nho pays and who benefits? examining the distributional...

\Nho Pays and Who Benefits? Examiningthe Distributional Consequences of the

Georgia Lottery for Education

Ross Rubenstein &Benjamin ScafidiAndreio Young Sdwolof Policy Studies,Georgia StateUniversity, Atlanta,GA 30303

National Tax JournalVol. LV, No. 2June 2002

Abstract - Tliis paper examines the incideyice of the implicit lot-tery tax and the distribution of benefits from lottery-funded pro-grams in Georgia. Georgia's lottery is unique in that revenues areearmarked for three educational programs—HOPE College Schol-arships, universal pre-kindergarten, and education infrastructure.We estimate separate models of household-level lottery purchasesand of household benefits from lottery-funded programs. Our esti-mates suggest that lower income and non-white households tendto have higher purchases of lottery products while receiving lowerbenefits, as compared to higher income and white households. Ben-efits of HOPE Sclwlarships, in particular, accrue disproportion-ately to higher income and more educated hojiseholds.

INTRODUCTION

While state-sponsored lotteries have been a part of theU.S. public finance landscape since the nation's birth

(Clotfelter and Cook, 1989), perhaps few lotteries have en-joyed the enormous popularity of Georgia's "Lottery for Edu-cation." Georgia's lottery is unique in that most lottery rev-enues are earmarked for HOPE (Helping Outstanding Pu-pils Educationally) College Scholarships and universal pre-kindergarten, both of which began with the lottery and arefunded solely with lottery revenues. Any lottery revenuesleft over are used for K-12 infrastructure spending. In the1998 campaign for Georgia governor, both major party can-didates pledged to support and continue the state's lottery-funded programs. Moreover, the enthusiasm has spread be-yond Georgia's boundaries. At least a dozen states have con-sidered or adopted programs similar to Georgia's most well-known lottery-funded program, HOPE scholarships. Legis-lators in many of these states make no secret of their debt toGeorgia's program, and some have even adopted the HOPE"brand name" (Selingo, 1999).

While it may be clear that Georgia's lottery has been anunequivocal political success (Henry and Gordon, 1999),important questions remain unanswered. This paper ad-dresses two fundamental questions surrounding the Geor-gia lottery:

223

NATIONAL TAX JOURNAL

• What is the incidence of householdpurchases of lottery products?

• 'V\T:io benefits from lottery-fundedprograms?

To answer the first question, we estimatea model of household-level lottery pur-chases using the Heckman two-step estimation procedure. We pay par-ticular attention to lottery spending be-havior by income, race, and educationlevel. To analyze the distribution of ben-efits from lottery-funded programs, weestimate lottery-funded program expen-ditures received by households fromthe three lottery-funded programs:HOPE scholarships, universal pre-kinder-garten, and K-12 infrastructure spending(technology and construction). Finally,we estimate the net budgetary incidenceof the lottery by comparing predictedhousehold lottery spending to predictedbenefits.

The next section provides backgrour\don the Georgia lottery, and the third sec-tion briefly reviews previous researchon lotteries as public revenue sources.The fourth section describes the data,while the fifth section provides a descrip-tion of the empirical models and results.Alternative financing methods for the lot-tery-funded programs and their incidenceare described in the sixth section, and theseventh section provides concluding re-marks.

THE GEORGIA LOTTERY FOREDUCATION

Georgia's lottery owes its widespreadrecognition not to any unique aspects of thelotter}' itself—Georgia's lottery operates ina similar manner to other state-run systems.Instead, it comes from the fact that Georgiaearmarks net lottery revenues for three ex-penditure areas: HOPE scholarships, uni-versal pre-kindergarten for four year-olds,and education infrastructure (technologyand construction of facilities). Table 1 dis-plays lottery expenditures by program forFY1994 through FY1999. Since the lottery'sinception, the largest share of proceeds (30percent) has been spent on pre-kindergar-ten, followed by HOPE Scholarships (29percent). The relative share of spending de-voted to HOPE and pre-kindergarten hasbeen steadily increasing, with almost 74percent of total FY 1999 appropriations ear-marked for those two programs (Brackett,Henry, and Weathersby 1999).'

As shown in Table 1, the lottery has alsoprovided substantial revenue for K-12construction and technology investment.While the lottery's early years saw sub-stantial infrastructure investments, thesehave declined over time as the HOPE andpre-kindergarten programs have grownin size. While Georgia's lottery revenuebase has been surprisingly stable, it islikely that sales wiUbegin to decline if sur-rounding states enact their own lotteries.-

' The pre-kindergarten program sen'es over 60,000 children annually in public, private, and not-for-profitchild care centers. l\'hile the program was originalK' intended for "at-risk" children, the family incomeeligibilit}' cap was lifted in 1995. All children are now eligible to attend a state provided or approved pre-kindergarten center at no cost to the child's family, though attendance is not compulsory. The HOPE collegescholarship program provides subsidies for students attending both public and private institutions of highereducation in Georgia. The Public College Scholarship component of the program provides full tuition, fees,and an allowance for books at any public institution of higher education in Georgia. The Private CollegeTuition Equalization Grant provides a grant (currently $3,000) to offset tuition at any private college or uni-versity itv Georgia. To earn either of these scholarships, a student must ha%'e a 3.0 cumulative grade pointaverage (GPA) in his or her core high school courses, and must maintain the 3.0 annually in college in order toretain the scholarship. While the program originally had an income cap for oligibilit)'. the restrictionwas lifted in F '̂ 1996 and any Georgia resident can now receive the scholarship based solely on his or hergrades.

- In 1999, Alabama voters rejected a lottery referendum, and South Carolina votes approved a lottery referen-dum in November 2000.

224

Examining the Distributional Consequences of the Georgia Lottery

TABLE 1LOTTERY EXPENDITURES, Fi' 1994-99

ScholarshipsPre-kindergartenTechnologyConstructionTotal

Amount ($)

S74,982,422928,531,837637,624,853627,617,944

3,068,757,056

Percent

29302120

100

Source: Brackett, Henry, and Weathersby, 1999.

Spurred in part by concerns over declin-ing future revenues, Georgians voted in1998 to amend the state Constitution toformally specify pre-kindergarten andHOPE as the lottery's funding priorities,leaving the future of construction andtechnology outlays in doubt.

Given the earmarking of lottery pro-ceeds to specific programs, lottery expen-ditures in Georgia are easily traced to theprograms they support. The close linkbetween the lottery-funded programs andtheir revenue source has allowed theGeorgia lottery to escape the commoncriticism that lottery revenues supplantrather than supplement other resources(Clotfelter and Cook, 1989; Borg and Ma-son, 1990; Borg, Mason, and Shapiro, 1993;Spindler, 1995), Georgia did not have astate-funded pre-kindergarten programbefore the lottery. Although many stateshave limited public funding for pre-kin-dergarten, only New York state haspledged to implement universal pre-kin-dergarten like Georgia's program. Never-theless, New York's pre-kindergarten wiUnot be universal until 2003 at the earliest(National Center for Policy Analysis,2000).

The available evidence suggests thatsupplanting of funds has been limited ornon-existent in the lottery-funded pro-grams. While Georgia provided no fund-ing for pre-school education prior to thelottery's implementation, the state did

fund some higher education student fi-nancial aid and K-12 school construction.In the three fiscal years prior to the startof the lottery, budgeted state expendituresfor student financial aid from the stateGeneral Fund averaged 0.30 percent oftotal General Fund expenditures and theaverage increased slightly to 0,31 percentin the four years after the HOPE programbegan (State of Georgia, 1997; 1996; 1995;1994; 1993; 1992; 1991), Adjusted by theConsumer Price Index, total real financialaid expenditures in Georgia from lotteryproceeds and the General Fund grew byapproximately $270 per student between1993 and 1996, while no other state in theSoutheast had an increase of more than$110 per student, and most states had de-clines (Southern Regional EducationBoard, 2000). Need-based aid in Georgia(constant dollars) remained relatively flatover this period, while the growth in non-need-based aid from the HOPE Scholar-ships accounted for the large increase,-Data are not available to track school con-struction and technology expendituresfrom the pre-lottery period. However, to-tal state General Fund expenditures on K-12 education, which include constructionand technology, grew at a faster rate thantotal General Fund expenditures (36 per-cent versus 33 percent) in the years afterthe lottery was introduced.

RESEARCH ON LOTTERiES INOTHER STATES

Thirty-seven states including Georgiacurrently run lotteries and many issuessurrounding these enterprises have beenwell documented by researchers. Rev-enues raised by lotteries can be viewed asan excise tax on one item—lottery play.Georgia retains 35 cents for every dollar

In the year examined in this paper, HOPE awards ivere reduced by the amount of any federal Pell Grantsavailable to lower income students, Tn the 2000-2001 academic year, this policy was changed to allow stu-dents to collect both HOPE Scholarships and Pell Grants.

225


spent on the lottery. These lottery pro-ceeds retained by the state represent animplicit tax on purchases of lottery prod-ucts. If a competitive market existed forlottery products, it is imlikely that eachfirm's profits would approach 35 percentof revenues. Although purchases of lot-tery tickets are "voluntary," the implicittax on a dollar spent on a lottery productis not volutitary, just as sales taxes paidon purchased goods are not voluntary.Given this implicit taxation, questions re-lated to the incidence of the implicit lot-tery tax are similar to those surroundingan income, sales, property, or other tax.Although the implicit taxes paid on lot-ter>' products are less than the full priceof the lottery product, the implicit lotterytax rates are much higher than sales taxrates on most other items.

The incidence of the lottery's implicittax has been one of the most frequentlystudied aspects of state lotteries. Mikesell(1989), using aggregate data from Illinois,examines county lottery sales over athree-year period and finds income elas-ticities close to one, suggesting a constant"tax" rate over all income levels and there-fore a proportional burden. In contrast.Price and Novak (1999), using zip code ag-gregated data, find that Texas' lotterygames are substantially more regressivethan sales taxes, and that the instant and"numbers" games are the most regressive.In their study, the percentage of the popu-lation that is African-American is posi-tively related to sales of the most regres-sive games.

Other studies, which use household-level data, find lotteries to be highly re-gressive revenue generators. For exam.ple,Clotfelter and Cook (1987; 1989; 1990a;1990b), using data from several states, findthat lottery expenditures, net of averageexpected winnings, decrease as educationlevels rise, and that African-Americanconsumers spend significantly more thanwhites. They find little systematic relation-ship, though, between household income

and lottery spending. This relatively flatspending across income groups producesa regressive implicit tax burden, withlower income households spending alarger share of their income on the lottery.Clotfelter, et. al. (1999), in a report to theNational Gambling Impact Study Com-mission, present tabulations from a na-tional survey of household lottery spend-ing and find that males, African Ameri-cans, and less educated and lower-incomehouseholds have large expenditures onlottery products. Borg, Mason, andShapiro (1991) find that spending in-creases somewhat with income, and thatthe small purchasing increases lead to aregressive incidence.

Scott and Garen (1994) use a Heckmantwo-stage estimator rather than the tobitapproach used in the previous studies toseparate the effects of demographic vari-ables on the probability of lottery partici-pation from their effects on the level ofplay, given participation. They find thatdemographic variables have differentialeffects on probabilities of play and levelsof play. Nevertheless, they also find thatthe incidence of implicit lotter}' taxationis regressive. Stranahan and Borg (1998a;1998b) review Scott and Garen's method-olog\' and suggest that a probit model toestimate the probability of participationand a truncated tobit model to estimatespending (contingent on participation)provide a better approach since lotteryspending, net of average expected win-nings, is truncated at zero. Using thismethod they again find a regressive bur-den, although more so for instant gamesthan for lotto.

An important but often overlookedquestion regarding lotteries is whether thedistribution of benefits from lottery-funded programs reduces or exacerbatesthe apparent regressivity of lottery pur-chases by players. That is, while lower-income households may bear a dispropor-tionately large share of the burden fromimplicit lottery taxation, they may also

226


receive a disproportionately small share ofthe benefits from lottery-funded programs.Borg and Mason's (1988) and Borg, Mason,and Shapiro's (1991) work is the only re-search to explore this issue. Using Floridaand Illinois data, respectively, they estimatethe direct benefits for households (calcu-lated as per-student education spendingfunded by the lottery) from the lottery inrelation to household lottery expenditures.In both states, lottery revenues are used toprovide funding for K-12 public education.They estimate the number of children perhousehold receiving public education us-ing characteristics (such as age, marital sta-tus, income, and race) tliat affect householdsize and the likelihood of public rather thanprivate school attendance. In Florida, theauthors find a positive net benefit (directbenefits minus household lottery expendi-tures) for all groups except those in the low-est income category, and that net benefitsgenerally rise with income. In Illinois theyfind that the implicit tax outweighs thebenefits of education for all income groups,particularly those at the lowest and high-est income levels.

Estimates of lottery benefits are compli-cated by the possibility that earmarkedlottery revenues supplant, rather thansupplement, other state revenues. Borgand Mason (1991) examine state expendi-tures for education in lottery and non-lot-tery states and find no evidence that lot-teries led to increases in state spending.Spindler (1995) uses an ARIMA model toexplore the fungibibty of lottery revenuein seven states that earmark revenues foreducation and finds that lotteries pro-duced negative cumulative effects oneducation spending in four states, al-though the effects varied across states andyears.

The dearth of research on the distribu-tion of lottery benefits relative to house-hold purchases may be a result of the rela-tive difficulty of estimating the net ben-efits of broadly defined public programssuch as education, particularly when lot-tery revenues supplant other state spend-ing. Georgia's earmarking of revenues fornewly created programs that are fundedexclusively with lottery revenue, though,facilitates our incidence analysis.

DATA

The data used to estimate equationsexplaining the probability of playing thelottery, net lottery expenditures condi-tional on playing, and household benefitsfrom lottery-funded programs come froma variety of sources. Each data set is ex-plained below.

Household Survey Data

To estimate household spending on lot-ter)' products we use household surveydata collected from a stratified randomsample of 803 adult residents of the stateof Georgia.'' The survey data wereweighted such that the weighted samplemeans would mirror Census Bureau esti-mates for the entire state of Georgia.Among the 803 respondents, we were ableto use 548 cases in the analyses in thispaper. Hereafter, we refer to these 548cases as the "usable" sample. Most of the255 other respondents were dropped dueto missing data, primarily household in-come. Table 2 displays descriptive statis-tics for the sample.

To check whether our usable sample of548 cases remains representative of thestate, we compare our weighted usable

' The sun'ey data come from the quarterly Georgia Poll conducted by the Applied Research Center at GeorgiaState University. All households with a working telephone are eligible for the poU as the sample includes un-listed as well as listed telephone numbers. Using the American Association for Public Opinion Research's defi-nitions and criteria (APOR, 1999), the response and cooperation rates for the sur\'ey were excellent. Of aU theworking residential telephone numbers called, 49.5 percent resulted in completed interviews (response rate). Inaddition, among persons who answered the phone, 78.2 percent participated in the sur\'ey (cooperation rate).

227


TABLE 2GEORGIA StATE POLL DESCRIPTIVE StA"rrStlCS'(S:'="5i8l

Variable Definition Mean Std. Dev. Min Max

NET_SPENDING household lottery spending minus winnings 81.53NONWHITE equals 1 if respondent non-white 0.26INCl income < $15K 0.13INC2 income >= $15K and < S25K 0.10INC3 income >= $25K and < $35K 0.26INC4 income >= $35K and < S50K 0.19INC5 income >= $50K and < $75K 0.18INC6 income >= $75K 0.14EDUCATION years of educarion 13.66.AGE age in years 42.75VOTER equals 1 if registered voter 0.80RELIGIONl attends religious ser\'ices every week 0.42REPUB equals 1 if republican 0.24IKDEP equals 1 if independent 0.39CHILDREN equals 1 if children under 18 in household 0.44OWNER equals 1 if owner-occupant 0.79EMPLOYED equals 1 if employed 0.71ATLANTA equals 1 if resident of metro Atlanta 0.32PLAY equals 1 if plays the lottery 0.33

Source: January 1999 Georgia Poll, quarterly poU conducted by the Applied Research Center (ARC) at GeorgiaState Uruversit}'. Results are weighted by the household weights provided by the ARC.

396.720.440.330.300.440.400.380.352.1416.350.400.490.430.490.500.410.460.470.47

^,00000000001018000000000

1,6321111111IS921]1111111

sample means to the Census Bureau esti-mates for Georgia. Approximately 74 per-cent of the usable sample is white and 26percent non-white, compared to the Cen-sus Bureau's estimate of 70 percent whiteand 30 percent non-white residents inGeorgia (U.S. Bureau of the Census, 1998).The mean years of education is 13.66,equivalent to a high school degree withsome college education, compared to theCensus estimate of 13.12 years (Table 3).The largest group of respondents (26 per-cent) report household income in the$25,000-$35,000 income range, followedby the $35,000-$50,000 category (19 per-cent). CK'erall, the usable sample's char-acteristics appear very similar to CensusBureau estimates of Georgia's population.

Each respondent was asked about hisor her household's lottery spending hab-its and winnings. Using this informa-tion, we constructed the variableNET_SPENDING, which equals house-hold lottery spending minus householdlottery winnings.^

[1] NET_SPENDING = annual spendingon lottery products - annualwinnings.

The mean of NET_SPENDING is about$82 per household per year." About 33percent of respondents report playingthe lottery. For these households,mean NET_SPENDING is about $250 peryear.

Other studies (Borg and Mason, 1988; Stranahan and Borg, 1998a, 1998b; Clotfelter and Cook, 1989; Scott arid Garen,1994) use spending on lottery products minus average expected winnings as their measure of lotteryspending. Thesurvey used in this study asked respondents how much the household had won from the lottery in the past year.We omit from the sample eight households who reported over $2,080 per year in lottery expenditures (spend-ing before winnings). The minimum lottery expenditures of these eight obsen-ations is almost $5,000. Nohouseholds report net spending between 82,080 and $5,000 per year. Given this natural break in the data andthe estimation problems they cause, we drop these eight observations from the analysis. The meanNET_SPENDING for our sample ($82) is low, but if respondents systematically underreport lotterj' purchases,the data should not biased. Herring and Bledsoe (1994) suggest that underreporting of lotterypurchases iusurvey data is common, but that the underreporting may be widespread across sociodemographic groupsand therefore would not bias analyses of distributional effects (Herring and Bledsoe, 1994). Including theoutliers, the mean of NET_SPENDING rises to a level that is too large to be considered a reasonable estimate.

228


TABLE 3, STATE LOTTERY EXPENfDITURES AND DEMOGRAPHICS BY COUNTY (N = 159)

Variable Definition Mean Std, Dev, Minimum Maximum

EXP Total lotterj' expenditure perhousehold (FY1998)

EXPl HOPE expenditures per householdEXP2 Pre-k expenditures per householdEXP3 K-12 lottery exp, per householdMEDU Mean Years of EducationMINC Mean household income (000)% NONWHITE Non-white (%)

$216,32 83,49 66,80 815,42

$73,75$84,10S58,47

13,12$48,982

29,98

27,3225,5260,76

,6711,58017,16

15,760

9,5511,80

$25,124,38

199,05174,77623,0414,16

$75,01679.77

Note: Data weighted by number of households per county. Data on FY 1998 lottery- expenditures come from thestate of Georgia, Demographic data come from the 1990 Census of Population and Housing,

Lottery-funded Program Data

Data on lottery-funded program expen-ditures, hereafter lottery expenditures,were collected from various Slate of Geor-gia sources. All lottery expenditure dataare county-level aggregates for FY 1998.Technology and construction allocations toschool districts are supplied by the Geor-gia Office of Planning and Budget, HOPEScholarship allocation data are from theGeorgia Student Finance Commission, andpre-kindergarten data are from the Geor-gia Office of School Readiness. Institutionsof higher education receive payments fromthe state Board of Regents for each enrolledHOPE scholar, and we allocate the HOPEbenefit to each student's county of homeresidence (rather than college location).The pre-kindergarten program ser\'es stu-dents in both public and private pre-kin-dergarten centers, and the Office of SchoolReadiness supplies data on allocations topublic school systems and to private pro-viders. In addition, we merge the lotteryexpenditure data with county-level demo-graphic data from the 1990 U.S. Census(Census Bureau, 1998),

Table 3 displays descriptive statistics forthe variables used in the model. As shown,household lottery benefits (expendituresreceived) average $216,32 per county, witha range of approximately $750 across coun-ties. On average, households receive thelargest benefits from the pre-kindergartenprogram, followed by HOPE Scholarships,

EMPIRICAL MODELS TO ESTIMATETHE NET BUDGETARY INCIDENCE

We estimate the net budgetary inci-dence of the Georgia Lottery for Educa-tion by the following six steps:

i. Estimate a model that relates house-hold lottery spending (NET_SPENDING) to household charac-teristics. This model is estimatedusing the household-level surveydata.

ii. Use the estimates from step 1 to pre-dict lottery spending for each house-hold in the survey data.

iii. Estimate a model that relates the lot-tery-funded program expenditures

In addition, by excluding these eight observations, our usable sample compares favorably to the survey dataused in Clotfelter and Cook (1989) and Scott and Garen (1994), Data reported in these two papers allowcomparisons with our usable sample: the mean of lottery expenditures conditional on plapng the lottery isS393 in Scott and Garen, $347 in Clotfelter and Cook, and $392—spending before winnings—in our usablesample (sample means were adjusted for inflation using the Bureau of Labor Statistics' CPI_U), If we includethe eight outliers, this figure rises to $676, In addition, Clotfelter and Cook report that in their sample, the top10 percent of lottery players accounts for 65 percent of lottery expenditures. In our usable sample, the top 10percent of lottery players accounts for 67,3 percent of lotterj' expenditures. If we include the eight outliers,this figure would rise to 84 percent. Four outliers are white, and four are non-white. In addition, four of theoutliers report incomes under $50,000, and four report incomes over $50,000,

229


to household characteristics. Thismodel is estimated with the count}'level data.

iv. Use the estimates from step 3 to pre-dict "benefits" from lottery-fundedprograms for each household in thesurvey data.

V. Subtract the predicted value ofNET_SPENDING from the pre-dicted benefits:

[2] NETJBENEFIT = BENEFIT-NET_SPENDING.

vi. Compare NET_BENEFIT across in-come, education, and racial groupsto analyze the net budgetary inci-dence of the Georgia Lottery forEducation.

Wiio Pays?

This section explores who plays theGeorgia Lottery and how much theyspend. Following Scott and Garen (1994),we estimate a model of household lotteryspending using the Heckman two-stepestimation procedure (Heckman, 1979).Other studies have used each household'sobserved level of lottery spending minusa measure of average expected winningsto measure net lottery spending by house-holds—this measure is always positive,which suggests a truncated tobit approachis appropriate (Stranahan and Borg, 1998).In our data, NET_SPENDING equals re-ported annual lottery spending minus re-ported lottery winnings. Thus, a house-hold can have negative NET_SPEND]NG,when it wins more thaii it loses. SinceNET__SPENDING can be negative, theHeckman two-step approach is appropri-ate for our estimation, rather than a trun-cated tobit approach.

The average expected winnings equalsthe mean payout rate for the lottery. Asfound in other studies, the propensity toplay particular lottery games is related tohousehold demographic characteristics

and the odds of winning differ consider-ably across games. Thus, expected win-nings will differ systematically across de-mographic groups. For example,Stranahan and Borg (1998b) found thatlower income households were morelikely to play instant games relative tohigher income households. Since we havethe data on household winnings availableto us, we use that information—ratherthan average expected winnings—tocompute net household lottery spending.

Lottery spending, net of winnings{NET_SP^ENDING), is decomposed intothe probability of playing the lottery mul-tiplied by lottery expenditures conditionalon playing the lottery:

[3] NET_SPENDING. = E[PLAy;\ *

E[NET_SPENDING^ IPLAY^ = 1],

where PLAY, equals 1 if household; playsthe lottery and 0 otherwise. If X̂ and Z^are vectors of household characteristics,where X, e Z_, and a and ji are vectors ofparameters, then the model explaining nethousehold lottery spending is:

;r,,.

[4] ,

NET_SPEND1NG =fiX,

where

/ Q{aZ.),

where the ;!:'s are jointly normally distrib-uted, 0(») is the standard normal densityfunction, and Q{») is the standard normaldistribution function.

Probit and OLS estimates from thismodel are listed in Tables 4 and 5, respec-tively The NET_SPENDING equation isidentified by the non-linear probit estima-tor and measures of attendance at reli-gious services and political affiliation.Following Scott and Garen (1994), thisspecification assumes that regular atten-dance at religious sendees and party af-filiation are likely to affect an individual'slikelihood of playing the lottery, but do

230


TABLE 4PROBIT RESULTS FOR LIKELIHOOD OE PLAYING THE LOTTERY

Variable Definition Coefficient Std.Error t-statistic

ConstantNONWHITE1NC2INC3IN'C4INC5IN;C6EDUCATIONAGEVOTERRELIGIONlREPUBINDEPCHILDRENOWNEREMPLOYEDATLANTA

E-statistic

equals 1 if respondent non-whiteincome >= S15K and < S25Kincome >= S25K and < S35Kincome >= S35K and < S50Kincome >= $50K and < S75Kincome >= $75Kyears of educationage in yearsequals 1 if registered voterattends religious sen'ices every weekequals 1 if republicanequals 1 if independentequals 1 if children under 18 in householdequals 1 if o^\Tier-occupantequals 1 if employedequals 1 if resident of metro Atlanta

3.93

0.1550.948

-0.0420.2940.5040.9660.780

-0.054-.005

-0.275-0.333-0.146-0.012-0.113

0.3680.0350.058

0.4S61.4400.2780.2270.2360.2480.2660.030

.0040.1530.1210.1640.1360.1250.1670.1520.131

0.330.660.151.302.13"3.89"*2.94*"1.76*1.051.80"2.75'"0.890.090.902.21"0.230.44

Note: Dependent variable equals PLAY. N = 548.•"Significant at p < .01•* Significant at p < .05* Significant at p < .10

TABLE 5SECOND-STAGE OLS RESULTS EOR LOTTERY SPENDING (LOTTERY PLA'i'ERS ONLY)

Variable Definition Coefficient Std. Error t-statistic

ConstantNONWHITE equals 1 if respondent non-whiteINC2 income >= $15K and < $25KINC3 income >= $25K and < $35KINC4 income >= $35K and < $50KINC5 income >= $50K and < $75KINC6 income >= S75KEDUCATION years of educationAGE age in yearsVOTER equals 1 if registered voterCHILDREN equals 1 if children under 18 in householdOWNER equals 1 if owner-occupantEMPLOYED equals 1 if employedATL.'̂ NTA equals 1 if resident of metro AtlantaLAMBDA inverse mills ratio

Adjusted W .078F-statistic 2.00

-56.37428.78

-105.13-246.27

81.16-164.43-109.80-13.46

4.71-240.29

1.47267.86149.58

-115.97241.33

670.43147.90339.37316.79375.51486.52439.9136.775.06

174.20133.25232.83164.62136.18522.23

0.082.90*0.310.780.220.340.250.370.931.380.011.150.910.850.46

Dependent variable = NET_SPENDING. N = 162.***Significant at p < .01** Significant at p < .05* Significant at p < .10

not affect the size of lottery purchases,conditional on playing."

In the probit results (Table 4), respon-dents who are non-white, have higherincomes, own their own homes, are em-

ployed, and reside in metropolitan Atlantaare more likely to play the lottery, thoughsome results are not significeintly differ-ent from zero at conventional levels. Re-spondents who have more years of edu-

We assume that indi^dduals who are more religious may not play the lottery because of moral oppositionto gambling, and that Republicans may be less likely to play because of ideological opposition or because the

231


cation, are older, attend religious servicesevery week, have children, are registeredvoters, and are not Democrats are lesslikely to play the lottery, though again,some coefficients are not significantly dif-ferent from zero.

The estimates from the probit equationare used to create the inverse mills ratio(A) used in the second stage OLS. Only lot-tery players are included in the second-stage equation. Estimates from the OLSregression (Table 5) of lottery players sug-gest that, conditional on playing the lot-tery, non-whites spend more than whitesand the results are statistically significant.The remaining coefficients are not signifi-cantly different from zero.' The estimatesfrom the first and second stage equationsare used to construct a predicted value ofNETJPENDING for households withvarious demographic characteristics.

Who Benefits?

As described above, Georgia earmarkslottery revenues for narrowly definededucational programs. Ideally, we wouldestimate benefits using data at the house-hold level measuring the characteristics ofbeneficiaries and non-beneficiaries. Un-fortunately, no such data are available forthis study. Therefore, we use aggregatedata describing educational attainment,race, and income for county residents, aswell as per-household lottery-fundedprogram expenditures by county, to esti-mate the distribution of lottery benefitsacross socio-economic groups.

To assess the incidence of lottery ben-efits, we regress total lottery expendituresper household by county {EXPENDI-TURES) on mean county income (MINC),mean educational attainment of county

residents (MEDU), and race of count}' resi-dents (%NONWHITEy.

[5] EXPENDITURES^. = ^ ^

+ 8MEDU, + S.(%NONWHITE) +;],,

where c indexes counties, the 5's are pa-rameters to be estimated, and rj is a nor-mally distributed error term.

Table 6 displays the results of the WLSregression (weighted by county house-holds). In the model using total state lot-tery expenditures per household as thedependent variable, mean income has apositive sign, indicating that higher lev-els of lottery expenditures flow to coun-ties with higher income households, ceteris•parihus. Mean education has a negativesign, indicating that, ceteris parihus, higherlottery expenditures flow to counties withless educated residents. Mean income andmean education are, of course, highly col-linear. An F-test for the joint significanceof mean income and mean education isstatistically significant at p <.O5. The per-centage of non-white residents is signifi-cant with a negative sign, suggesting thatcitizens in counties with higher percent-ages of white residents tend to receivelarger benefits from lottery-funded pro-grams. Tests for robustness of the modelusing non-linear specifications yield vir-tually identical predictions of net benefitsacross demographic groups.

Examining each lottery program indi-vidually produces similar but slightlydifferent results. Because the HOPE Schol-arship program primarily benefits stu-dents attending post-secondary institu-tions, we expect the benefits to accruedisproportionately to students from moreeducated, higher income families, since

lottery is strongly identified with Democratic former Governor Zell Miller, In the probit regressiori, the coef-ficients on both variables are negative as expected, and the religious attendance variable is significantly dif-ferent from zero. An F-test on the excluded instruments is statistically significant at p < .03.The estimated coefficient on /I is positive, as expected, and large in magnitude, but it is not statistically signifi-cant. This suggests that although we estimate a positive correiation betiveen jz^ and n,. this estimate is notprecisely measured.

232


TABLE 6WfEIGHTED LEAST SQUARES REGRESSION RESULTS

DEPENDENT VARIABLES

Constant

MINC

MEDU

%NONWHITE

Adj. R-squareF statistic

Total lottery exp.per household

(EXP)

301.04(219.78)

2.56"(1.20)

-12.67(21.16)

-145.65***(41.85)

.17111.90

HOPE expend,per household

(HXPl)

-95.61(61.93)

.114(.339)

14.55"(5.96)

-90.79*"(11.79)

.38634.06

Pre-k expendper household

(EXP2)

189.27(73.57)

.436(.403)

-9.47(7.08)

-7.54(14.01)

.0061.34

K-12 Lotter\' exp.per household

(EXP3)

207.39(169.30)

2.01**(.927)

-17.76(16.30)

^7 .32(32.24)

.0725.07

(Std. error in parentheses)N = 159.***Significant at p < .01*• Significant at p < .05* Significant at p < . 1 0

they may be more likely to attend college.The results of the analysis support thishypothesis; the coefficients estimated onMING and MEDU are positive. However,only the coefficient on education is statis-tically significant. MING and MEDU arejointly significant at p < .05, The percent-age of non-white residents is negative andsignificant, which is also consistent withexpectations.

While the pre-kindergarten program is,like HOPE, an entitlement available to allfamilies, there is no clear a priori assump-tion regarding the incidence of the ben-efits. The results in Table 6 suggest a weakrelationship between pre-kindergartenallocations and county characteristics.Mean household income is positive whileeducation and percentage of non-whiteresidents are negative, but none is statis-tically significant and the equation's ad-justed R- is only .006. Finally, Table 6 dis-plays the results of the model using K-12(technology and construction) expendi-tures as the dependent variable. Meanhousehold income is positive and signifi-cant atp< .05, while mean education andpercent non-white are both negative and

are not statistically significant. MINC andMEDU are jointly significant, p < .05.

We use the results from the second col-umn of Table 6 (total lottery-funded pro-gram expenditures per household) to con-struct a predicted value of lottery-fundedprogram expenditures {BENEFIT) for eachhousehold in the survey data.

Net Budgetary Incidence

Using the predicted benefits and thepredicted net spending for each house-hold, we can compute net benefits fromthe Georgia Lottery for Education:

[6] NET_BENEFIT = BENEFIT-NETJPENDING.

As shown in Table 7, we estimate thatthe households in Georgia receive an av-erage NET_BENEFIT of about $50 fromthe Georgia Lottery for Education. It ispossible that the actual averageNET_BENEFIT for Georgia residents ispositive because residents of other statespurchase lottery products, but residentsof other states do not receive any direct

233


TABLE 7DISTRIBUTION OF HOUSEHOLD SPENDING ON LOTTERY PRODUCTS, BENEFITS

OF LOTTERY FUNDED PROGRAMS, AND THE NET BUDGETARYINCIDENCE OF THE GEORGL^ LOTTERY FOR EDUCATION

GroupMean Predicted

NETJPENDINGMean Predicte'd

BENEFITMean PredictedNETJENEFIT

All Households

WhitesNon-Whites

Income < $15KIncome >= S15K and < $25KIncome >= S25K and < S35KIncome >= $35K and < $50KIncome >= $50K and < $75KIncome >= $75K

Less than high schoolHigh school gradsSome collegeCollege gradsGraduate / professional

$155.52

$132.99$220.68

$270.84S323.16

S90.45$236.57S143.62-$39.46

$162.29$132.35$164.94$172.22$136.58

S205.12

S248.39$80.01

$110.29$138.35$169.89$196.15S257.43S344.43

S185.43S225.99$196.99$194.36$231.22

$49.60

$115.40-S140.67

-$160.55-$184.81

$79.44-S40.42S113.81S383.89

$23.14$93.64$32.05$22.14$94.65

benefits from the lottery-funded pro-grams.

Table 7 also shows the weighted meanvalue of net benefits for households in dif-ferent socio-demographic groups. Thepoint estimates suggest a clear pattem ofnet benefits, although differences acrossgroups are not statistically significant. Weestimate that whites, on average, receivea substantially higher level of expectedexpenditures from lottery programs thannon-whites—the predicted mean of BEN-EFIT equals $248.39 for whites and $80.01for non-whites. In addition, non-whitehouseholds are estimated, on average, tospend more than white households onpurchasing lottery products—the pre-dicted mean of NET_SPENDING equals$132.99 for whites and $220.68 for non-whites. Thus, for the net budgetary inci-dence of the lottery, non-white householdsare predicted, on average, to spend ap-proximately $141 more on lottery productsthan they receive in benefits while whites,on average, are predicted to receive a posi-tive net transfer of about $115 per year.

Among income groups, householdsearning more than $35,000 per year are es-timated to spend less on lottery productsthan households earning less than $25,000

per year. Except for the $25,000 to $35,000income category', spending on lottery' prod-ucts is not only regressive, but higher in-come groups tend to spend a lower abso-lute amount on lottery products than dolower income groups. State lottery-fundedprogram expenditures (BENEFIT) are esti-mated to rise with income. Householdsearning less than $25,000 and between$35,000 and $50,000 are estimated to re-ceive, on average, negative net benefits(higher household lottery spending thanbenefits received) from the Georgia Lottery.All other income groups have net benefitsthat are, on average, predicted to be posi-tive. Households earning between $15,000and $25,000 per year are predicted to re-ceive the lowest net benefits on average peryear (-$184.81), primarily because thisgroup is estimated to have the highestmean NET_SPENDING ($323.16) and sec-ond lowest BENEFTT ($138.35). The larg-est net benefits accrue to the highest in-come households (income above $75,000).

Surprisingly, NET_SPENDING does notdecrease as years of education increases.For example, high school graduates areestinnated to spend slightly less on lotteryproducts than respondents with higherlevels of educational attainment. The ear-

234


marking of lottery revenues for HOPEscholarships and pre-kindergarten maymake more highly educated Georgia resi-dents more likely to purchase Georgia lot-tery tickets because of the direct benefitsthey receive. We do find that more highlyeducated persons are more likely to pur-chase lottery products than is typicallyfound in other studies (Borg, Mason, andShapiro, 1991; Stranahan and Borg, 1998).

Alternative specifications for estimatingthe NET_SPENDING regressions and thepattems of NET_BENEFITS suggest thatthe results are highly robust across speci-fications. For example, estimating theNET_SPENDING equation using lotterypurchases minus average expected (ratherthan actual) winnings produces a similardistribution of NETJENEFITS, thoughNETJBENEFITS rise even more sharplywith education levels. Including theNET_SPENDING outliers with imputedspending values (equal to the mean ofNET_SPENDING plus two standard de-viations) also produces very similar pat-terns of NETJBENEFITS across demo-graphic groups.'

ALTERNATIVE FINANCiNG FORLOTTERY-FUNDED PROGRAMS

Despite the estimated regressivity ofimplicit lottery taxation, most states havelotteries. Why? Our data, and data used inother household level studies of lotteries,suggest that a minority of households bearany burden from the implicit lottery tax,because most households do not play thelottery. In addition, a small minority of lot-tery players bears the majority of the lot-

tery tax burden. Since a small minority ofthe population bears much of the burdenof implicit lottery taxation and because itis generally illegal for a private entity torun a lottery, both lottery non-players andplayers may consider themselves betteroff with a lottery than without one. Non-players may consider themselves better offbecause they may derive a benefit from lot-tery-funded government services, andlottery players may consider themselvesbetter off with a lottery because they get toplay lottery games they enjoy.

In this political economy environment,Georgia has chosen to finance new edu-cational programs (HOPE, pre-k, and K-12 infrastructure) through dedicatedfunding from lottery proceeds rather thanthrough increases in broad-based taxes.The estimates in the previous section sug-gest that the net budgetary incidence ofthe Georgia Lottery is regressive. Target-ing the lottery-funded programs to lowincome households would help to reducethe overall regressivity of the GeorgiaLottery. Other methods to reduceregressivity include elinmiating market-ing, providing information to potentialplayers (warning labels), and increasingpayout rates (Clotfelter, et, al., 1999).

Georgia could also reduce the overallregressivity of the lottery-funded pro-grams by focusing on the revenue source—it could increase state sales and/or in-come taxes to replace revenues generatedby its lottery. To raise an amount of rev-enue equivalent to what is currently raisedby its lottery, we estimate that Georgiacould raise its sales tax rate from 4,0 per-cent to 4.69 percent,'" This additional sales

' For example, using the imputed values of NETJPENDING for the outliers in the analysis leads to a differenceof S284 betiveen the estimated NETJBENEFTTS that accrue to white and non-white households. The twolowest income groups have negative estimated NETJBENEFITS, the next three higher income groups haveNET_BENEFITS between $100 and S200, while the highest income group is estimated to have NETJBEKEFITSover S300, Finally, net benefits are estimated to rise with education levels, although the highest-educationgroup has net benefits lower than those of the next highest education group. This pattern of NETJENEFITSis very similar to the pattern reported in Table 7,

'- This computation is based upon an estimate of the elasticity of the sales tax base in Georgia found in Hawkins(1996), Hawkins estimated the elasticity to be -0,3 for Georgia, Estimating such elasticities using current tech-niques is controversial, and there are important methodological issues to consider (Merriman and Skidmore, 2000),

235


tax of approximately seven-tenths of acent on the dollar represents an increaseof 17.25 percent over current rates. Someof the burden of a sales tax increase inGeorgia would be exported to residentsof other states, just as some of the burdenof revenues raised by the lottery is ex-ported to otlier states. Although Georgia'ssales tax is particularly regressive becauseser\'ices are exempt from taxation, esti-mates suggest that the regressivity of salestaxation is less than that of implicit lot-tery taxation (Clotfelter and Cook, 1989;Price and Novak, 1999).

Another revenue neutral alternative isfor Georgia to raise its average state in-come tax rate from 3.0 percent to 3.385percent—a 12.8 percent increase in thestate income tax burden." An increase inthe income tax is likely to be less regres-sive than an increase in the sales tax, andfar less regressive than household spend-ing on lottery products. Given the lowermarginal tax rates, both sales and incometaxation would be much more efficientrevenue sources than implicit lottery taxa-tion {Clotfelter and Cook, 1989).

CONCLUSIONS

The Georgia Lottery for Education pro-vides a unique opportunity to directlyexplore not only the conventional ques-tions surrounding the incidence of house-hold lottery spending but to also exam-ine how spending relates to the benefitsreceived by households. While the vastmajority of research has found lotteries tobe a highly regressive method of raisingrevenue, it is important to also examinehow the benefits of lottery-funded pro-grams are distributed. Consistent withnumerous other studies, we find thatspending on lottery products is highlyregressive. We also find that higher in-

come households tend to receive a higherlevel of benefits from lottery-funded pro-grams than do lower-income households,though these benefits represent a higherproportion of income to lower-incomehouseholds. Taken together, we find ahighly regressive pattern of net benefits.Lower income households (those report-ing under $25,000 annual income) spendmore on the lottery than they receive inbenefits, while higher income households(those reporting over $50,000 annual in-come) receive a positive net benefit. Whitehouseholds tend to spend less on playingthe lottery than non-white householdsbut receive substantially higher benefitsfrom lottery-funded programs, on aver-age.

Our results suggest that much of theregressivity of net benefits is caused bypatterns of spending on lottery productsand by the HOPE Scholarship program.Since the program provides subsidies tostudents pursuing higher education, it isnot surprising that more educated, higherincome families would receive a dispro-portionate share of the benefits. The ben-efits of the other lottery-funded programs—particularly pre-kindergarten—appearto be only weakly related to race, educa-tion, and income.

If the publicity surrounding HOPEscholarships spurs more lower incomestudents to apply to and attend college,then our results may underestimate theactual benefits of the program to studentsfrom low-income families. It is also wortlinoting that students are required to ap-ply for Pell Grants at the same time asHOPE and in FY 1998 any Pell awardswere used to offset HOPE grants. Begin-ning in FY 2001, students will be able tocollect both HOPE and Pell Grants, andthis policy change may reduce theregressivity of net benefits by providing

This calculation assumes an elasticity of the income tax base with respect to income taxation of -0.1. Estimatesof labor supply elasticities for adult males are typically around 0, while elasticities for other demographicgroups (especially married women) suggest that there will be some erosion of the income tax base with a risein income taxes. See Pencavel (1999) and Killingsworth and Ileckman (1999) for good sur\-eys of the topic.

236


more state funding to lower-incomestudents.

Voters face choices about alternativerevenue sources for public programs. InGeorgia, HOPE Scholarships and univer-sal pre-kindergarten have become closelyidentified with their funding source—theGeorgia Lottery—and these programs areextremely popular. It is important,though, that voters have adequate infor-mation about the implications of variousfunding decisions. This paper providescitizens and policymakers (includingthose in other states seeking to emulateGeorgia's experience) with information toassess both the incidence of lottery pur-chases and the benefits of lottery fundedprograms.

Acknowledgments

The authors wish to thank CharlotteSteeh and staff at the Applied ResearchCenter for providing us with sur^'ey data,Edmund Sauer for expert research assis-tance, James Aim, Douglas Holtz-Eakin,David Sjoquist, Sally Wallace, Judy Temple,Dennis Zimmerman, seminar participantsat the annual conference of the NationalTax Association in Atlanta, GA (October1999), and two anonymous referees forhelpful comments, and the Eiscal ResearchProgram at Georgia State University forassistance and support. All responsibilityfor errors and omissions is ours,

REFERENCES

American Association for Public OpinionResearch,

Stmidari Definitions: Final Dispositions of CaseCodes and Outcmyie Rates for RDD TelephoneSurveys and In-Person Household Surveys.Ann Arbor, MI, 1999,

Borg, Mary."Earmarking Lottery Revenues: PositiveWindfalls or Concealed RedistributionMechanisms?" Journal of Education Finance15 No, 3 (Winter, 1990): 289-301,

Borg, Mar}', and Paul M, Mason,"The Budgetary Incidence of a Lottery toSupport Education." Natiaiml Tax Journal 41No, 1 (March, 1988): 75-85,

Borg, Mary O,, Paul M, Mason, and StephenL, Shapiro,

The Economic Consecjjiences of State Lotteries.New York: Praeger, 1991,

Borg, Mar\', Paul Mason, and Stephen L,Shapiro,

"The Cross Effect of Lottery Taxes on Alter-native State Tax," Public Finance Quarterly21 No, 2 (April, 1993): 123^0,

Brackett, Margaret H,H,, Gar}- T, Henry, andJeanie Weathersby,

Report on the Expenditure of Lottery Funds Fis-cal Year 1999. Atlanta: CouncU for School Per-formance, Georgia State Universit)', 1999,

Clotfelter, Charles T., and Philip J, Cook."Implicit Taxation in Lottery Finance," Na-tional Tax Journal 40 No, 4 (December, 1987):533-46,

Clotfelter, Charles T, and Philip J, Cook,Selling Hope: State Lotteries in America. Cam-bridge, MA: Harvard Universit)' Press, 1989,

Clotfelter, Charles T, and Philip J, Cook,"On the Economics of State Lotteries," Jour-nal of Economic Perspectives 4 No, 4 (Fall,1990a): 105-19,

Clotfelter, Charles T, and Philip J, Cook,"Redefining 'Success' in the State LotteryBusiness," Journal of Policy Analysis and Man-agement 9 No, 1 (Winter, 1990b): 99-104.

Clotfelter, Charles T, Philip J, Cook, Julie A,Edell, and Marian Moore,

"State Lotteries at the Tum of the Century:Report to the National Gambling ImpactStudy Commission," June 1,1999,

Hawkins, Richard,A Household Model of the Determinants of theIncome Elasticity for tlie Sales Tax. Ceorgia StateUniversity, Unpublished dissertation, 1996,

Heckman, James,"Sample Selection Bias as a Specification Er-ror," Econometrica 42 No.l (January, 1979):153-61,

Henry, Car}' T, and Craig Cordon,"Georgia's HOPE Scholarships: Good Poli-tics?" Ceorgia State University, Mimeo, 1999,

237


Herring, Mar}', and Timothy Bledsoe."A Model of Lottery Participation: Demo-graphics, Context and Attitudes." PolicyStudies Journal 22 No.2 (April, 1994): 245-57.

Killingsworth, Mark R., and James J. Heckman."Female Labor Supply: A Survey." In Ha7id-book of Labor Economics, Vol. 1, edited byOrley Ashenfelter and David Card, 103-204.Amsterdam, The Netherlands: Elsevier,1999.

Merriman, David E, and Mark Skidmore."Did Distortionarv' Sales Taxation Contrib-ute to the Growth of the Service Sector?"National Tax Journal 53 Ko. 1 (March, 2000):124-42.

Mikesell, John."A Note on the Changing Incidence of StateLottery Finance." Social Science Quarterly 70No. 2 (June, 1989): 513-31.

National Center for Policy Analysis."Will Public Schools Add Universal Pre-school to K-12?" Available at http://www.ncpa.org /pi /edu /pdlll398e.html, 2000.

Pencavel, John."Labor Supply of Men: A Survey." In Hand-book of Labor Economics, VoL 1, edited byOrley Ashenfelter and David Card, 3-102.Amsterdam, The Netherlands: Elsevier, 1999.

Price, Donald L, and E. Shawn Novak."The Tax Inciderice of Three Texas LotteryGames: Regressivity, Race and Education."National Tax Journal 52 No. 4 (December,1999): 741-52.

Scott, Frank, and John Garen."Probabilit)' of Purchase, Amount of Pur-chase, and the Demographic Incidence ofthe Lottery Tax." Jounml of Public Ecommiics54 No. 1 (May, 1994): 121-13.

Selingo, Jeffrey."Seeking Dollars to Further Their Dreams,Scholarship Supporters Push for Lotteries."Chronicle of Higher Education (April 16,1999)-.A36-39.

Southem Regional Education Board.Data Library on Studetit Financial Aid. Avail-able at http://wwvv.sreb.org/main/EdData /DataLibrary /h ighered/financialaid /financialaid.asp. 2000.

Spindler, Charles."The Lottery and Education: Robbing Peterto Pay Paul?" Public Budgeting and Finance15 No. 3 (Fall, 1995): 54-62.

State of Georgia.Governor's Budget Report, FY 1997. Atlanta:Office of Planning and Budget, 1996.

State of Georgia. Goi^ernor's Budget Refiort, FY1996. Atlanta: Office of Planning and Bud-get, 1995.






Stranahan, Harriet, and Mary O. Borg."Horizontal Equit}' Implications of the Lot-tery Tax." National Tax Journal 51 No.l(March, 1998a): 71-82.

Stranahan, Harriet, and Marj' O. Borg."Separating the Decisions of Lottery Expen-ditures and Participation: A Truncated TobitApproach." Public Finance Review 26 No. 2(March, 1998b): 99-117.

U.S. Bureau of the Census.Statistical Abstract of tlie United States: 1998.Washington, D.C, 1998.

238

nho pays and who benefits? examining the distributional...

Documents