of public tuition on col-lege attendance varies by family income.Exploiting variation in tuition across andwithin states, Kane finds that a $1,000drop in tuition increases the attendance

6 All dollar figures in the paper are converted to real 1998 values.7 Reyes (1995) examines the effect of subsidized loans on middle– and upper income students, but her estimate

will not inform us of the effect of grant aid if youth are debt–averse or do not fully recognize the subsidy valueof a low–interest loan.


Hope for Whom?

633

rate of lower–income youth 5.2 percent-age points more than it does that ofmiddle– and upper–income youth.

Modeling the Relationship between Aid,Income and Schooling

In policy discussions of the effect of aidon education, it is often assumed that low–income students are most responsive toaid. It is loosely argued that low–incomestudents are liquidity–constrained—thatis, that they face interest rates that arehigher than market rates—and will there-fore respond most elastically to a givensubsidy to college costs. However, it isstraightforward to demonstrate that whileadding liquidity constraints to a simplehuman capital model does predict that li-quidity–constrained youth will invest lessin education, it does not yield the predic-tion that constrained and unconstrainedyouth will respond any differently to agiven subsidy to college costs. We requirea more nuanced treatment of the interac-tion of income, interest rates, and collegecosts in order to develop theoretical in-sight into how high– and low–income stu-dents will respond to aid.

In this section, I develop a model ofschooling investment that suggests thatthe effect of aid on schooling will vary byincome. The key assumption of the modelis that the level of debt that a college stu-dent must assume for an additional yearof schooling is a decreasing function of hisfamily’s income. Higher levels of debtlead to higher interest payments, whichincrease the college costs of low–incomestudents relative to those of high–incomestudents. By reducing both the presentprice of college and the level of debt onwhich interest is paid, aid increases theoptimal level of schooling. The effect ofaid on schooling decisions is predicted todrop or remain constant as income rises;which of these predictions holds dependson whether marginal interest rates rise orremain constant as the level of debt in-

creases. The rest of this section lays outthe detail of the model and discusses itsimplications.

Earnings are assumed to be an increas-ing, concave function of schooling:

[1] y(S) = α + βS – δS2

In this equation, S is years of schooling,y(S) is earnings and both β and δ arepositive. This equation describes the ob-served empirical relationship betweenearnings and schooling. DifferentiatingEquation (1) with respect to schoolingyields the marginal benefit of a year ofschooling:

[2] MB = y′(S) = β – 2δS

In other words, an additional year ofschooling yields a return that drops witheach year of schooling obtained. The mar-ginal cost of schooling consists of a year’stuition net of any student aid, interestcharged on any borrowing and an indi-vidual–specific parameter that reflectsheterogeneity in the ease with which anadditional year of school is completed:

[3] MC = p(1 – A) + R(B) + γi

Here, p is annual tuition and A is the per-centage of annual tuition that is offset bystudent aid. γi is the individual–specific,non–financial cost of college. For example,γi may reflect academic preparation forcollege–level work. R(B) is the total inter-est rate paid when borrowing amount B.I will discuss the characteristics of theR(B) function shortly.

A student’s borrowing is determined bythe price of college, living expenses, theamount of aid, and the amount that hisfamily can contribute to his education.Parents devote a fixed proportion of in-come, α, to their child’s education. Theamount borrowed is then:

[4] B = p(1 – A) + C – αY



634

Here, Y is parental income and C is livingexpenses other than tuition.8 A positivevalue of B indicates a student whoborrows for college, while a negativevalue indicates a combination of prices,family income, and aid that allows afamily to forgo borrowing and insteadsave.

Setting equal the marginal benefit andcost of college, solving for S and takingthe derivative of schooling with respectto aid yields:

[5]

This term is unambiguously positive: aidincreases schooling. The effect of aidworks through two channels, which I willrefer to as the price effect and the liquid-ity effect. The price effect of aid is repre-sented by the first term in the numerator:aid reduces the price of a year of collegeand thereby increases demand for school-ing. The liquidity effect is represented bythe second term: aid reduces borrowingand total interest paid and thereby in-creases the impact of aid on schoolingdecisions. Both of these effects are scaledby the δ term, which represents the rateat which the return to schooling drops asthe level of schooling rises. If returns dropquickly (that is, δ is large), then schoolingchoice will be relatively insensitive to aid.Conversely, if returns drop slowly thenaid will have a relatively large effect onschooling choices.

We are interested in how the responseof schooling choice to aid varies by fam-ily income. Taking the derivative of Equa-tion [5] with respect to income yields:

[6]

The sign of this term depends on the shapeof the interest function, R(B). Recall thatR(B) is the total interest that is paid whenborrowing amount B. If R′′(B) is positive,then Equation [7] is negative, implyingthat the effect of aid decreases as incomerises. R′′(B) is positive if students face mar-ginal interest rates that rise with borrow-ing.

There is evidence that students andfamilies face rising interest rates whenborrowing for college.9 The cheapestsource of funds for most families is feder-ally subsidized student loans: the interestrate is about 7 percent, and the govern-ment pays all interest while the student isin school. For a student borrowing themaximum of $17,125 and repaying overten years, a loan with a nominal rate of 7percent and an in–school interest subsidyis equivalent to a standard loan with anominal rate of 4.5 percent.10 For mostfamilies, housing equity is the next cheap-est source of funds.11 Current mortgagerates are about 8 percent; the preferentialtax treatment of mortgage interest impliesan effective rate of 6 percent for those inthe 28 percent tax bracket. If housing eq-uity has been exhausted, families can turnto unsubsidized federal loans, whichcharge a rate of 7–8 percent but requirethat interest be paid while the student isin school. As a last resort, families can turnto more expensive sources of funds, suchas unsecured personal loans, retirementsavings, and credit cards. It is plausible,then, that students face interest rates that

d2SdAdY

= – 2δ

[R′′(B)]αp

8 Note that living expenses affect the level of debt but do not directly enter the marginal cost equation. This isbecause living expenses are incurred whether or not an individual attends college, and so are not considereda marginal cost of attending college. However, interest is paid in order to fund living expenses only if a personattends college and so in this case constitutes a marginal cost of schooling.

9 This paragraph draws on Kane (1999).10 Recent changes in tax law further increase the subsidy value of federal loans. As of 1998, borrowers can de-

duct $1,000 a year in loan interest from taxable income if their adjustable gross income falls below $40,000(single taxpayers) or $60,000 (married taxpayers).

11 Housing equity is the cheapest source of capital for high-income families because of their high marginal taxrates. Further, those from the highest–income families are not eligible for subsidized loans.

dS=

p + pR′(B)

dA 2δ


Hope for Whom?

635

rise with their borrowing. Equation [6], inthis case, predicts that the effect of aid onschooling will diminish as income rises.

The effect of aid does not vary by in-come if interest rates remain constant asthe level of borrowing rises. In this case,Equation [6] is equal to zero. The effect ofaid will actually increase with income ifmarginal rates decline with the level ofborrowing, but I ignore this case sincesuch a pattern of interest rates is implau-sible. The model therefore predicts that,for each individual, the effect of aid eitherdrops or remains constant as income rises.

However, as I will explain with an ex-ample, Equation [6] does not unambigu-ously predict that at least as large a shareof low–income youth as high–incomeyouth will be induced to attend collegeby a given subsidy.12 Say that the indi-vidual–specific, non–financial cost ofschooling (γi) is identically and normallydistributed within the low–income andhigh–income populations.13 The collegeattendance margin will cut at a higherpoint in the γi distribution among high–income youth than it will among low–income youth. This is because amonghigh–income youth the reduced level ofdebt (and therefore interest payments)that their parents’ financial contributionto their schooling allows can offset rela-tively high non–financial costs of college.14

How does this affect our parameter ofinterest, the relative shares of high– andlow–income youth induced by a subsidyto attend college? The share of an incomegroup that is pushed over the college at-tendance margin is a function not only of

the sensitivity of that group to aid but alsothe proportion of the group near the mar-gin of college attendance. If the distribu-tion of γi is normal (or, more generally,non–uniform) it is ambiguous whether theshare that is close to the margin of atten-dance is larger among low– or high–income youth. For example, the college at-tendance margin among high–incomeyouth might appear at the mean of the γi

distribution, where the largest share of thegroup is concentrated. The college atten-dance margin among low–income youthwill then appear below the mean of the γi

distribution, where a smaller share of thegroup is concentrated. In this case, a givensubsidy could easily push into college alarger share of high–income youth thanlow–income youth. It is simple to con-struct a scenario in which the opposite istrue.

The effect of aid on the schooling deci-sions of middle– and high–income youthmust therefore be determined empirically,rather than extrapolated from aid’s effecton low–income youth. In the next section,I discuss the policy experiment that willbe used to estimate the impact of aid onthe college attendance rates of upper–income youth.

GEORGIA’S HOPE SCHOLARSHIP

In 1991, Georgia Governor Zell Millerrequested that the state’s General Assem-bly consider the establishment of a state–run lottery, with the proceeds to be de-voted to education. The Georgia GeneralAssembly passed lottery–enabling legis-

12 Thanks to David Autor for drawing this to my attention. Stanley (1999) discusses this point in the context ofthe G.I. Bill, which he finds had its largest impact on the schooling of veterans who grew up in families of highsocioeconomic status.

13 Normalcy is not required here; any non–uniform distribution will produce the same conclusion.14 Ellwood and Kane (1999) show that even after controlling for test scores (a measure of the non–financial costs

of schooling, γi) low–income students are less likely to go to college than high–income students. This is evi-

dence that the college attendance margin cuts at a higher point in the γidistribution of high–income youth

than low–income youth. The paper also shows that low–income families contribute less money to theirchildren’s education than high–income families and that this differential is not fully offset by the greaterlevels of aid received by low–income students. This indicates that low–income students do, therefore, facehigher borrowing requirements than do high–income students.



636

lation during its 1992 session and for-warded the issue to voters, who approvedthe required amendment to the state’s con-stitution in November of 1992. The first lot-tery tickets were sold in June of 1993. Since1993, $2.5 billion in lottery revenue hasflowed into Georgia’s educational institu-tions (Byron and Henry, various years).The legislation and amendment enablingthe lottery specified that the new fundswere not to crowd out spending from tra-ditional sources. While it is not possibleto establish conclusively that such crowd–out has not occurred, spending on educa-tion has risen substantially since the lot-tery was initiated, both in absolute dollarsand as a share of total state spending.Roughly equal shares of lottery funds havegone to 4 programs: the HOPE Scholar-ship, educational technology for primaryand secondary schools, a new pre–kinder-garten program, and school construction.

Residents who have graduated since1993 from Georgia high schools with atleast a 3.0 grade point average are eligiblefor HOPE.15 The first scholarships were dis-bursed in the fall of 1993. Participation inHOPE during its first year was limited tothose with family incomes below $66,000;the income cap was raised to $100,000 in1994 and eliminated in 1995. HOPE paysfor tuition and required fees at Georgia’spublic colleges and universities. Those at-tending private colleges are eligible for anannual grant, which was $500 in 1993 andhad increased to $3,000 by 1996. Theseamounts are offset by other sources of aid.A student who receives the maximum Pell

Grant gets no HOPE Scholarship but re-ceives a yearly book allowance of $400.16 A$500 education voucher is available to thosewho complete a General Education Di-ploma (GED). Public college students mustmaintain a GPA of 3.0 to keep the scholar-ship; a similar requirement was introducedfor private school students in 1996.

Georgia education officials, concernedthat students would forgo applying forfederal aid once the HOPE Scholarshipwas available, created an application pro-cess designed to prevent this outcome.Those from families with adjusted grossincomes lower than $50,000 must completethe Free Application for Federal StudentAid (FAFSA) in order to apply for HOPE;the rationale for the $50,000 income thresh-old is that few students above that cutoffare eligible for need–based federal aid.17

The four–page FAFSA requests detailedincome, expense, asset, and tax data fromthe family. Those with family incomesabove $50,000 fill out a short, one–pageform that requires no information aboutfinances other than a confirmation thatfamily income is indeed above the cutoff.

In 1998–9, 140,000 students received$189 million in HOPE Scholarships. Fifty–four percent of those students attended atwo– or four–year college, while the bal-ance attended a technical institute. Thebulk of spending (81 percent) goes to theminority of students at two– and four–year schools, however, since their tuitionsare substantially higher than those of thetechnical institutes. Georgia politicianshave deemed HOPE a great success,

15 The high school GPA requirement is waived for those enrolled in certificate programs at technical institutes.For high school seniors graduating after 2000, only courses in English, math, social studies, science, andforeign languages will count toward the GPA requirement. More than 40 percent of those who currentlyreceive the HOPE Scholarship would be ineligible under this definition.

16 As a result of this provision and the scaling back of the state’s need–based State Student Incentive Grants(SSIGs), some low–income students have actually seen their state aid reduced slightly since HOPE was intro-duced (Jaffe, 1997). This contemporaneous shift in SSIG spending has the potential to contaminate the paper’sestimates, especially the specifications in which low–income youth are used as a control group for upper–income youth. However, SSIG spending was so miniscule—$5.2 million in 1995, before the program wasscaled back—that the impact of its elimination on the estimates is likely inconsequential.

17 In 1995, only 3.7 percent of dependent students from families with incomes over $40,000 received federalgrant aid, while 57 percent of those from families with income under $20,000 did so (U.S. Department ofEducation, 1998a).


Hope for Whom?

637

pointing to the steady rise in the numberof college students receiving HOPE. Thekey question is whether the program isactually increasing college attendance orsimply subsidizing students who wouldhave attended college even in the absenceof HOPE. In the next sections, I discussthe data and empirical strategy I will useto answer this question.

DATA AND EMPIRICALMETHODOLOGY

Data

The data for the analysis are the Octo-ber Current Population Survey (CPS) andthe Integrated Postsecondary EducationData System (IPEDS). The CPS is amonthly, national household survey thateach October gathers detailed informationabout schooling enrollment. IPEDS inte-grates into a single data set information

from a variety of surveys of post–second-ary institutions conducted by the U.S. De-partment of Education.

The CPS will be used for the bulk of theanalysis, as its detailed demographic dataallow for the identification of youth whoare most likely eligible for HOPE. I havemerged annual, state–level unemploy-ment statistics from the Bureau of LaborStatistics with the CPS data. Means for theCPS data–set are in Table 1.

The CPS, while the best available re-source for the purposes of this paper, hasits flaws. First, state samples are small: forthe period 1989 to 1997, there are a totalof 470 18– to–19–year-olds from Georgiain the October CPS. As a result, year–to–year changes in enrollment rates withinGeorgia are fairly noisy.18 The CPS’ smallwithin–state samples also preclude any in-formative analysis of detailed schoolingchoices, such as whether college studentsare induced by HOPE to attend public vs.

TABLE 1SAMPLE MEANS

OCTOBER CPS, 1989–9718–19–YEAR–OLDS

Black

Family Income < $50K

Metro Area Resident

Age 18

State Unemployment Rate

N

0.377(0.486)

0.754(0.432)

0.661(0.475)

0.474(0.500)

5.53(0.709)

183

Georgia

0.265(0.441)

0.740(0.439)

0.682(0.467)

0.492(0.500)

6.24(1.73)

3,231

SoutheasternStates

0.325(0.469)

0.611(0.489)

0.703(0.458)

0.522(0.500)

5.73(1.06)

287

GeorgiaSoutheastern

States

0.260(0.438)

0.666(0.472)

0.716(0.451)

0.503(0.500)

5.36(1.37)

3,110

Note: Means are weighted by CPS sample weights. Standard deviations are in parentheses. The income mean isfor the 70.2 percent of 18– to 19–year–olds that both appear on their parents’ CPS record and have a valid re-sponse to the family income question. The Southeastern states consist of the South Atlantic and East SouthCentral Census Divisions: Alabama, Delaware, District of Columbia, Florida, Kentucky, Maryland, Mississippi,North Carolina, South Carolina, Tennessee, Virginia, and West Virginia.

1989–92 1993–7

18 I could more than double the sample by extending the age cutoff to 22. However, older youth were not eligiblefor HOPE during its early years. In fact, later in the paper, older youth will be used as a control group to studythe response of younger Georgia residents to HOPE.



638

private schools, or four–year vs. two–yearschools. The IPEDS allow for limited ex-ploration of these questions.

Second, information about a youth’sfamily background is not consistentlyavailable in the CPS. Family backgroundvariables, such as parental income, areavailable only for those youth that appearon their parents’ CPS record. A youthappears on their family’s record for oneof two reasons: they live with their fam-ily or they are away at college. The prob-ability that a youth has family backgroundinformation available is therefore afunction of their propensity to attend col-lege. This form of sample selection willproduce bias in analyses where collegeattendance is an outcome of interest.19 Oneof the estimation strategies I test requiresfamily income information, and for thatanalysis the sample is limited to those whoappear on their parents’ record. I will ex-plore the sensitivity of these results tosample selection. The bulk of the analysisis based on the full sample of 18– to 19–year-olds and is not subject to this sourceof bias.

Third, the CPS identifies neither thestate in which a person attended highschool nor the state in which they attendcollege. However, within a group thisyoung, migration across state linesother than to attend college is minimal.And when a youth does go out of stateto college, CPS coding standards arethat they are recorded as residents oftheir home state.20 Since the CPS does notprovide the state in which the student at-tends college, I am unable to use thesedata to detect if HOPE has altered not justthe rate of attendance but the proclivityof youth to attend college in–state.The IPEDS allows us to gain some insightinto this issue, as the Department of

Education every other year gathers fromcolleges data about their students’ statesof residence.

Empirical Methodology

The empirical approach of the paper isstraightforward. I examine changes in col-lege attendance rates over time withinGeorgia, looking for discontinuities atthe time of HOPE’s introduction. A con-trol group is required in order to netout any secular trends in college atten-dance. A natural control group is the otherstates of the southeastern United States.I use as a control group the South Atlan-tic and East South Central Census Divi-sions, which consist of Georgia plusAlabama, Delaware, the District ofColumbia, Florida, Kentucky, Maryland,Mississippi, North Carolina, South Caro-lina, Tennessee, Virginia, and WestVirginia. As will be shown below, the re-sults are robust to the choice of controlgroup.

The effect of HOPE is identified by dif-ferences between Georgia and the rest ofthe southeastern United States in the timepattern of college attendance rates. I usedifference–in–differences estimation,comparing attendance rates before andafter HOPE was introduced, within Geor-gia and in the rest of the region. This cal-culation can be made using ordinary leastsquares:

[7] yi = α1 + β1(Georgiai * Afteri)

+ δ1Georgiai + θ1Afteri + νi1

where the dependent variable is a binarymeasure of college attendance, Georgiai isa binary variable that is set to one if ayouth is a Georgia resident and Afteri is a

19 Cameron and Heckman (1999) discuss this point.20 Such youth enter the sample if their parents’ home has been selected as a CPS household. Youth who leave

home and set up independent households do not show up on their parents’ record and are recorded as resi-dents of whatever state they live in. The overwhelming majority (about 90 percent) of 18– to 19–year-olds doshow up on their parents’ record, so these coding rules appear to hold in practice.


Hope for Whom?

639

binary variable that is set to one in thesample years in which HOPE was in place(1993 forward). This specification controlsfor the time trend in college attendance(θ1), as well as for the average effect onattendance of being a Georgia resident(δ1). The reduced–form effect of the HOPEScholarship is identified by β1. The iden-tifying assumption is that any relativeshift in the attendance rate of Georgiayouth is attributable to the introductionof HOPE.

I also undertake a strategy that useswithin–state control groups. This triple–differencing approach exploits key insti-tutional aspects of HOPE. I first take ad-vantage of the fact that HOPE was initiallyopen to only the youngest high schoolgraduates. In September of 1993, for ex-ample, the only HOPE recipients weremembers of the high school class of 1993.Older youth therefore form a natural con-trol group against which to measure theeffect of HOPE on more recent high schoolgraduates. I pool the sample of older (23to 24) and younger (18 to 19) college–ageyouth and run the following regression:

[8] yi = α2 + β2(Georgiai * Afteri)

+ δ2Georgiai + θ2Afteri

+ φ(Georgiai * Afteri * Youngi)

+ η(Georgiai * Youngi)

+ λ(Afteri * Youngi) + πYoungi + νi2

In this equation, Youngi is a dummy thatindicates whether a person is aged 18 to19. This specification controls for the maineffects of being a Georgia resident (δ2) andbeing aged 18 to 19 (π) as well as their in-teraction (η). The specification further netsout national trends in the college atten-dance of 18– to 19–year-olds (λ) and

shocks to the schooling decisions ofGeorgia’s college–age population (β2). Thecoefficient of interest is (φ), which identi-fies the effect of the HOPE Scholarship onthe college attendance rate of 18– to 19–year–olds. This approach has a key advan-tage over the difference–in–differences ofEquation [7] in that the estimated effectof HOPE will not be biased by any Geor-gia–specific shocks to the college atten-dance decisions of young people that oc-curred after HOPE was introduced.

A second triple–differencing strategytakes advantage of HOPE’s family incomeeligibility rules. As was explained earlier,recent Georgia high school graduates withannual family incomes over $50,000 whomeet the high school grade requirementautomatically qualify for HOPE by fillingout a simple, one–page form. Those withlower income, by contrast, apply for fed-eral aid with a complex, four–page formand wait several months to learn the sizeof their grant award, which is then de-ducted from their HOPE Scholarship. Asa result, lower–income students receiveHOPE Scholarships that are both smallerand more uncertain than those receivedby their better–off peers. We would there-fore expect that the introduction of theHOPE program had a smaller impact onlower–income youth than higher–incomeyouth.

In order to exploit this aspect of HOPE,I divide the sample into those with annualfamily incomes above and below $50,000.I then run the same triple–difference speci-fication as that of Equation [8] with thevariable Youngi replaced by a dummy in-dicating that a youth is from an upper–income family.21 This specification identi-fies the effect of HOPE as the change inthe college attendance rate of upper–income over lower–income youth in Geor-gia relative to the same change in thecontrol states. This approach has the ad-

21 The use of family income data in the Current Population Survey in this context potentially produces biasedestimates. I discuss this point below.



640

vantage of controlling for any Georgia–specific shock that affects equally thecollege attendance decisions of upper–income and lower–income youth. But thisapproach is imperfect, as it is unable todistinguish between two key reasons whyHOPE’s impact may differ by income: theincome eligibility rules described aboveand differences in academic performancein high school. If lower–income youth per-form worse in higher school, then, even ifthey are offered the same HOPE Scholar-ship as upper–income youth, they will re-spond less, since fewer will be able to getinto and succeed in college. Evidence in-dicates that there is a correlation betweensocioeconomic status (SES) and highschool performance. Among high schoolseniors in 1992 who intended to go to col-lege, 24.4 percent of those of high SES hada grade point average of at least 3.5, whilejust 10.0 percent of those from low SESfamilies had grades that high.22

All estimates are undertaken using or-dinary least squares. Probit produces simi-lar results. The CPS sample weights areused in all the regressions. The standarderrors are adjusted for heteroskedasticitydue to the binary dependent variable.Standard errors are also adjusted for cor-relation within state and year.

RESULTS

Difference–in–Differences Estimates

Table 2 shows college attendance ratesfor youth that are residents of Georgia and

the rest of the Southeast, before and afterthe Georgia HOPE Scholarship was intro-duced in 1993. Previous to the introduc-tion of HOPE, the enrollment rate in Geor-gia of 18– to 19–year-olds was relativelylow: 30.0 percent, as compared to 41.5percent in the rest of the Southeast. AfterHOPE was introduced, the enrollmentrate in the rest of the Southeast did notchange appreciably. However, the Geor-gia enrollment rate rose to 37.8 percent.

These two differences are differenced inthe last column of Table 2. The impliedeffect of HOPE on the college enrollmentrate is 7.9 percentage points. In Table 3, Imake the same calculation using ordinaryleast squares. In the first row of Column(1) is the estimate that corresponds to thatof Table 2.23 The estimate of 7.9 percent issignificant at the one–percent level. Thisis a fairly large effect, given an initial at-tendance rate in Georgia of 30 percent. Theresult implies that HOPE increased atten-dance probabilities by about 26 percent(7.9 percentage points/30 percentagepoints). Further, the estimates suggest thatHOPE nearly closed the gap betweenGeorgia and the rest of the Southeast incollege attendance. Later, I will put thiseffect in perspective by comparing it toprevious estimates of the response of col-lege attendance to subsidies.

In the second column of Table 3, I add aset of covariates to the regression. For rea-sons discussed earlier, I limit myself tocovariates that are available for the entiresample. Variables whose generation re-

22 U.S. Department of Education (1995).23 The two estimates are necessarily the same, since it is computationally equivalent to take differences in the

means in Table 1 and to regress attendance against the Georgia dummy, the after dummy, and their interaction.

TABLE 2DIFFERENCE–IN–DIFFERENCES

SHARE OF 18–19–YEAR–OLDS ATTENDING COLLEGEOCTOBER CPS, 1989–97

GeorgiaRest of Southeastern StatesDifference

Note: Means are weighted by CPS sample weights. The Southeastern states are defined in the note to Table 1.

Difference

0.078–0.0010.079

Before 1993

0.3000.4150.115

1993 and After

0.3780.4140.036


Hope for Whom?

641

quires that the youth and parents appearon the same record, such as parental in-come and education, are not included. Iinclude indicator variables for residencein a metro area, being black, survey year,and age. The estimate drops slightly, from7.9 percentage points to 7.5 percentagepoints, with a standard error of 3.0 per-centage points. The estimate is significantat the two–percent level.

I next test whether the estimate is sen-sitive to the choice of control states. InTable 4, I run the difference–in–differenceregression using as control states, in turn,the entire Southeast, the states that bor-der Georgia, and the entire United States.The border–state estimate is 8.7 percent-age points, as compared to 7.9 percentagepoints for the Southeastern states. The es-timate is significant at the one–percentlevel. In Column (3), where the controlgroup is the entire United States, the co-efficient drops to 7.0 percentage points but

is still significant at the one–percent level.The estimates are therefore relativelystable across choice of control group, rang-ing from 7.0 percentage points to 8.7 per-centage points. None of the estimates ismore than a standard error away from theother two. Since the results of the paperare consistent across control group, in theremainder of the paper I will only showresults that use the Southeastern states asthe control group.

Controlling for Georgia–SpecificEconomic Shocks

Georgia may have experienced eco-nomic shocks around the time of HOPE’sintroduction that were not shared with itsneighboring states. In this case, the col-lege attendance rate in Georgia may havediverged from that of its neighbors forreasons unrelated to the introduction ofHOPE. I attempt to address this problem

TABLE 3COLLEGE ATTENDANCE OF 18–19–YEAR–OLDS

OCTOBER CPS, 1989–97CONTROL GROUP: SOUTHEASTERN STATES

(1)Difference–in–

Differences

(2)Add

Covariates

(3)Add Local EconomicConditions Controls

After*Georgia

Georgia

After

Age 18

Metro Resident

Black

State Unemployment Rate

Year DummiesR2

N

0.079(0.029)

–0.115(0.023)

–0.001(0.018)

0.0036,811

0.075(0.030)

–0.100(0.019)

–0.042(0.014)

0.042(0.016)

–0.134(0.014)

Yes0.0236,811

0.070(0.030)

–0.097(0.018)

–0.042(0.016)

0.042(0.015)

–0.133(0.015)

0.005(0.007)

Yes0.0236,811

Note: Regressions are weighted by CPS sample weights. Standard errors are adjusted for heteroskedasticity andcorrelation within state–year cells. The Southeastern states are defined in the note to Table 1.



642

in two ways. First, in Column (3) of Table3, I add to the difference–in–differencesregression the unemployment rate in theyouth’s state of residence during the sur-vey year. The coefficient on the unemploy-ment rate is insignificant and positive. Thecoefficient of interest is unaffected: thedifference–in–differences estimate is 7.0percentage points and is significant at thethree–percent level.24

An alternative, non–parametric methodof controlling for Georgia–specific eco-nomic shocks is to use a within–state con-trol group. Youth in their early twentiesare likely to experience the same shocksto the opportunity costs of college asyouth in their late teens. The HOPE Schol-arship differentially affects these twogroups, however. HOPE eligibility isbased on graduating from a Georgia highschool in 1993 or later. Even in 1997, thosewho are aged 23 to 24 would have gradu-ated from high school before HOPE wasintroduced, and so were generally not eli-gible for the program. It should be notedthat changes in the program rules in 1995did open HOPE to older Georgians who

had completed two years of college witha 3.0 average. But since this older grouphas never been eligible for subsidies intheir first two years of college, they stillform a valid control group when the out-come is attendance at the freshman andsophomore level.25

Results for this analysis are in Table 5.The first two columns show separate esti-mates for the younger, eligible group andthe older, ineligible group. The impact ofHOPE on freshman and sophomore en-rollment among 18– to 19–year-olds issimilar to its effect on enrollment at anylevel of college: 8.1 percentage points,with a standard error of 3.0 percentagepoints. Among older students, whoshould be unaffected by HOPE, the effectis zero: 0.7 percentage points with a stan-dard error of 1.1 percentage points. InColumn (3), I pool the two age groups intoa single regression and test for the statis-tical significance of the difference betweenthese two coefficients. In Georgia, 18– to19–year-olds increased their attendancerelative to 23– to 24–year-olds by 7.5 per-centage points more than they did in the

TABLE 4ALTERNATIVE CONTROL GROUPS

COLLEGE ENROLLMENT OF 18–19–YEAR–OLDSOCTOBER CPS, 1989–97

After*Georgia

Georgia

After

R2

N

(1)Southeastern States

0.079(0.029)

–0.115(0.023)

–0.001(0.018)

0.0036,811

(2)States Bordering Georgia

0.087(0.031)

–0.100(0.025)

–0.008(0.021)

0.0034,275

(3)United States

0.070(0.024)

–0.135(0.021)

0.009(0.009)

0.00132,266

Note: Regressions are weighted by CPS sample weights. Standard errors are adjusted for heteroskedasticity andcorrelation within state–year cells. The states that border Georgia are Alabama, Florida, North Carolina, SouthCarolina, and Tennessee. The Southeastern states are defined in the note to Table 1.

24 I have also experimented with specifications that include lags of the unemployment rate. The results aresubstantively unchanged.

25 The prospect of receiving HOPE in the third year could, however, affect the probability that an older studententers college. This will tend to bias toward zero my estimate of HOPE’s effect when older students are usedas the control group.


Hope for Whom?

643

other Southeastern states. This triple–difference estimate is significant at thethree–percent level. If the identifying as-sumption of the analysis is correct, thenthis result suggests that the impact ofHOPE eligibility was to increase the col-lege attendance rate by 7.5 percentagepoints.

These last results effectively control forany trends in employment opportunityand college costs (e.g., tuition prices) thataffect both the younger and older membersof the college–age population. The estimateobtained from this specification is statisti-cally the same as that obtained from thesimplest difference–in–differences analysisin Table 3. The conclusion that HOPE in-creased college attendance of Georgia’syoung people by about seven to eight per-centage points is therefore robust to a va-riety of specifications and control groups.

Using Income Data to Identify theEligible Population

As was discussed earlier, the eligibilityrequirements for Georgia’s HOPE Schol-arship vary by income. The analysis so farhas measured increases in relative atten-dance among all Georgia youth. In orderto attempt to narrow in on the group thatwas most likely eligible for the subsidy, Idraw on family income data. Georgia uses$50,000 as the income threshold abovewhich students are automatically eligiblefor the HOPE Scholarship, as long as theymeet the high school academic require-ments. In Column (1) and Column (2) ofTable 6 are regression results for samplemembers from families with annual in-comes above and below $50,000, respec-tively. Among those from higher–incomefamilies, the difference–in–differences es-

TABLE 5TRIPLE DIFFERENCE, BY AGE

FRESHMAN AND SOPHOMORE COLLEGE ENROLLMENTOCTOBER CPS, 1989–97

CONTROL GROUP: SOUTHEASTERN STATES

(1)Difference–in–Differences:

Age GroupAffected by Scholarship

(18–19)

(2)Difference–in–Differences:Age Group Not Affected

by Scholarship(23–24)

(3)Difference–in–Differences–in–

DifferencesPooled Regression

After*Georgia*Age 18–19

After*Georgia

Georgia

After

Age 18–19

Georgia*Age 18–19

After*Age 18–19

R2

N

0.081(0.030)

–0.101(0.023)

–0.016(0.018)

0.0026,811

0.007(0.011)

–0.021(0.010)

0.009(0.005)

0.0017,445

0.075(0.034)

0.007(0.011)

–0.021(0.010)

0.009(0.005)

0.364(0.012)

–0.080(0.027)

–0.025(0.019)

0.18514,526




644

timate is 11.4 percentage points, with astandard error of 5.4 percentage points. Bycontrast, the estimate for lower–incomestudents is –1.4 percentage points and isnot statistically different from zero. InColumn (3) of Table 5, I pool the two in-come groups and test whether the differ-ence in their responses to HOPE is statis-tically significant. A full set of main effectsfor Georgia residence, income, and timeis included in the regression, along withtheir second–order interactions. The tripleinteraction of Georgia residence, being oflow income, and time identifies the effectof the HOPE scholarship in this regres-sion. The difference in response acrossincome groups is significant at the 5 per-cent level. These results indicate that, inGeorgia, higher–income youth increasedtheir attendance relative to lower–income

youth by 12.7 percentage points more thanthey did in the other Southeastern states.26

Since the group of youth for whom fam-ily income is available is a function of thecollege attendance rate, it is prudent tocheck the sensitivity of these results tosample selection. In Column (4) of Table6 is replicated from the previous sectionthe difference–in–differences estimate forthe full sample of 18– to 19–year-olds: 7.9percentage points. In Column (5) I run thesame regression with only the 70 percentof 18– to 19–year-olds that appear on theirparents’ record and have family incomedata available. The sub–sample estimateof 4.5 percentage points is statistically dif-ferent from the full–sample estimate, sug-gesting that selection bias is a problem inthis context. Further, as expected, the biasis toward zero.

26 Adding a set of covariates (race, urbanicity, and age) to the specifications of Table 5 does not affect the esti-mates, although their precision is increased slightly.

TABLE 6TRIPLE DIFFERENCE, BY INCOME



(1)Parents’Income> $50K

(2)Parents’Income< $50K

(3)

TripleDifference

(4)

FullSample

(5)Limit to ThoseWith Parents’Income Data

After*Georgia* > $50K

After*Georgia

Georgia

After

> $50K

Georgia* > $50K

After* > $50K

R2

N

0.114(0.054)

–0.159(0.041)

–0.070(0.030)

0.0091,401

–0.014(0.062)

–0.067(0.038)

–0.037(0.018)

0.0043,380

0.127(0.062)

–0.014(0.062)

–0.067(0.038)

–0.037(0.018)

0.350(0.023)

–0.091(0.030)

–0.033(0.035)

0.0944,781

0.079(0.029)

–0.115(0.023)

–0.001(0.018)

0.0036,811

0.045(0.043)

–0.095(0.034)

–0.022(0.018)

0.0024,781



Hope for Whom?

645

Note that, in this context, selection biasin the CPS is likely most severe amonglow–income youth. If the college atten-dance of low–income youth is particularlysensitive to college costs, then the rate atwhich they appear on their parents’records will also co–vary particularlystrongly with college costs. Selection bias,which will, in this case, bias downwardthe estimated effect of a subsidy on col-lege attendance, will then be most severefor low–income students. This suggeststhat some of the difference in attendanceresponse across income groups found inTable 6 is driven by differing degrees ofbias in the estimated coefficients. How-ever, the bias would have to be extremelylarge in order to negate the conclusion thatHOPE has increased the college atten-dance of upper-income youth more thanthat of lower-income youth.

The Differential Impact of HOPE onBlacks and Whites in Georgia

Given the focus of the Georgia HOPEScholarship on middle– and upper–income families, it is probable that the pro-gram has had a differential impact on thecollege attendance of blacks and whites.To get a sense of the correlation betweenrace and the income guidelines of theGeorgia HOPE Scholarship, I examinedthe family incomes of 16– to 17–year–oldsin the 1989–97 October CPS.27 In Georgiaduring 1989 to 1997, 94 percent of blackand 62 percent of white 16– to 17–year-olds lived in families with incomes lessthan $50,000.28 The numbers for the restof the United States are similar: 88 and 64percent, respectively.29 These figures indi-cate than very few black youth in Geor-

gia can, given adequate high school aca-demic performance, automatically qualifyfor a HOPE Scholarship, while about 40percent of white youth can. While race istherefore only a rough proxy for income,an analysis by race does skirt the selec-tion bias problems seen in the previoustable.

In Table 7, I show the results of split-ting the difference–in–differences analy-sis by race; the estimates for the entiresample are in the first column for ease ofcomparison. In Column (2) are the esti-mates for whites. College attendanceamong whites rose 12.3 percentage pointsfaster over this period in Georgia than inthe rest of the southeastern United States.The estimate is significant at the 1 percentlevel. By contrast, college attendanceamong blacks did not rise significantly inGeorgia relative to the other southeasternstates: the difference–in–differences esti-mate for blacks is –2.7 percentage points,with a standard error of 5.2 percentagepoints. In Column (5) I pool blacks andwhites and test for a statistically signifi-cant difference in their responses toHOPE. The responses of whites and blacksare different at the 6 percent level of sta-tistical significance.

The evidence presented here clearlysuggests that HOPE has widened the ra-cial gap in college attendance in Georgia.This is likely due both to HOPE’s differ-ential impact by income and its highschool academic requirements. Sinceblacks have lower incomes, they both facea more complicated HOPE applicationprocess and are eligible for more gener-ous federal grants, which are deductedfrom the HOPE Scholarship. Blacks alsohave lower average grades in high school,

27 I choose youth of this age because almost all (92 percent) show up in the same record as their parents.28 Note that this refers to the nominal income distribution. This is appropriate, since the Georgia rules are writ-

ten in nominal rather than real terms.29 These figures for the share with income below $50,000 may appear high. This is because the unit of observa-

tion is not the family but the child. Since lower–income families have more children, the distribution of familyincome within a sample of children has a lower mean than the distribution of family income within a sampleof families.



646

which means a smaller proportion willmeet HOPE’s academic requirements:among those members of the high schoolclass of 1992 intending to go to college, 21percent of whites had a high school GPAof 3.5 or above, while only 4 percent ofblacks had such high grades.30 The avail-able data do not allow us to disentanglewhether it is the income or academic rulesthat drive the differential effect of HOPEon blacks and whites.

Is the Timing Right?

The results so far suggest that HOPEhas had a significant impact on collegeattendance rates in Georgia. This sectionprobes the robustness of this result by ex-amining more closely the timing of therelative rise in Georgia’s attendance rate.A sharp relative increase in attendance

rates in Georgia in the years after 1993 isconsistent with the hypothesis that HOPEinduced the increase in college–going thatthe difference–in–differences analysis haspicked up. By contrast, a slow relative risein Georgia’s attendance rates that beginsbefore HOPE was introduced suggeststhat HOPE is not responsible for this in-creased attendance. It should be said atthe outset that the small size of the year–state cells in the CPS sample makes this asuggestive exercise, as it is quite difficultto differentiate within–state changes inattendance rates that are due to a programchange from those that are due to randomnoise.

In Figure 1, I graph the coefficients froma specification in which college attendanceis regressed on the interaction of yeardummies and the Georgia dummy, alongwith state unemployment rate and dum-

TABLE 7TRIPLE DIFFERENCE, BY RACE



After*Georgia*White

After*Georgia

Georgia

After

White

Georgia*White

After*>White

R2

N

0.079(0.029)

–0.115(0.023)

–0.001(0.018)

0.0036,811

(1)Full Sample

(2)Whites

(3)Blacks

(4)Triple Difference

0.123(0.045)

–0.109(0.039)

–0.002(0.022)

0.0024,974

–0.027(0.052)

–0.088(0.030)

–0.000(0.026)

0.0071,837

0.149(0.079)

–0.027(0.052)

–0.088(0.030)

–0.000(0.026)

0.126(0.021)

–0.020(0.058)

–0.001(0.030)

0.0196,811


30 U.S. Department of Education (1995).


Hope for Whom?

647

Gra

phed

are

the

coe

ffic

ient

s on

the

int

erac

tion

of

year

dum

mie

s w

ith

the

Geo

rgia

dum

my.

Coe

ffici

ents

are

obt

aine

d f

rom

a s

ingl

e re

gres

sion

tha

t in

clud

es t

he G

eorg

ia*y

ear

inte

ract

ions

as

wel

l as

the

sta

te u

nem

ploy

men

t ra

te a

nd d

umm

ies

for

year

, ag

e, r

ace,

and

urb

anic

ity .

Dep

end

ent

vari

able

is

an i

ndic

ator

for

col

lege

att

end

ance

. T

he 1

992

inte

ract

ion

has

been

nor

mal

ized

to z

ero.

Fig

ure

1.

Est

imat

ed E

ffec

t of

HO

PE

Eli

gib

ilit

y o

n C

olle

ge

Att

end

ance

, By

Yea

r



648

mies for age, year, urbanicity, and race.The 1992 interaction has been normal-ized to zero. The 95 percent confidenceinterval for each point estimate isalso plotted. This is equivalent to the dif-ference–in–differences specification inColumn (3) of Table (3), except that theGeorgia effect is now allowed to vary byyear.

In this graph, we can clearly see therelative rise in attendance rates in Geor-gia that we have picked up with thedifference–in–differences regressions.The timing of this rise is less clear. Thereis a sharp relative increase in 1992, theyear before HOPE was introduced. Therelative attendance rate rises again in1994 and then especially sharply in 1996.A sharp drop in 1997 sends Georgia’srelative attendance rate back to itspre–HOPE level. As will be discussedlater, this drop–off in the program’s effectmay be due to the fact that a majority offreshman HOPE recipients do not receivethe HOPE Scholarship in their sophomoreyear, because they drop out of school and/or fail to meet the college GPA require-ment.

Figure 1 does not provide strong sup-port for the hypothesis that HOPE causedthe relative rise in Georgia’s attendancerate. An alternative explanation is that, forsome reason, an upward trend in relativeattendance rates in Georgia began in 1992and simply persisted when HOPE wasput in place. We can attempt to distinguishbetween these competing hypotheses byfocusing on the attendance rates of thosemost likely eligible for HOPE. Figure 2replicates the previous figure, except thatthe sample is now limited to those fromfamilies with incomes more than $50,000.For this group, the time pattern of rela-tive attendance rates is roughly consistent

with the timing of HOPE. Relative atten-dance rates are flat through 1993, the firstyear of the program. This is to be expected,since in 1993 the family income cap on par-ticipation was $66,000, thereby excludingfrom HOPE much of the populationwhose relative attendance rates are shownin this figure. There is, however, a largerelative increase in this group’s attendancerate in 1994, the year that the income capwas raised to $100,000. There is anotherslight relative increase in 1995, the yearthat the income cap was eliminated com-pletely. The interaction term drops backsubstantially in 1996, to pre-HOPE lev-els.31

The Effect of HOPE on College Choice

This section briefly examines the effectof HOPE on students’ choice of college.High school students are on a variety ofmargins when making their decisionsabout college. Some youth are on themargin of attending college at all. Thetheoretical effect of HOPE on these youthis to push them in to college, most likelyinto less–than–two–year or two–yearschools. Other youth may be set onattending a two–year school. HOPE willpush some of them toward a four–yearcollege, by driving down its relative cost.32

The net impact of these two effects on theshare of college–going youth attendingless–than–four–year schools is ambigu-ous, as the first effect will increase thenumber of youth at less–than–four–yearschools and the second effect will pushstudents at those same schools into four–year colleges. A last group of youth is seton attending a four–year school, andHOPE will shift some of them towardchoosing to attend college within theirhome state.33

31 I will discuss a possible explanation for this observed drop–off in HOPE’s effect later.32 Two–year colleges are generally cheaper than four–year colleges. The HOPE Scholarship makes them both free.33 Students at four–year colleges, as compared to those at two–year schools, are more likely to be on the margin

of attending out of state. Nationwide, about 25 percent of four–year college students go to school outside theirhome state, while only about three percent of two–year college students do so.


Hope for Whom?

649

Gra

phed

are

the

coe

ffic

ient

s on

the

int

erac

tion

of

year

dum

mie

s w

ith

the

Geo

rgia

dum

my.

Coe

ffici

ents

are

obt

aine

d f

rom

a s

ingl

e re

gres

sion

tha

t in

clud

es t

he G

eorg

ia*y

ear

inte

ract

ions

as

wel

l as

the

sta

te u

nem

ploy

men

t ra

te a

nd d

umm

ies

for

year

, ag

e, r

ace,

and

urb

anic

ity.

Dep

end

ent

vari

able

is

an i

ndic

ator

for

col

lege

att

end

ance

. T

he 1

992

inte

ract

ion

has

been

nor

mal

ized

to z

ero.

Fig

ure

2.

Fam

ily

In

com

e >

$50K

, Est

imat

ed E

ffec

t of

HO

PE

Eli

gib

ilit

y o

n C

olle

ge

Att

end

ance

, By

Yea

r



650

Analysis of annual data from the Uni-versity System of Georgia (USG) and bi-ennial data from the Department ofEducation’s Residence and Migration Sur-vey (a component of IPEDS) yields resultsthat are consistent with these theorizedeffects. First, evidence suggests that HOPEhas shifted students from two–year col-leges into four–year colleges. Figure 3graphs data from the USG on the share ofits students from Georgia that are attend-ing four–year colleges. This share declinedthrough 1992–3, was level for a year andthen began to increase during HOPE’ssecond year of operation. This graph sug-gests that HOPE has had its second theo-rized effect, that of encouraging those whowould have gone to a two–year college toinstead attend a four–year college. In fact,relying on this graph alone, we wouldconclude that the second effect of HOPE(shifting youth from less–than–four–yearinto four–year schools) dominates the first(shifting youth from no college to less–than–four–year schools). However, theUSG data do not inform us about enroll-ment in the private sector, especially atless–than–two–year schools. About half ofthose receiving HOPE Scholarships areenrolled at these schools, which generallydo not grant degrees and are run as for–profit enterprises. Data on enrollment atthese schools is quite poor: while theIPEDS surveys all degree–grantingschools, it only includes a sample of thenon–degree schools and the samplingmethodology appears to vary from yearto year. We therefore cannot directly mea-sure the effect of HOPE on enrollment atthese institutions.

Second, data from both the USG andResidence and Migration Survey suggestthat HOPE has had the third theorizedeffect of encouraging Georgia residentswho would have attended four–year col-lege out of state to instead stay in Geor-gia. Data from the Residence and Migra-tion Survey indicate that in 1992 about5,000 Georgians were freshmen at two–

and four–year colleges in the states thatborder Georgia. This represented an av-erage of 3.4 percent of the border states’freshmen enrollment. By 1998, just 4,500Georgians crossed state lines to enter col-lege in the border states, accounting foran average of 2.9 percent of freshmen en-rollment in those states. This drop in mi-gration was concentrated in a group ofborder schools that have traditionallydrawn large numbers of Georgians. At theten schools drawing the most Georgiafreshmen in 1992, students from Georgianumbered 1,900 and averaged 17 percentof the freshman class. By 1998, the ten topdestinations enrolled 1,700 Georgians,who represented 9 percent of freshmanenrollment. Jacksonville State College inFlorida, for example, drew 189 Georgianfreshmen in 1992 and only 89 in 1998; theshare of the freshman class from Georgiadropped from 17 to 11 percent.

Further supporting the conclusion thatGeorgia’s four–year college students arenow more likely to attend college in stateis a shift in the composition of Georgia’sfour–year colleges. In Figure 4 is grapheddata from the USG on the share of fresh-men enrollees that are Georgia residentsat Georgia’s two– and four–year publiccolleges. The data are separately plottedfor the two–year, four–year and the elitefour–year colleges in the state. Here wesee a definite shift toward Georgia resi-dents since HOPE was introduced, withthe effect most pronounced at four–yearcolleges (especially the top schools) andleast evident at the two–year schools. Thispattern fits with our understanding thatfour–year students are most mobile whenmaking college attendance decisions.

Discussion of Results

The analysis of this section suggests thatHOPE increased attendance rates in Geor-gia. The attendance rate of Georgia’s 18–to 19–year–olds has risen by 7.0 to 7.9 per-centage points relative to that in the rest


Hope for Whom?

651

Gra

phed

is th

e sh

are

of U

SG s

tud

ents

from

Geo

rgia

who

are

eno

lled

in fo

ur–y

ear

colle

ges.

Sou

r ce:

Uni

vers

ity

Syst

em o

f Geo

rgia

Ten

–Yea

r E

nrol

lmen

t Rep

ort,

vari

ous

year

s.

Fig

ure

3.

Un

iver

sity

Sy

stem

of

Geo

rgia

Stu

den

ts, S

har

e of

Geo

rgia

Res

iden

ts E

nro

lled

in F

our–

Yea

r C

olle

ges



652

Fig

ure

4.

Un

iver

sity

Sy

stem

of

Geo

rgia

Stu

den

ts, G

eorg

ia R

esid

ents

as

Shar

e of

Tot

al E

nro

llm

ent


Hope for Whom?

653

of the Southeast. We can convert thisresult into an estimate comparable toprevious studies of the effect of collegecosts on attendance by dividing thechange in attendance by the subsidyprovided by HOPE. HOPE paid for tu-ition and mandatory fees at Georgia’spublic colleges and universities, which av-eraged $1900 from 1993 to 1997.34 The es-timates of this paper therefore translateinto an increase in the attendance rate of3.7 to 4.2 percentage points for each $1,000in subsidy. How do these estimates com-pare to previous research on the effect ofaid on attendance? Reyes (1995) and Kane(1994) find that a $1,000 drop in collegecosts increases attendance by 1.5 and 3.8percentage points, respectively. In previ-ous work (Dynarski, 1999), I have esti-mated that $1,000 in grant aid increasesattendance of a low– to middle–incomepopulation by 4 percentage points. Thisplaces the estimate of HOPE’s effect onmiddle– and upper–income youth at themid–range of previous estimates of theeffect of aid.

In light of the conventional wisdom thatmiddle– and high–income youth are in-fra–marginal consumers of higher educa-tion, this is a surprisingly large effect.There are two possible explanations. First,as was discussed earlier, a larger propor-tion of upper– than lower–income stu-dents may be close to the margin of col-lege attendance. A given subsidy maytherefore cause a relatively large share ofhigh-income students to spill over themargin into college. Second, particularcharacteristics of Georgia and the HOPE

Scholarship may intensify the program’seffect. This possibility directly affects thevalidity of the paper’s estimates in pre-dicting the effect of other middle–class aidprograms. I next turn to a discussion ofhow confidently we can extrapolate esti-mates based on the Georgia HOPE Schol-arship to programs such as the federalHope Scholarship.

WHAT CAN THE GEORGIA HOPESCHOLARSHIP TELL US ABOUTTHE EFFECT OF AID ON THEMIDDLE CLASS?

The paper so far has shown that aid canaffect the schooling decisions of middle–income youth. Ideally, we could simplyapply the estimates of the paper to a pro-gram such as the federal Hope Scholar-ship. Indeed, there are key similaritiesbetween the Georgia and federal pro-grams. They are of roughly equal finan-cial value and focus their subsidies onroughly the same portion of the incomedistribution. The average value of theGeorgia HOPE Scholarship for those at-tending a public college or university is$1,900, while the maximum federalHope Scholarship is $1,500 and the maxi-mum Lifetime Learning Credit is $1,000.35

Both programs largely exclude low–income students by linking the subsidy tohow much outside aid is received.36 Nei-ther program excludes the well–off: theGeorgia program has no income cap onparticipation while the federal incomecaps are set quite high in the income dis-tribution.

34 Those attending private colleges are eligible for an annual grant, which was $500 in 1993 and gradually in-creased to $3,000 by 1996. For the 1993–7 period, $1,900 is a fair approximation of HOPE’s average subsidy toprivate college attendance.

35 The Lifetime Learning Credit provides a credit of 20 percent of up to $5,000 in tuition, which makes it lessvaluable than Hope for those attending low–tuition schools. Hope allows a credit equal to the sum of 100percent of the first $1,000 of tuition and 50 percent of the second. The Hope Scholarship is available for just thefirst two years of college, while the Lifetime Learning Credit is available at any level.

36 The Georgia program reduces its subsidy one dollar for each dollar of other grant aid. The federal programonly allows reimbursement of college costs net of other grant aid. A low–income youth who attends a typicalpublic college and receives an average Pell grant will not get a federal tax credit. By contrast, a middle–classstudent who attends the typical private school and is not eligible for a Pell grant will get a credit.



654

Despite these similarities, the results ofthe Georgia analysis should be extrapo-lated to the federal Hope Scholarship withcaution. There are key institutional differ-ences between the Georgia and federalsubsidies that will likely drive a wedgebetween their effects. In the balance of thissection, I will discuss these differences,which, as will become clear, generallyimply that the impact of the federal HopeScholarship will be substantially lowerthan that of the Georgia program.

Information and Transaction Costs of theGeorgia and Federal Programs

It is obvious that a youth, or their fam-ily, needs to know about and apply for anaid program if aid eligibility is to affecttheir schooling decisions. A program thatis well–publicized, easily comprehended,and requires little paperwork will have agreater impact on schooling decisionsthan one that is obscure, complicated, andimposes a large burden of paperwork.The latter program has high transactioncosts, which reduce the value of aid to aneligible youth. Seeking out informationabout an aid program, comprehending itsrules, and obtaining and completing anynecessary forms are all transaction coststhat are imposed upon potential students.Transaction costs can also be imposedupon schools, if they are required tohandle aid–related paperwork. If schoolsraise tuition in the face of these increasedcosts, the impact of a given subsidy willbe further reduced.

The transaction costs of the federal pro-gram are quite high. The subsidy is deliv-ered through the federal tax code, notknown for its transparency or simple pa-perwork. The size of the subsidy is uncer-tain when the college–attendance decision

is made, since taxpayers do not know thesize of their credit until their tax liabilityand eligible educational expenses for agiven year have been determined.37 Theprogram is costly to schools, as well,which are required to collect informationabout who is responsible for a givenstudent’s tuition (parents and outsidescholarship providers, for example) andmail to the Internal Revenue Service a list-ing of the responsible parties, their tax-payer identification numbers, and an an-nual accounting of the portion of tuitionand fees eligible for the tax credit.38

By contrast, Georgia’s program has un-usually low transaction costs. Informationabout the program is effectively dissemi-nated through the high schools. Fifty-ninepercent of high school freshmen, whenasked to list some requirements of HOPE,volunteer that a high school GPA of 3.0 isnecessary; more than seventy percent canname the program without prompting.39

The paperwork is minimal, at least for stu-dents from families with incomes above$50,000. The application for the 1998–9academic year for that group consists of asingle page with about a dozen questions,of which the only financial query is: “Wasyour family’s Adjusted Gross Income for1997 $50,000 or more?” Lower–incomestudents fill out the a four–page form thatis roughly as involved as the typical taxreturn. In order to ease this process, Geor-gia college officials assist applicants incompleting this form, check it for accu-racy, and mail it in.

Tuition Effects of the Georgia andFederal Programs

Opponents of the federal tax creditshave expressed concern that they willdrive up tuition prices, as schools seek to

37 The federal credit is therefore delivered well after educational expenses have been incurred. This aspect of theprogram will likely reduce its impact among liquidity–constrained families.

38 Bowing to complaints that the new reporting requirements are expensive and burdensome, the IRS has waivedthis requirement until after the 2000 tax year (Hebel, 1999).

39 Henry et al. (undated).


Hope for Whom?

655

capture the subsidy. The California legis-lature has discussed raising tuitions sothat its students can qualify for the fullHope Scholarship.40 To gauge whetherHOPE has driven up tuition prices inGeorgia, I examine trends in schoolingcosts in Georgia and the U.S. Plotted inFigure 5 is the natural log of the averagetuition, fees, room, and board paid by stu-dents at public four–year colleges in Geor-gia and the rest of the U.S. for the aca-demic years 1986–7 through 1997–8. Fig-ure 6 contains the corresponding graph forprivate four–year colleges.41

Public college costs were relatively flatin Georgia before HOPE, with costs in1993–4 only about 6 percent higher thantheir level in 1986–7. Real prices in Geor-gia actual dropped during the years im-mediately preceding HOPE. By contrast,real public schooling costs in the U.S. rosesteadily between 1986–7 and 1993–4, fora total increase over this period of around15 percentage points. After HOPE wasintroduced, the situation was reversed,with public college costs in Georgia ris-ing at a rate higher than that of the U.S.Between 1993–4 and 1997–8, schoolingcosts rose about 21 percent in Georgia and8 percent in the rest of the U.S. To a lesserdegree, the same pattern emerges from theplot of private school costs in Figure 6.Private schooling costs rose slightly fasterin the U.S. than Georgia before HOPE (18vs. 16 percentage points, respectively) but

the situation was reversed after HOPEwas introduced (8 vs. 12 percentagepoints, respectively).

These results suggest that HOPE hashad an inflationary effect on college costsin Georgia, especially on the publicschools.42 The inflationary effect of the fed-eral tax credit on tuition is likely to be evenstronger. In Georgia, the state governmentboth distributes the subsidy and sets tu-ition prices for the public sector, whichshould at least moderate schools’ ten-dency to raise prices in order to capturethe subsidy. There is no such brake in thefederal program, since the federal govern-ment has no control over prices in thehigher–education market. We wouldtherefore expect the inflationary effects ofthe federal scholarship to be more se-vere.43

Academic Requirements of the GeorgiaProgram

Georgia’s subsidy requires a 3.0 GPA inboth high school and college; the federalprogram has no grade requirement. Thehigh school GPA requirement theoreticallyhas two countervailing effects. It couldmagnify HOPE’s effect by encouragingstudents to increase their academic effortand decrease its effect by denying eligi-bility to those who are slightly below thegrade cutoff but who can handle college–level work. The requirement might also

40 Basinger and Healy (1998). A student must incur at least $2,000 in eligible costs in order to get the full Hopecredit. Many community colleges, and some state universities, charge tuition lower than that threshold.

41 These data are from Table 312 in U.S. Department of Education (1998a) and Table 81 in U.S. Department ofEducation (1998b). The Georgia series are much noisier than the national series, which is likely a function oftheir relative sample sizes: a single school’s large tuition increase can shift the Georgia mean appreciably,while the actions of a single school are not likely to visibly affect the national trend.

42 While the dollar value of the HOPE subsidy is roughly the same in private and public sectors, schooling costsare higher in the private schools. We would therefore expect HOPE to induce a smaller percentage increase inprivate than in public schooling costs, which is consistent with the results seen here. HOPE would also tend tohave a smaller level effect on prices in the private than the public sector, since the scholarship covers only afixed dollar portion of private tuition but all of public tuition, thereby giving public schools more latitude inraising prices.

43 Further, if the federal Hope Scholarship leads to tuition increases, the net effect of this program may be todecrease black and low–income attendance, since these populations will face the full impact of higher tuitionbut largely be ineligible for the subsidy. Institutions may choose to ameliorate this effect, however, by usingtheir increased tuition revenues to cross–subsidize needy students who are ineligible for Hope.



656

Fig

ure

5.

Log

of A

ver

age

Tuit

ion

, Fee

s, R

oom

, an

d B

oar d

at

Pu

bli

c 4–

Yea

r Sc

hoo

ls (

$199

8)


Hope for Whom?

657

Sour

ce: T

able

312

in U

.S. D

epar

tmen

t of E

duc

atio

n (1

998a

) and

Tab

le 8

1 in

U.S

. Dep

artm

ent o

f Ed

ucat

ion

(199

8b).

Fig

ure

6.

Log

of A

ver

age

Tuit

ion

, Fee

s, R

oom

, an

d B

oar d

at

Pri

vat

e 4–

Yea

r Sc

hoo

ls (

$199

8)



658

encourage grade inflation in high school,which would expand the pool of studentsthat, at least on paper, meet minimumcollege entry requirements. How this af-fects the number of students HOPE in-duces to attend college depends on howable colleges are to detect grade inflation.

The effect of the college GPA require-ment is less ambiguous. The B–averagerequirement cuts off financial assistance tostudents who could make it through col-lege, albeit with mediocre grades, and istherefore almost certain to reduce HOPE’simpact on college attendance. Sixty–fourpercent of freshmen who received HOPEduring academic year 1997–8 lost theirscholarships the following year.44 The col-lege GPA requirement therefore culls fromHOPE eligibility not just those who can’thandle college, but the median college stu-dent. Further, the college GPA requirementappears to hit blacks harder than whites.Blacks at the University of Georgia aretwice as likely as whites to lose their schol-arship after the freshman year (Healy,1997). A recent study of students at theGeorgia Institute of Technology also foundthat blacks were substantially more likelythan whites to lose their scholarships,though this differential disappeared afteraccounting for differences in ability asmeasured by SAT scores.45

The substantial rate of attrition fromHOPE may explain a result observed ear-lier: as seen in Figures 1 and 2, the effectof HOPE on college attendance appearsto have dropped in recent years. It is pos-sible that young people on the margin ofcollege attendance have observed the veryhigh rate at which their older peers havelost their HOPE Scholarships and decidedthat the expectation of one year of free

tuition is not enough to make collegeworthwhile.46

Other Differences between the Federaland Georgia Programs

There are two more key differences be-tween the Georgia and federal subsidyprograms. The first is the conditions intowhich these two programs have been in-troduced. The college attendance rate inGeorgia when HOPE was introduced wasmuch lower than that in the rest of the U.S.A large reservoir of youth not attendingcollege may have contributed to theprogram’s large impact. It is likely that theeffect of the federal Hope Scholarship willvary geographically, producing a largerimpact in states where attendance is lowand a smaller impact where attendance ishigh.

Second, the federal program is nationalin scope: a person can use the credit any-where in the U.S. By contrast, Georgia re-quires that a student stay in the state inorder to receive the scholarship. However,the vast majority (83 percent) of freshmenattends college in their home state.47 Thisfigure is likely even higher for youth onthe margin of college attendance. As a re-sult, this particular difference between thetwo programs is most likely inconsequen-tial to their relative impact on college at-tendance rates.

Predicted Net Impact of the FederalHope Scholarship

Most of the differences between thetwo programs point to the federal HopeScholarship having a lesser impact on col-lege attendance than has Georgia HOPE.

44 Data from Steve Thomkins of the Georgia Student Finance Commission.45 Dee and Jackson (1999).46 It is true that the number of HOPE Scholars has risen steadily over time, which would seem to contradict the

finding that its effect on attendance is diminishing. However, HOPE take–up can rise due to two shifts instudent behavior that have no impact on the college attendance rate: students who would have gone to col-lege anyway choose to attend in Georgia and students who would have gone to college anyway increase theirhigh school grades marginally so as to meet the HOPE academic requirement.

47 U.S. Department of Education (1998a), Table 203.


Hope for Whom?

659

The federal Hope Scholarship delivers itssubsidy after tuition has been paid; is ad-ministered through the complex U.S. taxcode; threatens to push up tuition prices;and is being offered to a U.S. populationwhose college attendance rates are alreadysubstantially higher than those in Geor-gia. These characteristics combine to re-duce the effect of the federal program rela-tive to that of Georgia’s. The results there-fore provide an upper bound on the im-pact of the federal Hope Scholarship onrecent high school graduates, suggestingthat each $1,000 in tax credits could in-crease their college attendance rate by asmuch as 4 percentage points.

The Georgia experience indicates thatany impact of the federal Hope Scholar-ship on college attendance will come withthe price of exacerbating already substan-tial racial and income gaps in collegeattendance. In Georgia, the HOPE Schol-arship has increased overall collegeattendance but widened the gap in atten-dance rates between whites and blacksand between rich and poor. The resultsin Table 6 and Table 7 indicate that,since HOPE was introduced, gaps in col-lege attendance between blacks andwhites and between upper– and lower–income youth in Georgia have risensubstantially more than they have in therest of the Southeast. Nationwide, thegap in attendance rates between recenthigh school graduates in the bottomand top quartiles of the family incomedistribution is 30 percentage points.Even after controlling for ability, as mea-sured by standardized test scores, this gapremains quite large: among the middlethird of test scorers, the gap betweenhigh– and low–income youth is 22 per-cent. Further, differences in college atten-dance across income groups have beengrowing over time.48 Programs that pri-marily subsidize the college attendance ofmiddle– and upper–income youth, like

the federal Hope Scholarship and Geor-gia HOPE Scholarship, will only exacer-bate this trend.

CONCLUSION

The federal government and the stateshave recently enacted a slew of new stu-dent aid programs aimed at youth frommiddle– and high–income families. Therehas been little research on the sensitivityto college costs of this group’s attendancerates. In this paper, I estimate the impactof aid on the college attendance ofmiddle– and upper–income youth byevaluating the Georgia program that is thenamesake and inspiration of the new fed-eral Hope Scholarship: the Georgia HOPE(Helping Outstanding Pupils Education-ally) Scholarship. The results suggest thatGeorgia’s program has had a surprisinglylarge effect on the college attendance rateof middle– and high–income youth. Us-ing a set of nearby states as a controlgroup, I find that Georgia’s program haslikely increased the college attendancerate among 18– to 19–year–olds by 7 to 8percentage points.

I further find that the program’s effectis concentrated among Georgia’s whitestudents, who have experienced a 12.3percentage point rise in their attendancerate relative to whites in comparisonstates. The black attendance rate in Geor-gia has not increased relative to that incomparison states since HOPE was intro-duced. The racial gap in college atten-dance in Georgia has therefore increasedrelative to its level in the rest of the South-east. The evidence also suggests thatGeorgia’s program has widened the gapin college attendance between those fromlow–income and high–income families.The federal Hope Scholarship, which fo-cuses on the same slice of the family in-come distribution as Georgia’s program,is also likely to exacerbate already large

48 The figures on college attendance, family income, and test scores are from Ellwood and Kane (1999).



660

racial and income gaps in college atten-dance in the U.S.

Acknowledgments

I appreciate the helpful commentsof Daron Acemoglu, Joshua Angrist,David Autor, Julie–Anne Cronin, AmyFinkelstein, Jonathan Gruber, ThomasKane, Steve Pischke, James Poterbaand participants of the NBER SummerInstitute and Economic Effects of Taxationmeetings. Support from NBER Aging andHealth Pre– and Post–Doctoral Fellow-ships and the Smith–Richardson Founda-tion is gratefully acknowledged.

REFERENCES

Basinger, Julianne, and Patrick Healy.“Will New Federal Tax Breaks HurtCalifornia’s Colleges?” The Chronicle ofHigher Education (March 6, 1998): A36.

Byron, Kris, and Gary Henry.“Report on the Expenditure of LotteryFunds Fiscal Year 1996.” Georgia StateUniversity. Unpublished Manuscript,1996.

Byron, Kris, and Gary Henry.“Report on the Expenditure of Lottery FundsFiscal Year 1997.” Georgia State University.Unpublished Manuscript, 1997.

Byron, Kris, and Gary Henry.“Report on the Expenditure of LotteryFunds Fiscal Year 1998.” Georgia StateUniversity. Unpublished Manuscript, 1998.

Cameron, Stephen, and James Heckman.“Can Tuition Policy Combat Rising Wage In-equality?” In Financing College Tuition: Gov-ernment Politics and Educational Priorities, ed-ited by Marvin Kosters. Washington, D.C.:American Enterprise Institute, 1999.

College Board.Trends in Student Aid. New York: CollegePublications, 1998.

Dee, Thomas, and Linda Jackson.“Who Loses Hope? Attrition from Georgia’sCollege Scholarship Program.” Southern Eco-nomic Journal 66 No. 2 (October, 1999): 379–90.

Dynarski, Susan.“Does Aid Matter? Measuring the Effect ofStudent Aid on College Attendance andCompletion.” NBER Working Paper No.7422. Cambridge, MA: National Bureau ofEconomic Research, 1999.

Ellwood, David, and Thomas Kane.“Who is Getting a College Education?Family Background and the GrowingGap in Enrollment.” Harvard University.Unpublished Manuscript, 1999.

Healy, Patrick.“HOPE Scholarships Transform the Univer-sity of Georgia.” The Chronicle of Higher Edu-cation (November 7, 1997): A32.

Hebel, Sara.“IRS Delays Reporting Requirements forTwo New Tax Credits for College Costs.”The Chronicle of Higher Education (August 6,1999): A36.

Henry, Gary, Steve Harkrender, PhiloHutcheson, and Craig Gordon.

“Hope Longitudinal Study, First-Year Re-sults.” Georgia State University. Unpub-lished Manuscript, August, 1998.

Hollenbeck, Scott, and Maureen Kahn.“Individual Income Tax Returns, 1997: EarlyTax Estimates.” Statistics of Income Bulletin18 No. 3 (Winter, 1998–99): 126–50.

Jaffe, Greg.“Free for All: Georgia’s Scholarships areOpen to Everyone, and That’s a Problem.”The Wall Street Journal (June 2, 1997): A1.

Kane, Thomas.“Rethinking the Way Americans Pay for Col-lege.” Milken Institute Review (1999a).

Kane, Thomas.“College Entry by Blacks since 1970: The Roleof College Costs, Family Background, and theReturns to Education.” Journal of PoliticalEconomy 102 No. 5 (October, 1994): 878–911.

Leslie, Larry, and Paul Brinkman.The Economic Value of Higher Education. NewYork: Macmillan, 1988.

Reyes, Suzanne.“Educational Opportunities and Outcomes:The Role of the Guaranteed Student Loan.”Harvard University. Unpublished Manu-script, 1995.


Hope for Whom?

661

Selingo, Jeff.“Copying Georgia’s HOPE: States that areWeighing Tuition-Scholarship Proposals.”The Chronicle of Higher Education (April 16,1999): A37.

Stanley, Marcus.“College Education and the Mid-CenturyG.I. Bills.” Harvard University. Unpub-lished Manuscript, 1999.

U.S. Department of Education.National Center for Educational Statistics.Making the Cut: Who Meets Highly Selective

College Entrance Criteria? Washington, D.C.:Government Printing Office, 1995.

U.S. Department of Education.National Center for Educational Statistics.Digest of Education Statistics. Washing-ton, D.C.: Government Printing Office,1998a.

U.S. Department of Education.National Center for Educational Statistics.State Comparisons of Education Statistics: 1969-70 to 1996-97. Washington, D.C.: Govern-ment Printing Office, 1998b.



Documents

Hope for Whom? Financial Aid for the Middle Class and Its ...users.nber.org/~dynarski/2000 Hope for Whom.pdf · and high-income families. I estimate the impact of aid on the col-lege