who are the ineligible eitc recipients?€¦ · spending on temporary assistance for needy families...

Who Are the Ineligible EITC Recipients?

1165

National Tax JournalVol. LIII, No. 4, Part 2

Abstract - The IRS estimates that more than 20 percent of EarnedIncome Tax Credit (EITC) payments are made in error. By match-ing the Current Population Survey (CPS) to tax return data, thispaper finds that a large portion of the overpayments went to fami-lies with children. Depending on the exact measure used, only 11to 13 percent of EITC recipients lacked children in their householdat the time they received the EITC. While some of these erroneouspayments to households with children are received by householdswith incomes above EITC eligibility levels, many of these ineligiblefamilies with children are likely to be quite similar to eligible EITCfamilies.

INTRODUCTION

This year, 19 million families are expected to receive theEarned Income Tax Credit (EITC) at a total cost to the

federal government of $30 billion (Committee on Ways andMeans, 1998; Office of Management and Budget, 2000). Thecost of the EITC is now more than total federal and statespending on Temporary Assistance for Needy Families(TANF). Research on the EITC suggests that the programsucceeds in transferring income to needy families whilemaintaining low administrative costs, encouraging the la-bor force participation of single parents, and having littleor no impact on hours conditional on working.1

Research has also shown, however, that many EITC recipi-ents are ineligible for the credit. Tabulations from the 1985and 1988 IRS Taxpayer Compliance Measurement Program(TCMP) surveys first presented by Holtzblatt (1991) andScholz (1990) found that one–third of EITC recipients werenot eligible for the credit, primarily because they did not haveeligible children. More recent estimates from tax year 1994returns suggest that 21 percent of EITC dollars are currentlybeing paid in error, including payments to ineligible taxpay-

Who Are the IneligibleEITC Recipients?

Jeffrey B. LiebmanJohn F. Kennedy Schoolof Government,Harvard University,Cambridge, MA 02138

and

National Bureau ofEconomic Research,Cambridge, MA 02138

1 On administrative costs see U.S. General Accounting Office (1995) andScholz (1997). On labor force participation see Dickert, Hauser, and Scholz(1995), Eissa and Liebman (1996), Meyer and Rosenbaum (1999), andEllwood (2000). On hours worked see Triest (1996), Holtzblatt, McCubbin,and Gillette (1994), Dickert, Hauser, and Scholz (1995), Eissa and Liebman(1996), Eissa and Hoynes (1998), and Meyer and Rosenbaum (1999).

NATIONAL TAX JOURNAL

1166

ers and overpayments to eligible taxpay-ers.2 In comparison, quality control datasuggest an overpayment rate of 6 percentfor Aid to Families with Dependent Chil-dren (AFDC).3

Given the excess burdens involved inincome transfers, a program that transfersone–fifth of its dollars to the wrong popu-lation could be prohibitively expensive.4

However, it is far from certain that theseoverpayments are reaching families thatare very different from the eligible fami-lies. IRS rules regarding filing status andthe claiming of children are complicated,and it is possible that many ineligibleEITC recipients are making innocent er-rors in claiming the EITC.

In order to determine what share ofEITC payments are erroneously going tofamilies without children and the charac-teristics of these ineligible families thispaper uses an exact match of the March1991 Current Population Survey (CPS) toinformation from the 1990 tax returns ofCPS adults. This match makes it possibleto estimate the number of taxpayers whoclaimed the EITC but did not have anychildren living with them. In addition toproviding an independent assessment ofthe extent and nature of the largest partof the EITC compliance problem, thisanalysis yields two other benefits that gobeyond what can be learned from IRS

compliance studies.5 First, it directly an-swers an important policy question: whatshare of EITC dollars are reaching house-holds with children? For many purposesthis is a more relevant number than thenoncompliance rate, and it is a numberthat is not available from the governmentstudies. Second, the analysis provides amuch richer description of the demo-graphic characteristics of both eligible andineligible EITC recipients than is availablefrom tax data alone.

I find that between 11 and 13 percent ofall tax year 1990 EITC recipients did nothave a child in their CPS household at thetime that they received the credit. By fur-ther matching back to the March 1990 CPS,I determine that 10 percent of EITC recipi-ents also did not have a child in theirhousehold one year before they receivedthe credit, and therefore were very un-likely to have been eligible for the credit.Noncompliance rates appear to be par-ticularly high among males filing ashousehold heads. One–third of malehousehold heads claiming the EITClacked children in their CPS households.

This paper is organized as follows. Thesecond section describes the evolution ofthe EITC compliance problem and high-lights the key uncertainties about the na-ture of EITC noncompliance. The thirdsection describes the two data sets em-

2 See Scholz (1997), U.S. General Accounting Office (1998), and McCubbin (2000). From the 1994 study, Treasuryhas released figures only on overpayment rates, and not on the number of ineligible taxpayers. However, inthe 1988 TCMP data, the percentage of ineligible taxpayers and the overpayment rates were quite similar,suggesting that in 1994 around 20 percent of EITC recipients were ineligible for the credit.

3 AFDC quality control audits may not be of the same intensity as IRS audits. Evidence of possibly higherAFDC error rates come from the work of Edin (1993), who found that essentially all AFDC recipients in aChicago housing development had unreported income (although not necessarily enough to make them ineli-gible for the program), and Hill, Hotz, Mullins, and Scholz (1997), who matched administrative welfare andearnings data for four California counties and found that at least 14 percent of AFDC recipients underreportearnings to the welfare system.

4 Liebman (1995) provides a simple framework for calculating when it is more efficient to make transfers throughthe tax system than through the welfare system in light of differing administrative costs, participation rates,and noncompliance rates. The IRS estimates a noncompliance rate of 16 percent for the overall tax system(Internal Revenue Service 1996), so the estimated EITC overpayment rates are not that far above the overallnoncompliance rate.

5 The analysis reported in this paper was completed well before the data from the 1994 compliance study wereavailable, at a time when it was unclear how much of an EITC compliance problem remained after the elimi-nation of the support test in 1990.


1167

ployed in the match and the imputationmethods used. The fourth section presentsresults on how many EITC taxpayers lackchildren in their CPS households and sub-families and describes the characteristicsof the childless EITC recipients. The fifthsection discusses the implications of theresults for the design of transfer programsfor low income families. The sixth sectionconcludes.

THE EVOLUTION OF THE EITCCOMPLIANCE PROBLEM

In the decade since the EITC complianceproblem was first recognized, there havebeen important changes in the EITC thathave altered the nature of the complianceproblem. The large expansions of thecredit, legislated in 1990 and 1993, haveincreased the return to erroneously claim-ing the credit; modifications of eligibilityrules have simplified enforcement; andthe IRS has taken important steps to re-duce EITC errors and fraud.

In tax year 1988, a taxpayer was entitledto claim the Earned Income Tax Credit ifhe or she met three requirements. First, thetaxpayer must have had earned income(wage and salary income plus businessand farm self–employment income) above$0 and below $18,576, as well as adjustedgross income below $18,576. Second, thetaxpayer must have had a child living withhim or her for more than half of the year.6

Third, the taxpayer was required to use a

filing status of married filing jointly, headof household, or surviving spouse.

Married taxpayers were generally re-quired to claim their children as depen-dents in order to be eligible for the EITC.7

Thus the taxpayer must have provided atleast half the cost of supporting the child.The head of household and survivingspouse filing statuses require that the tax-payer provide half the cost of keeping upa home for the child.8 Therefore, a supporttest applied for all EITC claimants. AFDCincome is counted as support provided bythe state, not the taxpayer. This means thata taxpayer who received $8,000 in AFDCbenefits and $6,000 in earnings wouldhave failed the support test and would nothave been eligible to receive the EITC.

Estimates from the 1988 TCMP implythat one–third of the 10.4 million taxpay-ers claiming the EITC in that year werenot eligible for the credit (see Table 1).9

The 3.4 million ineligible taxpayers re-ceived $1.9 billion of the $5.6 billion spenton the EITC. Forty percent of head ofhousehold filers claiming the EITC werenot entitled to the credit, compared with21 percent of married filers. These headof household returns accounted for almostthree–quarters of the taxpayers who hadtheir EITC claims disallowed. Table 2 dis-plays a tabulation of the filing statusclaimed by the taxpayer versus the filingstatus determined by the IRS auditor forall returns on which an EITC claim wasdisallowed. The table shows that most of

6 Under pre–1991 EITC rules, a child was the taxpayer’s son, daughter, stepchild, legally adopted child, ordescendent of the taxpayer’s child. In addition, married taxpayers could claim foster children who under IRSrules are children “you cared for as you would your own child.” However, the foster child (which for marriedtaxpayers includes grandchildren) had to live with the taxpayer for the full year.

7 Married taxpayers were entitled to claim the EITC without claiming a dependent exemption if the taxpayer’sspouse was not the child’s parent and the child’s other parent was allowed to claim the exemption under adivorce agreement.

8 Surviving spouses also had to be able to claim their children as dependents in order to claim the EITC, as wellas meeting a household maintenance test.

9 The Taxpayer Compliance Measurement Program samples roughly 50,000 individual tax returns every threeyears. IRS auditors conduct line by line audits of the sampled tax returns to assist the IRS in designingmethods for predicting which tax returns are efficient to audit. The 1991 TCMP was canceled due to budgetcutbacks. Tax returns were selected for the 1994 TCMP, but the audits never occurred first due to budgetcutbacks and then due to Congressional opposition to the “audits from hell.”


1168

the head of household filers in this popu-lation should have filed as single becausethey had all of their dependent childclaims disallowed. The filing status ofmarried taxpayers who had their EITCclaim disallowed was usually not changedby the TCMP auditor because losing de-pendent child exemptions does not affecta taxpayer’s eligibility for the married fil-ing status.10

The TCMP computer file does not ex-plain why EITC claims were disallowed.However, it does include information onincome, filing status, and dependent childexemptions both as claimed by the tax-payer and as revealed in the audit. In mostcases, the disallowal of all dependent ex-emptions would mean that the taxpayerwas not entitled to claim the EITC.11 Table3 uses the available evidence in the 1988

TABLE 1EITC AUDIT STATUS OF ALL TAX RETURNS IN 1988 TCMP

All tax returns

EITC not claimed

EITC established

EITC claimed

EITC not adjusted

EITC increased

EITC decreased

EITC disallowed

Audit Status

104,319,102(100.0 %)

93,822,851(89.9 %)

95,467(0.1 %)

10,400,784(10.0 %)

5,600,237(53.8 %)

428,040(4.1 %)

1,016,412(9.8 %)

3,356,095(32.2 %)

Number of Returns

5.627(100.0 %)

0.000(0.0 %)

0.000(0.0 %)

5.627(100.0 %)

3.056(54.3 %)

0.172(3.1 %)

0.544(9.7 %)

1.855(33.0 %)

Total Amount of EITC Claimed(billions of dollars)

Source: 1988 TCMP.Note: “EITC established” are tax returns not claiming the EITC that are eligible for the EITC.

TABLE 2RETURNS WITH EITC DISALLOWED IN 1988 TCMP

FILING STATUS CLAIMED VERSUS FILING STATUS EXAM

Filing Status(claimed)

Filing Status (exam)

SingleQualifying

Widow AllHead of

HouseholdMarriedSeparate

MarriedJoint

SingleMarried jointMarried separateHead of householdQualifying widow

All

26229,63818,263

1,959,8360

2,007,999

0797,187

000

797,187

028,418

0363,590

0

392,008

0796

0143,26214,506

158,564

0000

337

337

262856,03918,263

2,466,68814,843

3,356,095Source: 1988 TCMP.

10 The TCMP dropped from the sample non–married filers who in response to the audit opted to file as married.From the 1994 IRS study we know that such filers are rare.

11 A taxpayer could have legitimately claimed the EITC without claiming a dependent child if the taxpayer hadgiven up the dependent exemption in a divorce agreement.


1169

TCMP to explain why EITC claims weredisallowed. Eighty–two percent of taxpay-ers whose EITC claims were disallowedwere determined by the auditor either notto have any dependent children or not tobe eligible to use a filing status entitling thetaxpayer to the EITC (or both). Only 11.5percent had income exceeding the maxi-mum levels for EITC eligibility but wereotherwise eligible. Less than 1 percentlacked positive earned income. Two per-cent had both excessive income and eitherno dependent or an improper filing status.Four percent of the disallowed EITC claimsdo not fit into any of these categories.

The basic TCMP data suggest that mostnoncompliance on EITC tax returns in-volves the reporting of dependent chil-dren. Most of the disallowed EITC claim-ants had income levels that would havemade them eligible for the credit. How-ever, these basic data are open to a num-ber of interpretations.

First, it is possible that the TCMP doesnot accurately measure noncompliance. AGovernment Accounting Office (GAO)

study of erroneous dependent claims triedto examine the original audit sheets of 958tax returns that were coded in the 1988TCMP computer file as having the num-ber of dependents altered due to the au-dit. The GAO found that 180 of the auditsheets were missing, 51 had beenmiscoded on the computer file, and 85contained dependent exemptions thatwere disallowed by default because thetaxpayer never responded to the auditrequest (U.S. General Accounting Office,1993). In addition to doubts about dataquality, TCMP–based estimates of EITCnoncompliance are likely to be biased dueto the limits of what auditors can learn inan audit. On the one hand, TCMP auditsgenerally occur two to three years after atax return is filed and some low incometaxpayers who provided more than halfthe support for a child may not be able todocument the support a few years later.This would imply that the TCMP hasoverestimated the number ofnoncompliant EITC tax returns. On theother hand, auditors are unlikely to un-

TABLE 3TAX RETURNS WITH EITC DISALLOWED IN 1988 TCMP

REASONS WHY EITC CLAIM WAS DISALLOWED

No dependents and improper filing status, incomeeligible

No dependents only

Improper filing status only

AGI above $18,576 or earned income above $18,576,filing status and dependents eligible

Negative or zero earnings filing status anddependents eligible

(No dependents or improper filing status) and(AGI above $18,576 or earned income above $18,576)

Cannot explain why EITC was disallowed

Total returns with EITC disallowed

1,925,104(57.4 %)

376,832(11.2 %)

435,910(13.0 %)

386,471(11.5 %)

16,550(0.5 %)

66,113(2.0 %)

149,415(4.5 %)

3,356,395(100.0 %)

Number of Returns1.126

(60.7 %)

0.201(10.8 %)

0.264(14.2 %)

0.157(8.5 %)

0.011(0.6 %)

0.027(1.5 %)

0.069(3.7 %)

1.855(100.0 %)

Amount ofEITC Claimed

(billions of dollars)

Source: 1988 TCMP.


1170

cover all taxpayer fraud (Graetz andWilde, 1985), so the TCMP mayundercount the number of ineligible EITCrecipients.12, 13

Second, it is possible that most of theineligible 1988 EITC claimants failed thesupport test, but were otherwise eligiblefor the credit. If this were the case, then,ceteris paribus, current rates of EITC non-compliance would be much lower than1988 rates because under post–1990 rules,the support test no longer applies. Fromthe TCMP computer file it is usually im-possible to determine the reason that ataxpayer’s dependent child claim was dis-allowed. However, the original auditsheets do contain an explanation for whythe claim was disallowed. While I wasunable to obtain access to the audit sheets,the GAO study mentioned above exam-ined 554 TCMP audit sheets from tax re-turns that had one or more dependentchild claims disallowed. If the GAOsample were representative of all returnswith disallowed dependent exemptions,then roughly one–third of the GAOsample would be tax returns that claimedthe EITC. The GAO study estimated thatif the post–1990 EITC rules had been ineffect in 1988 then 890,000 of the 4 milliontax returns represented in their surveywould have failed the support test andbeen ineligible to claim a dependent child

while still being eligible for the EITC.14 Ex-trapolating to the entire population of 6.21million returns with disallowed depen-dents implies that 1.38 million EITC taxreturns would have been ineligible for thecredit under the 1988 rules, but eligibleunder the post–1990 rules. Thus, this smallGAO sample suggests that the supporttest accounted for 59 percent of the 2.3million EITC tax returns on which all de-pendent claims were disallowed and 41percent of all disallowed EITC returns.15

Third, it is possible that ineligible EITCtaxpayers did not have children livingwith them. One way in which this couldoccur is if noncustodial parents claimedtheir children. Since 1985, the custodialparent has been entitled to claim the de-pendent child exemption so long as bothparents combined paid more than half thechild’s support. Only if the custodial par-ent releases his or her right to the exemp-tion (or did so under a pre–1985 divorceor separation agreement) may the noncus-todial parent legally claim the exemption.Noncustodial parents are required toclaim the children as dependents livingaway from home. In the TCMP, less thanhalf of 1 percent of ineligible EITC returnscontained claims for children living awayfrom home. However, the IRS auditorswhom I interviewed suggested that non-custodial parents are savvy enough to re-

12 Two IRS auditors whom I interviewed told me that it is very difficult to catch a taxpayer who hides AFDCincome, since the auditor generally cannot obtain information from the state on benefits paid (Liebman,1994).

13 Another complaint about using TCMP estimates to measure EITC noncompliance comes from Greenstein(1995). He argues that not all taxpayers who erroneously claim the EITC receive the EITC, since the IRSmanages to catch some erroneous claims before mailing out refund checks. Greenstein’s point is more rel-evant for recent years when the IRS has begun using computerized matches of social security numbers beforesending out refunds than it is for the 1980s, when it was often many months after a refund check was mailedout that the IRS discovered that the return was suspect.

14 The GAO sample represents only 4 million of the 6.21 million tax returns with disallowed dependent exemp-tions. The sample does not represent the entire tax–filing population because some audit sheets could not befound and because discrepancies were found between some of the audit sheets and the TCMP computer file.

15 The GAO report estimated that 57 percent of returns that failed the support test failed for not providing halfthe support, while 43 percent failed for not keeping adequate records. According to the auditors whom Iinterviewed, a taxpayer’s dependent claim is often disallowed for not keeping adequate records when theauditor suspects but cannot prove that welfare income exists, but can show that spending is greater thanreported income. This implies that it is likely that many of the taxpayers whose claims were disqualified forinsufficient documentation in fact did not meet the support test.


1171

alize that claiming a child living awayfrom home without providing the re-quired paperwork would attract IRS at-tention, so that noncustodial parents mayclaim their children as living at home.Another way in which taxpayers withoutchildren might claim a dependent childis to invent a fictional one. The strongestevidence for this possibility is that in 1987,the first year in which taxpayers were re-quired to list social security numbers ofdependents on their tax returns, 7 millionfewer dependent children were claimedthan in the previous year (Szilagyi, 1991).Further evidence that nonexistent chil-dren may have been claimed comes fromthe 1988 TCMP. In 1988, taxpayers wererequired to list on their tax returns thesocial security numbers of all dependentswho were at least five years old. On taxreturns where the TCMP auditor disal-lowed an EITC claim, 39 percent of the dis-allowed dependent child claims were de-pendents for whom the taxpayer checkedthe box stating that the child was underfive and did not provide a social securitynumber—possibly because the childrendid not exist.

As was alluded to above, the 1990 leg-islation simplified EITC rules by remov-ing the support test and replacing it witha residency test and an adjusted gross in-come (AGI) tiebreaker. The EITC resi-dency test requires that the child live withthe taxpayer for more than six monthsduring the year (12 months for a fosterchild). The AGI tiebreaker rules state thatif a child could potentially be claimed bytwo taxpayers, only the person with thehigher adjusted gross income is eligibleto take the credit.

While eliminating the support test mayhave made some previously ineligible tax-payers eligible for the credit, it has also

created a new source of errors. The 1994IRS compliance study found an overalloverclaim rate of 21 percent, substantiallylower than the 35 percent rate for 1988.EITC claims by taxpayers without chil-dren continued to be a problem with 39percent of EITC overclaims due to taxpay-ers claiming children who did not meetthe residence test. However, an additional18 percent of overclaims were due to thefailure to correctly follow the new AGI–tiebreaker rules (GAO, 1998).

In addition, the 1994 study identifieda large source of errors that had not beennearly as important in the 1980s—mar-ried taxpayers who by law are supposedto file married returns but do not. Thirty–one percent of EITC overclaims were dueto these filing status errors, which occuramong separated couples who have notobtained a legal separation agreement oramong couples who are living togetherbut erroneously file separate returns withthe filing status of single or head ofhousehold (perhaps deliberately avoid-ing EITC marriage penalties).16 Whenthese separate returns are combined intoa joint return, the income rises, usuallyreducing the amount of the EITC thehousehold receives or eliminating eligi-bility altogether.

Over this period, the IRS adopted anumber of new enforcement procedures,including verifying the social securitynumbers of children claimed on tax re-turns before making payments. At firstglance, it might appear that these enforce-ment efforts have been highly successfulbecause the overpayment rate has fallenby roughly one–third. However, such ajudgement is complicated by the removalof the support test that may have simplyredefined many previously ineligible tax-payers as eligible taxpayers. Indeed, the

16 It seems likely that most of these errors involve married spouses who are separated from each other because theIRS study found that the appropriate post–audit filing status for most of these households was married filingseparately, not married filing jointly. While technically ineligible for the EITC, some of these separated house-holds may be very similar to the eligible single parent households that the EITC aims to transfers money to.


1172

GAO study discussed above indicatedthat the support test may have been re-sponsible for 41 percent of ineligible EITCclaims in 1988. This would imply that re-moving the support test directly reducedthe level of EITC noncompliance toaround 20 percent, raising the possibilitythat recent enforcement initiatives havehad no effect.

While it is impossible to be certain, itseems likely that recent IRS efforts haveindeed had a substantial impact. First, asdiscussed above, errors associated withthe AGI tiebreaker have likely replacedsome of the support test errors. Second,there is good reason to expect that in theabsence of new IRS efforts, the error ratewould have increased dramatically. Theresults of Liebman (1995), which used the1985 and 1988 TCMPs to study how EITCnoncompliance responded to the 1987 ex-pansion of the EITC, indicate that a 45 per-cent increase in the EITC increased thenoncompliance rate by 14 percent. It ishighly speculative to extrapolate fromthese results to the most recent EITC ex-pansions. Nonetheless, between 1990 and1994 the real value of the EITC more thandoubled, raising the payoff to erroneousclaims. Thus, we might have predicted anincrease in noncompliance of 33 percentfrom this increase. So even if we assumedthat elimination of the support test di-rectly reduced the noncompliance rate to21 percent, the expansion in the generos-ity of the credit would have raised thenoncompliance rate to 28 percent. If recentreforms have reduced the rate of noncom-pliance back to 21 percent, then they haveeliminated one–quarter of EITC noncom-pliance. If the impact of eliminating thesupport test was less than assumed here,the impact of IRS efforts may have beeneven greater.

DATA AND IMPUTATIONMETHODOLOGY

To produce an alternative measure ofEITC noncompliance, I became a specialsworn Census Bureau employee and useda data set which matched the March 1991CPS to tax return data.17 These data al-lowed me to estimate the percentage oftaxpayers who were likely to have claimedthe EITC on their 1990 tax returns (the re-turns most taxpayers filed in April, 1991)who told the Current Population Surveythat they did not have a child living withthem in March 1991. Since the 1990 EITCrules were identical to those for 1988 (ex-cept that the credit amounts were adjustedfor inflation), this provides a comparableestimate of the number of EITC claimantswho wrongly claimed to have a depen-dent child when they did not have a childliving with them. Since the support test isno longer a requirement for claiming theEITC and has been replaced by a residencytest, the number of EITC claimants whodo not have a child residing with them isalso a relevant number for thinking aboutthe current magnitude of child–relatedEITC noncompliance.

The March, 1991 CPS contains infor-mation on 158,477 individuals, of whom121,320 were adults (age 15 and over) andwere asked for their social security num-ber during the CPS interview (see Table4). The CPS obtained social security num-bers for 87 percent of the adults in thesample. Of the adults providing socialsecurity numbers, 77 percent matched taxreturns in the IRS’s Individual Master File,which contains selected information onevery tax return filed for tax year 1990.The taxpayers who have valid social se-curity numbers but do not match to taxreturns are considered nonfilers.

17 In order to protect census respondents, U.S. law does not permit the Census Bureau to share with the IRSmicro data in which individuals can be identified. Therefore, research matching census micro data with taxreturn data can only be done by Census Bureau personnel.


1173

The IRS Individual Master File (IMF)contains information from every tax re-turn filed during the year. Unfortunately,the 1990 IMF extract created for the matchdoes not record whether or not the tax-payer claimed the earned income taxcredit. However, it does contain the num-ber of dependent children claimed (sepa-rately for children claimed as living athome and for children living away fromhome), adjusted gross income, wage andsalary income, and a total income variablethat can be combined with schedule C andschedule F indicators to calculate self–employment income.18

I assume that a taxpayer claims the EITCif the taxpayer both has income qualify-ing him or her for the EITC and claims adependent child living at home. I havetested my methodology for identifyingEITC returns using the 1990 Statistics ofIncome (SOI) sample of tax returns. TheSOI sample contains all of the variables

that I use to predict whether a taxpayerclaims the EITC as well as the actualamount of EITC that the taxpayer claimed.I find that my methodology successfullyidentifies 95 percent of tax returns thatclaim the EITC and that only 2 percent ofthe taxpayers whom I predict to claim theEITC did not claim it.19 Thus, the measure-ment error introduced by not having a di-rect measure of EITC claims could at mostbias my estimates of improper claims up-ward by 2 percentage points (if 100 per-cent of the people claiming dependentchildren living at home but not claimingthe EITC did not have children at home).Most likely, the impact of this measure-ment error is negligible. For example, ifimproper child claims were twice as highamong the people whom I wrongly pre-dict to claim the EITC as among true claim-ers, my overall estimate of wrong claimswould be biased upwards by less thanthree–tenths of a percentage point.

TABLE 4DETAILS OF 1991 CPS —1990 IMF MATCH

Individuals in the March 1991 CPS

Children (age less than 15)Adults (age 15 and above)

Adults without valid SSNAdults with valid SSN

Filers (adults matching to a tax return)Nonfilers (adults not matching to a tax return)

158,477

37,157121,320

15,803105,517

81,62023,897

UnweightedNumber of

Observations248,807,213

55,347,888193,459,325

25,155,698168,303,627

130,118,29238,185,335

WeightedNumber of

Observations

Source: Author’s calculations from March 1991 CPS —1990 IMF match.

18 The total income variable is the sum of wage and salary, total interest (taxable and tax exempt), taxable divi-dends, alimony received, business income, farm income, pensions and annuities, net rents, royalties, estates,trusts, unemployment compensation, and social security benefits. The file contains separate items for theamount of wage and salary income, interest income, rent and royalty income, and social security income. Fortax returns that contained a schedule C, a schedule F, or a schedule SE, I define self–employment income asthe difference between total income and the sum of the separately listed components of income. To the extentthat these self–employed taxpayers have alimony, pension, or unemployment compensation income, my mea-sure of self–employment income is incorrect.

19 Three–fourths of the EITC claims that I miss are head of household filers who do not claim a dependent child.If I include all head of household filers with appropriate income in my sample of EITC filers, then my samplewould include 99 percent of all EITC claims. However, 6 percent of my sample would be taxpayers who didnot claim the EITC.


1174

In order to account for the 13 percent ofCPS sample members who did not pro-vide a valid social security number (SSN)to the CPS and therefore could not bematched to a tax return, I estimated theprobability of supplying a social securitynumber to the CPS conditional on theindividual’s observed CPS characteristicsand then reweighted the complete–datapart of the sample by the inverse of thisprobability. I used the method of momentsto produce standard errors that incorpo-rate the extra uncertainty that thereweighting procedure introduced intothe estimates.20

Even after accounting for nonmatches,the CPS–IMF sample has fewer EITC re-

cipients than would be expected based onIRS Statistics of Income data. The last tworows of Table 5 show the weighted num-ber of EITC returns from the CPS–IMFmatch and from SOI by filing status. Thetotal number of EITC returns in the CPS–IMF match is about 12 percent too low,and most of the discrepancy is a shortageof head of household returns.21 There area number of possible explanations for whythe number of EITC tax returns filed bythe entire March 1991 CPS sample is lessthan the number that were actually filedfor 1990. First, it is possible that the CPSundercounts low income families withchildren. Second, it is possible that somepeople file more than one tax return or file

TABLE 5ESTIMATES OF THE PERCENT OF EITC CLAIMANTS IN THE CPS–IMF MATCH

WITH CHILDREN PRESENT

Percent with no child in March,1991 CPS household

Percent with no child in March,1991 CPS subfamily

Percent with no child in the March,1991 CPS household and no childin March, 1990 household

Percent with no child in the March,1991 household or multiple taxreturns in household and moredependents claimed on tax returnsthan there are children in CPShousehold

Percent with no child and no adultchild in the 1991 CPS household

Weighted number of EITCclaimants in CPS–IMF match(millions)

Number of EITC claimants inStatistics of Income estimates(millions)

9.0 (0.4)

11.3 (0.4)

8.7 (0.8)

13.7 (0.5)

7.1 (0.7)

4.739

4.539

MarriedJoint

11.2 (0.6)

15.7 (0.7)

6.8 (0.9)

19.9 (0.8)

9.2 (0.5)

4.754

FemaleHead of

Households33.2 (1.8)

45.9 (1.9)

25.4 (3.3)

53.2 (2.0)

32.3 (1.8)

1.452

MaleHead of

Households13.2 (0.4)

17.8 (0.4)

10.1 (0.6)

21.4 (4.9)

11.3 (0.3)

10.978

12.540

Total

7.944

Source: Author’s calculations from the CPS–IRS match. Standard errors in parentheses.

20 An earlier version of this paper available in Liebman (1996) compares the estimates from the reweightingprocedure with those from two multiple imputation techniques, and presents the full details of all three meth-ods. The results were extremely similar across the three procedures.

21 About one–fifth of the undercounting is attributable to my method of identifying EITC returns. In the SOIsample, my method identifies 12.2 million EITC returns while there are in fact 12.5 million.


1175

without a proper social security number.Third, it could be that people who do notgive their SSN to the CPS are more likelyto file head of household tax returns thanthe prediction equations indicate. What-ever the cause, my inability to match theaggregate totals implies that my resultsare only representative of the 88 percentof EITC claims that are accounted for inthe entire CPS sample including imputa-tions.

THE NUMBER OF CHILDLESS EITCRECIPIENTS

Basic Results

Using the CPS–IMF match, I estimatethe proportion of tax returns predicted toclaim the EITC in which the taxpayer wasin a CPS household that did not containany children. In my methodology, a childmust be below 19 (below 24 if a full timestudent) but does not have to be the child(or even a relative) of the taxpayer in theCPS household who claimed the EITC.22

This is a lower bound estimate of theamount of noncompliance since there area number of reasons (most importantlythe AGI–tiebreaker) why a taxpayer witha child in the household may not be en-titled to claim that child for the purposesof the EITC. I also estimate the number ofEITC claimants who do not have a childin their CPS subfamily.23 This alternativeestimate gives an indication of families inwhich the choice of which adult claims thechild for EITC purposes may be ambigu-ous.24

The first two rows of Table 5 show thepercentage of (predicted) 1990 EITC taxreturns by gender and filing status wherethere was no child in the CPS householdor subfamily. Nine percent of couples fil-ing married joint tax returns claiming theEITC did not have a child in their CPShousehold. This estimate has a standarderror of 0.4. An additional 2.3 percent con-tained a child in the household but not inthe taxpayers’ subfamily. For female headof household returns, the numbers aresimilar to those for married joint returns.There were 11.2 percent who had no childin their household (with a standard errorof 0.6), while an additional 4.5 percent hadno child in their subfamily. For malehousehold heads, the numbers are muchlarger; 33.2 percent of male householdheads claiming the EITC did not have achild in their CPS household, and an ad-ditional 12.7 percent did not have a childin their subfamily. Among all EITC taxreturns, 13.2 percent did not have a childin the CPS household and 17.8 percent didnot have one in the subfamily.

Further Estimates

These results indicate that at the timeEITC taxpayers receive their tax refunds,a sizable portion do not have children liv-ing in their households. While this factmay itself be of policy relevance since theEITC aims to raise the living standards offamilies with children, it is not necessar-ily a good estimate of the number of EITCtaxpayers who are noncompliant due to

22 Currently a child must be under 19, under 24 and a full–time student, or permanently or totally disabled to bean EITC qualifying child. However, under 1990 rules some older children could have qualified their parentsfor the EITC. In the next subsection, I explore the possibility that taxpayers are claiming adult children inorder to claim the EITC.

23 I include the primary family (excluding any subfamilies that belong to it) as a subfamily. Thus, in this portionof the analysis a child will qualify only one of the household’s subfamilies for the credit.

24 There are two main cases in which children in different subfamilies might entitle a taxpayer to the EITC. First,in a multigenerational family a grandparent who was in a different Census subfamily could be eligible toclaim the child for the EITC. Second, under the foster child rules (which in this period only apply to marriedtaxpayers), a taxpayer can claim a child for the EITC if the child lives with the taxpayer for the entire year andif the taxpayer treats the child as if it were the taxpayer ’s own child.


1176

claiming children who do not live withthem. On the one hand the estimate is toolow since the requirements for being aqualifying child are stronger than simplyresiding in the same household or eventhe same subfamily. On the other hand,the estimates could overstate noncompli-ance to the extent that there are changesin household composition between the taxyear and the time of the CPS survey. Inaddition, it is possible that adult childrenmay have qualified some taxpayers for thecredit whom I categorize as ineligible. Toinvestigate these considerations, I performthree additional calculations. First, whenpossible, I match EITC taxpayers who lacka child in the March, 1991 CPS back to theMarch, 1990 CPS. It is unlikely that a tax-payer would lack a child in both of theCPS surveys and still have had a childwho met the six month residency require-ment for claiming the EITC.25 Second, Icalculate the number of EITC taxpayerswho lived in households where more thanone tax return was filed and where thetotal number of dependent childrenclaimed on all the tax returns exceeded thenumber of children present in the CPShousehold. Third, I calculate a lowerbound on the percentage of ineligibleEITC recipients by counting some taxpay-ers with adult children in their householdsas eligible for the credit.

There were 1,010 observations in theMarch, 1991 CPS who were predicted toclaim the EITC, but did not have a childin their household. In theory, roughly halfof them (those with months–in–samplegreater than four) can be matched to ob-

servations for the same individual in theMarch, 1990 CPS. In fact, I was able tomatch only 60 percent of the potentiallymatchable individuals to individuals inthe March, 1990 CPS.26

Since sample attrition in the CPS is notrandom (Welch, 1991), I estimated a logitregression to obtain the conditional prob-ability that someone with given charac-teristics who was a March, 1991 CPS adultand in his or her last four months in thesample would also be present in theMarch, 1990 CPS. I then reweighted the1991 observations that matched back to1990 by the inverse of the probability thatthey were in both samples.

In the third row of Table 5, I present es-timates from this procedure. Of the taxreturns claiming the EITC for 1990, 10.1percent were filed by taxpayers wholacked children in both their March, 1990and March, 1991 households. Thus my es-timate of noncompliance is 25 percentlower when I use this more restrictivemeasure. Female household heads showa much lower noncompliance rate withthis measure—6.8 percent rather than11.3.

One reason that my basic estimates ofnoncompliance might be too low is thatthe same CPS child could be claimed bymore than one taxpayer. Therefore, I esti-mate the percentage of households withEITC returns in which more than one re-turn claimed dependent children living athome and the number of children in thehousehold was less than the total num-ber claimed on all the tax returns filed bythe household. For this estimate, I used

25 While family dynamics are often complicated, in order for my methodology to miss a child who qualifies thetaxpayer for the EITC, the child would have to make two moves between March, 1990 and March, 1991—afirst move into the household after March, 1990 and a second move out of the household more than six monthslater but before March, 1991. Even if this possible measurement error makes my estimates less than idealmeasures of EITC noncompliance, the estimates are still relevant for the question of whether EITC paymentsare reaching households that have children living in them at the time the payments are made.

26 Within households, I first matched individuals in the March, 1991 household to individuals who were oneyear younger, of the same gender, and on the same CPS line number in the March 1990 household. If no matchwas found, I then dropped the line number requirement. If there was still no match, I matched the March,1991 individual to an (unmatched) individual in the March, 1990 household of the same gender who waseither the same age as the March, 1991 individual or two years younger.


1177

only households in which all adults sup-plied a social security number to the CPS.Of the 59,929 households in the March1991 CPS, 49,878 contained social secu-rity numbers for all adults. The percent-age of EITC tax returns in this subsamplethat were in households that lacked chil-dren was 12.8 percent compared to 13.2percent for the entire sample, so thesehouseholds seem to be reasonably repre-sentative of all households. Of the EITCtax returns in this subsample, 32 percentwere in households in which more thanone tax return was filed. In roughly 40percent of the EITC households in whichmore than one tax return was filed, morechildren were claimed on the tax returnsthan were present in the household. Thus,the total percentage of EITC tax returnsthat were in households with no childrenor that were in households in which mul-tiple tax returns were filed that claimeddependent children and in which the to-tal number of children claimed on tax re-turns exceeded the number of childrenpresent in the household in March, 1991was 21.4 percent (row 4 of Table 5). Therate for married tax returns was 13.7 per-cent, the rate for returns filed by femalehousehold heads was 19.9 percent, andthe rate for male household heads was53.2 percent. While it is of course possiblethat the tax return in the household claim-ing the EITC was in fact eligible to claimthe children while the other tax returnswere not, these estimates suggest at thevery least that there is sizableoverreporting of dependents on tax re-turns filed by EITC households, if notnecessarily on EITC returns.27

Under pre–1991 EITC rules, there wasno age restriction on children claimed forthe EITC. However, married taxpayersgenerally had to claim their child as a de-

pendent to receive the child and the rulesfor dependent children do contain age re-strictions. Moreover, since all of the chil-dren in my EITC sample were alsoclaimed as dependents, the age require-ments for dependent children apply evenfor the head of household returns. Underthe requirements for claiming dependentchildren, the child must be under 19 (orunder 24 if a full–time student) or thechild’s gross–income must be under a cer-tain limit ($2,100 in 1990). To eliminatehouseholds containing an adult child withincome below $2,100 from my estimatesof EITC noncompliance, I identified tax-payers with an adult child in the house-hold (using the CPS parent pointer vari-able). If the adult child either did not filea tax return or filed one with AGI below$2,100, then I assumed the taxpayer couldclaim the child as a dependent and qualifyfor the EITC. For adult children who didnot give social security numbers to theCPS, I used their total CPS income as aproxy for gross income. Since I do notimpose the support test or the joint returntest, this measure overestimates the num-ber of EITC eligible households. Account-ing for adult children lowers the estimatesof EITC noncompliance for married tax-payers and female household heads bynearly 20 percent. However, it lowersthe estimates of EITC noncompliance formale household heads by only 3 percent(row 5 of Table 5).

Demographic Characteristics of Eligibleand Ineligible EITC Recipients

Past descriptions of the characteristicsof EITC recipients have either had to makedue with the limited information availablein samples of tax returns or to imputeEITC recipiency based on survey informa-

27 There is another source of child–related EITC noncompliance that is not picked up in these estimates.Even when each child is only claimed once, if there are multiple tax returns filed by adults in the house-hold, the wrong taxpayer (the one with the lower AGI) could erroneously claim the child in order toreceive the EITC.


1178

tion.28 The estimates presented here arelikely to give a much more accurate pic-ture of who received the EITC in tax year1990 and allow us to examine how thecharacteristics of eligible and ineligibleEITC recipients differ. Tables 6 and 7present demographic and economic char-acteristics of taxpayers who were pre-dicted (based upon information from theirtax returns) to claim the EITC in 1990.Table 6 contains information on taxpay-ers who had a child in their March, 1991CPS household, while Table 7 presentsinformation on taxpayers who did nothave a child in their household. The tablesalso present separate characteristics by fil-ing status and gender (the demographiccharacteristics of the married taxpayersare the average of the characteristics of thetwo spouses).

Table 6 shows that 50 percent of eligible1990 EITC taxpayers are married, while 30percent are formerly married, and 20 per-cent have never been married. A little morethan half are white, a quarter are black, and18 percent are Hispanic. Of eligible EITCrecipients, 74 percent have a high schooleducation or less; 44 percent live in theSouth; and 36 percent live in a central city.Fifty–eight percent work 1,500 hours ormore, though this average is broughtdown by married couples in which onespouse does not work. Sixteen percent ofeligible EITC tax returns are filed by indi-viduals in households that receive welfareincome during the year and 26 percent arein households receiving food stamps.

Compared with eligible taxpayers, in-eligible EITC recipients are less likely tobe currently married and are more likelyto have never been married. Ineligible tax-payers are also more likely to be black, to

have less than high school education, tolive in the South, and to be at least 30 yearsof age. Ineligible taxpayers are much lesslikely to live in a household that receivedpublic assistance (3 percent), food stamps(8 percent), or Medicaid (11 percent) dur-ing the previous year. This is further evi-dence that no child was living with thetaxpayer during the previous year andthat the discrepancy between the numberof children claimed on the tax return andthe number present in the CPS householdis not caused by CPS measurement error.

The marital status tabulations in Table 7indicate that a substantial percentage oftaxpayers who filed head of household re-turns reported to the CPS that they were infact married. This is consistent with the re-sults of the 1994 IRS compliance study thatfound that many married taxpayers erro-neously used a non–married filing status.

IS THE EITC AN EFFECTIVE WAY TOTRANSFER INCOME?

The estimates in this paper indicate thatin 1990, between 11 percent and 13 per-cent of EITC recipients did not have achild living with them at the time theyreceived the credit. Thus it seems likelythat almost 90 percent of EITC dollars arereaching families with children.29

However, some of these families appearto have incomes above the intendedrange. Relatively few EITC taxpayers un-der report income, but because house-holds often contain more than one tax–fil-ing unit, the income measured on a taxreturn may not correspond well to the to-tal resources of the household. More thanone–third of EITC recipients in the 1990data live in households with multiple

28 Scholz (1994) matched the 1990 SIPP to a file that allowed him to determine which Survey of Income andProgram Participation (SIPP) individuals filed tax returns. He presents the characteristics of people who arepredicted based on SIPP information to be eligible for the EITC and who file a tax return. He also presentscharacteristics of people who told the SIPP that they received the EITC but who do not appear to be eligiblefor the credit based upon other SIPP information.

29 If the noncustodial parents who are erroneously receiving the EITC make income transfers to their children,the percentage of EITC dollars reaching families with children could be even higher.


1179

TABLE 6DEMOGRAPHIC AND ECONOMIC CHARACTERISTICS OF TAXPAYERS PREDICTED TO CLAIM

THE 1990 EITC WHO HAVE CHILDREN IN THEIR MARCH 1991 CPS HOUSEHOLD

Marital statusMarriedWidowedDivorcedSeparatedNever married

RaceNon–hispanic whiteNon–hispanic blackHispanicNon–hispanic other

EducationLess than high schoolHigh schoolMore than high schoolCollege degreeMore than college

Geographical regionNortheastMidwestSouthWest

Central cityIn central cityBalance of MSANot in MSA

FarmFarmNon–Farm

Age14 < age < 2019 < age < 2524 < age < 3029 < age < 4039 < age < 5049 < age < 65age > 64

Hours worked last year00 < hours < 500500 < hours < 10001000 < hours < 15001500 < hours < 20002000 < hours < 25002500 < hours < 3000hours > = 3000

Migration statusSame house last yearDiff. house last year

92.1 0.7 0.93.52.8

66.19.1

19.8 4.9

33.242.416.3 5.1 3.0

13.119.744.023.2

27.137.135.8

3.396.7

1.9 13.020.134.619.79.61.0

22.0 9.2 8.59.8

11.728.7 5.4 4.8

73.826.2

Married

35.6 2.420.99.2

31.8

34.635.726.9 2.8

39.241.014.7 3.0 2.1

14.714.646.324.5

43.731.524.8

0.899.2

4.115.524.932.016.36.9

0.4

5.7 5.6 5.013.715.745.1 5.1 4.0

65.634.4

MaleHousehold

Head

15.55.1

35.7 13.530.3

46.237.6

13.8 2.4

24.546.923.2 3.4 1.9

16.523.343.6

16.6

44.133.522.4

0.599.5

2.7 13.219.740.217.66.3

0.3

6.7 8.1 9.214.018.139.8 2.6 1.5

71.528.5

FemaleHousehold

Head

52.33.0

18.28.5

17.9

54.024.517.9 3.6

30.044.219.2 4.1 2.4

14.820.744.020.4

36.235.028.8

1.898.2

2.5 13.420.536.818.47.9

0.6

13.5 8.38.5

12.114.935.3 4.1 3.2

71.928.1

All

Tax Return Filing Status


1180

adults filing separate tax returns, andmany live in households with total in-comes above the EITC maximum. More-over, data from the 1994 IRS compliancestudy indicate that errors associated withthe AGI tiebreaker and with married tax-payers wrongly splitting their incomesamong two non–married returns are re-sponsible for almost half of EITCoverclaims, accounting for about 10 per-cent of EITC dollars.30

The difficulties that arise in verifyingthe residency of children claimed on taxreturns and the imprecise measure ofhousehold resources provided by a taxreturn are two drawbacks to using the taxsystem to transfer income in low–incomefamilies. In contrast, traditional welfareprograms employ caseworkers to verifyeligibility and often use more comprehen-sive measures of household resources indetermining eligibility. However, the

TABLE 6 (continued)DEMOGRAPHIC AND ECONOMIC CHARACTERISTICS OF TAXPAYERS PREDICTED TO CLAIM

THE 1990 EITC WHO HAVE CHILDREN IN THEIR MARCH 1991 CPS HOUSEHOLD

Married

MaleHousehold

Head

FemaleHousehold

Head All


Labor force statusWorkingWith job not workingUnemployed lookingUnemployed on layoffNot in labor force

Income on tax returnPhase–inFlatPhase–out

HH public assistanceYesNo

HH public assistancemean for PA>0

HH food stampsYesNo

HH food stampsmean for FS>0

HH medicaidYesNo

Weighted number of tax returns

Unweighted observations

61.3 2.47.2 1.7

27.4

25.815.258.9

11.288.8

$3185

22.078.0

$1593

20.679.4

4,240,106

4978

74.5 2.4 9.2 4.29.7

35.716.048.3

20.080.0

$2868

25.774.3

$1686

29.470.6

1,008,069

603

75.4 3.6 6.6 1.1

14.2

38.320.740.9

19.880.2

$2707

27.572.5

$2514

29.370.7

4,087,243

2552

68.5 2.9 7.1 1.7

19.7

32.317.749.9

15.984.1

$2878

24.875.2

$1631

25.474.6

9,355,391

8146

Source: Author’s calculations from match of March, 1991 CPS to 1990 IMF.Note: The weights for the married tax returns are split between the two spouses.

30 It is likely that a substantial share of the married taxpayers who filed non–married returns involve cases inwhich the spouses are separated and therefore that the incomes on the two returns are not accruing to a singlehousehold.


1181

TABLE 7DEMOGRAPHIC AND ECONOMIC CHARACTERISTICS OF TAXPAYERS PREDICTED TO CLAIM

THE 1990 EITC WHO HAVE NO CHILDREN IN THEIR MARCH 1991 CPS HOUSEHOLD

Marital statusMarriedWidowedDivorcedSeparatedNever married

RaceNon–hispanic whiteNon–hispanic blackHispanicNon–hispanic other

EducationLess than high schoolHigh schoolMore than high schoolCollege degreeMore than college

Geographical regionNortheastMidwestSouthWest

Central cityIn central cityBalance of MSANot in MSA

FarmFarmNon–Farm

Age14 < age < 2019 < age < 2524 < age < 3029 < age < 4039 < age < 5049 < age < 65age > 64

Hours worked last year00 < hours < 500500 < hours < 10001000 < hours < 15001500 < hours < 20002000 < hours < 25002500 < hours < 3000hours > = 3000

Migration statusSame house last yearDiff. house last year

Married

MaleHousehold

Head

FemaleHousehold

Head All


71.3 0.1 4.411.711.7

55.713.727.7 2.8

46.635.711.2 4.3 2.2

12.916.041.729.4

26.844.029.2

4.295.8

0.5 3.612.224.120.929.88.8

26.9 6.7 7.89.2

14.024.1 6.3 5.1

75.924.1

14.1 2.021.714.747.5

23.852.022.2 2.1

41.536.414.8 5.1 2.1

11.521.749.417.4

46.129.824.0

2.597.5

2.315.620.538.411.010.6 1.6

5.6 3.4 7.712.618.543.8 5.6 2.8

69.031.0

19.214.235.3 7.923.4

42.345.5

10.6 1.6

36.138.420.1 3.2 2.1

18.315.354.8 11.6

44.132.923.0

0.499.6

0.2 7.911.521.628.127.4 3.3

5.1 4.2 8.812.121.043.6 2.5 2.6

75.324.7

32.7 6.721.411.228.0

40.338.119.5 2.1

40.937.015.8 4.2 2.2

14.517.649.218.7

39.435.225.4

2.297.8

1.0 9.214.727.920.322.5 4.4

11.9 4.7 8.111.518.037.8 4.7 3.4

73.426.6


1182

greater accuracy of the welfare systemcomes at a cost. Administrative costs ofAFDC were 16 percent of benefits paid in1996 (Committee on Ways and Means,1998),31 while administrative costs for theEITC, although hard to isolate from therest of the IRS budget, appear to be lessthan 3 percent of total credits received bytaxpayers, probably substantially less.32

Moreover, the welfare system is gener-ally thought to impose a greater burdenon beneficiaries than the tax system, bothbecause the time cost of regular meetingswith a case worker is larger than the mar-ginal cost of adding an additional form toone’s tax return33 and because of thestigma associated with welfare receipt.These cost differentials are likely to be part

TABLE 7 (continued)DEMOGRAPHIC AND ECONOMIC CHARACTERISTICS OF TAXPAYERS PREDICTED TO CLAIM

THE 1990 EITC WHO HAVE NO CHILDREN IN THEIR MARCH 1991 CPS HOUSEHOLD

Married

MaleHousehold

Head

FemaleHousehold

Head All


Labor force statusWorkingWith job not workingUnemployed lookingUnemployed on layoffNot in labor force

Income on tax returnPhase–inFlatPhase–out

HH public assistanceYesNo

HH public assistancemean for PA>0

HH food stampsYesNo

HH food stampsmean for FS>0

HH medicaidYesNo

Weighted number of tax returns

Unweighted observations

Source: Author’s calculations from match of March, 1991 CPS to 1990 IMF.Note: The weights for the married tax returns are split between the two taxpayers.

59.2 3.39.3 1.0

27.2

33.820.545.7

2.397.7

$2152

8.191.9

$1134

13.286.8

419,902

402.5

69.7 3.8 9.2 6.211.1

36.318.045.7

3.896.2

$1588

8.691.4

$718

11.388.7

482,713

273

84.3 1.9 4.4 0.9 8.4

31.119.749.1

3.196.9

$2083

6.493.6

$862

8.191.9

518,785

328

71.9 2.9 7.4 2.7

15.0

33.619.347.2

3.196.9

$1881

7.692.4

$888

10.789.3

1,430,718

1010

31 To the extent that AFDC administrative dollars are spent on services beyond eligibility verification and ben-efit payments, the administrative costs of the two programs are not directly comparable.

32 This number is based on Scholz (1997). See Liebman (1998) for a discussion.33 According to Scholz (1997), most EITC recipients would file a tax return even in the absence of the EITC.


1183

of the reason that take–up rates appear tobe higher for the EITC than for traditionalwelfare programs.34 Moreover, evenwithin the EITC itself it is clear that re-cent efforts at improving compliance haveincreased administrative costs and havelikely discouraged some eligible recipientsfrom applying for the program.

Is it more cost–effective to administerpayments to low–income working fami-lies through the tax system with low ad-ministrative costs and high error rates orthrough the welfare system with high ad-ministrative costs and lower error rates?35

The answer depends heavily on how dol-lars transferred to ineligible taxpayers arevalued.

With an overpayment rate of 21 percentand administrative costs of 3 percent,every dollar that is transferred to an eli-gible EITC recipient leads to 27 cents ofpayments to ineligible taxpayers (.21/.79)and administrative costs of 4 cents (.03 x1.27).36 In contrast, if it were possible toadminister the EITC through the welfaresystem with an overpayment rate of 6percent and administrative costs of 16percent, then every dollar that was trans-ferred to eligible recipients would lead to6 cents in payments to ineligible taxpay-ers (.06/.94) and administrative costs of17 cents (.16 x 1.06).37

If both the administrative costs and thetransfers to ineligible taxpayers are treatedas having zero value, then administeringthe program through the welfare systemis cheaper; transfering one dollar to aneligible taxpayer costs $1.23 cents throughthe welfare system and $1.31 through thetax system.38 However, if many EITC er-rors are inadvertent and involve pay-ments to low–income families with chil-dren we may want to place some valueon the transfers to ineligible taxpayers,perhaps assigning them the same socialwelfare weight as a dollar given to theaverage taxpayer. In that case the dollarsthat go to the ineligible taxpayer are not acost (though the transfer does still requireadministrative costs) and transferring onedollar to an eligible taxpayer costs $1.17through the welfare system and $1.04through the tax system.39

CONCLUSION

The IRS estimates that more than 20percent of EITC payments are made inerror. However, by matching the CPS totax return data, this paper finds that alarge portion of the overpayments wentto families with children. Depending onthe exact measure used, only 11 to 13 per-cent of EITC recipients lacked children in

34 Scholz (1994) shows, using a variety of data sets and methodologies, that between 80 and 86 percent of EITC–eligible taxpayers receive the EITC. In comparison, Blank and Ruggles (1993) estimate that AFDC–eligiblefamilies received 75 percent of the total dollars to which these families are entitled.

35 Holtzblatt and Liebman (1999) discuss the UK’s recent conversion from a welfare–based system of in–workbenefits to a tax–based system.

36 There is also excess burden from raising the revenue, but in this simple example it is simply proportional tothe direct costs and does not affect the comparison.

37 As I discussed in footnote 3, there are reasons to suspect that AFDC noncompliance rates are higher than theofficial quality control estimates.

38 Placing a low or even negative weight on the dollars transferred to ineligible recipients could be justified iftheir activity may encourage others to evade taxes or is simply distasteful. Liebman (1995) finds that im-proper claiming of children rose after the 1986 expansion of the EITC, indicating that at least some ineligibleEITC recipients are responding to the economic incentive to wrongly claim children.

39 This analysis ignores the different participation rates of the two programs. If the social welfare value ofdollars to eligible taxpayers is high, the tax system could be preferable even at a higher cost because it man-ages to reach a higher percentage of its intended beneficiaries. Liebman (1995) contains a more elaborateanalysis of this issue that includes differential participation rates.


1184

their household at the time they receivedthe EITC. While some of these erroneouspayments to households with children aregoing to households with multiple adultsand combined incomes that exceed thatof the typical EITC family, many of theseineligible families with children are likelyto be quite similar to eligible EITC fami-lies.

Acknowledgments

I am grateful to John Coder for his as-sistance with the CPS–IMF match and toJoel Stubbs for providing me with esti-mates from the TCMP. I thank David Cut-ler, Nada Eissa, Dan Feenberg, MartinFeldstein, Larry Katz, Guido Imbens,Brigitte Madrian, Billy Pizer, and KarlScholz for numerous helpful discussions,and Janet Holtzblatt, George Yin, the edi-tors, and referees for helpful comments onan earlier draft. I am also grateful to JoyceCooper, Chuck Nelson, Bill Prosser, andDaniel Weinberg for facilitating my re-search at the Census Bureau. This researchwas supported by a National ScienceFoundation Graduate Research Fellow-ship and by an Alfred P. Sloan DoctoralDissertation Fellowship.

REFERENCES

Blank, Rebecca M., and Patricia Ruggles.“When do Women Use AFDC & FoodStamps? The Dynamics of Eligibility vs.Participation.” NBER Working Paper No.4429. Cambridge, MA: National Bureau ofEconomic Research, 1993.

Committee on Ways and Means.1998 Green Book. Washington, D.C.: Govern-ment Printing Office, 1998.

Dickert, Stacy, Scott Hauser, and John K.Scholz.

“The Earned Income Tax Credit and Trans-fer Programs: A Study of Labor Market andProgram Participation.” Tax Policy and theEconomy 9 (1995): 1–50.

Edin, Kathryn.There’s a Lot of Month Left at the End of theMoney: How Welfare Recipients Make EndsMeet in Chicago. New York: Garland Publish-ing, 1993.

Eissa, Nada, and Hilary Hoynes.“The Earned Income Tax Credit and the La-bor Supply of Married Couples.” NBERWorking Paper No. 6856. Cambridge, MA:National Bureau of Economic Research, 1998.

Eissa, Nada, and Jeffrey B. Liebman.“Labor Supply Response to the Earned In-come Tax Credit.” Quarterly Journal of Eco-nomics 111 No. 2 (May, 1996): 606–37.

Ellwood, David T.“The Impact of the EITC on Work and Mar-riage.” National Tax Journal 53 No. 4, Part 2(December, 2000): 1063–1106.

Graetz, Michael J., and Louis L. Wilde.“The Economics of Tax Compliance: Factand Fantasy.” National Tax Journal 38 No. 3(September, 1985): 355–63.

Greenstein, Robert.“The Earned Income Tax Credit: A Target forBudget Cuts.” Washington, D.C.: Center onBudget and Policy Priorities, 1995.

Hill, Carolyn J., V. Joseph Hotz, Charles H.Mullin, and John Karl Scholz.

“EITC Eligibility, Participation and Compli-ance Rates for AFDC Households: Evidencefrom the California Caseload.” Draft Report.University of Chicago, 1997.

Holtzblatt, Janet.“Administering Refundable Tax Credits:Lessons from the EITC Experience.” In Pro-ceedings of the Eighty–Fourth Annual Confer-ence on Taxation, 180–5. Washington, D.C.:National Tax Association, 1991.

Holtzblatt, Janet, and Jeffrey B. Liebman.“The EITC Abroad.” In Proceedings of theNinety-First Annual Conference on Taxation,198–207. Washington, D.C.: National Tax As-sociation, 1999.

Holtzblatt, Janet, Janet McCubbin, andRobert Gillette.

“Promoting Work Through the EITC.”Washington, D.C.: U.S. Department of Trea-sury, June, 1994.


1185

Internal Revenue Service.Federal Compliance Research: Individual IncomeTax Gap Estimates for 1985, 1988, 1992. Wash-ington, D.C.: U.S. Department of the Trea-sury, 1996.

Liebman, Jeffrey B.Interview with Kevin Belanger and SeanSchomer. 9 a.m., September, 6, 1994. 18Tremont Street, Boston.

Liebman, Jeffrey B.“Noncompliance and the Earned IncomeTax Credit: Taxpayer Error or TaxpayerFraud?” Harvard University. 1995.

Liebman, Jeffrey B.The Impact of the Earned Income Tax Credit onLabor Supply and Taxpayer Compliance.Harvard University. Ph.D. Thesis, 1996.

Liebman, Jeffrey. B.“The Impact of the Earned Income Tax Crediton Incentives and Income Distribution.” TaxPolicy and the Economy 12 (1998): 83–118.

McCubbin, Janet.“Noncompliance and the Optimal Design ofthe Earned Income Tax Credit,” National TaxJournal 53 No. 4, Part 2 (December, 2000):1135–1164.

Meyer, Bruce D., and Dan T. Rosenbaum.“Welfare, the Earned Income Tax Credit, andthe Labor Supply of Single Mothers.” NBERWorking Paper No. 7363.Cambridge, MA:National Bureau of Economic Research, 1999.

Office of Management and Budget.Budget of the United States Government, FiscalYear 2001. Washington, DC: U.S. Govern-ment Printing Office, 2000.

Scholz, John K.“The Participation Rate of the Earned In-come Tax Credit.” Institute for Research onPoverty Discussion Paper 928–90. Madison,WI: University of Wisconsin–Madison, 1990.

Scholz, John K.“The Earned Income Tax Credit: Participa-tion, Compliance, and Antipoverty Effec-tiveness.” National Tax Journal 48 No. 1(March, 1994): 63–85.

Scholz, John K.Testimony before U.S. House of Represen-tatives, Committee on Ways and Means.Washington, D.C., 1997.

Szilagyi, John.“Where Some of Those Dependents Went.”In 1990 Research Conference Report: How DoWe Affect Taxpayer Behavior, 162–3. Washing-ton, D.C.: Internal Revenue Service, March,1991.

Triest, Robert K.“The Efficiency Cost of IncreasedProgressivity.” In Tax Progressivity and In-come Inequality, edited by Joel Slemrod, 137–69. Cambridge, UK: Cambridge UniversityPress, 1996.

U.S. General Accounting Office.Tax Administration, Erroneous Dependent andFiling Status Claims. GAO/GGD-93-60,Washington, D.C., March, 1993.

U.S. General Accounting Office.Earned Income Credit, Noncompliance and Po-tential Eligibility Revisions, Statement of LyndaD. Willis. GAO/T-GGD-95-179, Washington,D.C., June, 1995.

U.S. General Accounting Office.Earned Income Tax Credit: IRS’ Tax Year 1994Compliance Study and Recent Efforts to ReduceCompliance. GAO/GGD-98-150, Washing-ton, D.C., June, 1998.

Welch, Finis.“Matching the Current Population Sur-veys.” Stata Technical Bulletin 12 (1991):7–11.

who are the ineligible eitc recipients?€¦ · spending on temporary assistance for needy families...

Documents