TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES DAVID JOULFAIAN * & MARK RIDER *

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES DAVID JOULFAIAN * & MARK RIDER * Abstract - This paper examines the impact of marginal tax rates, which incorporate the earned income tax credit (EITC) as it existed in 1988, on the reporting of income by low-income taxpayers. We generally find that misreported income is not affected by tax rates, except for proprietors. Negative marginal tax rates, which occur in the phase-in range of the EITC, do not appear to affect the amount of income overreported, irrespective of the source of income. Furthermore, the amount of income underreported does not appear to be affected by the relatively high marginal tax rates which occur in the phase-out range, except for proprietors. Even in the case of proprietors, the effect on the understatement of income is modest. INTRODUCTION Since the seminal work of Allingham and Sandmo (1972), a number of studies have examined the influence of tax rates, audit rates, and penalties, * Office of Tax Analysis, U.S. Department of the Treasury, Washington, D.C. 222. among other policy instruments, on tax compliance. The availability of the earned income tax credit (EITC) allows us to extend such studies by examining an interesting and generally overlooked question: the effect of negative tax rates on reported income. 1 Furthermore, the EITC enables us to examine the response of low-income taxpayers to relatively high marginal tax rates. Because the EITC initially increases with income, marginal tax rates are negative for some recipients. Some low-income taxpayers, however, face relatively high positive tax rates as the credit is phased out. In 1988, the year of our focus, marginal tax rates ranged from.14 to.37 for credit recipients. 2 In order to study the response of taxpayers to the negative tax rates and the high positive tax rates under the EITC, we employ a random sample of income tax returns examined by the Internal Revenue Service (IRS) as part of the 1988 Taxpayer Compliance Measurement Program (TCMP). The TCMP data provide detailed information on sources of income, deductions, credits, and tax liabilities, among other taxpayer attributes, before and after adjustment upon audit. The subsample used in the present study consists of 3,219 lowincome households, with about one- 553

NATIONAL TAX JOURNAL VOL. XLIX NO. 4 third ineligible for the credit but included in the sample as a control group. Our empirical strategy permits us to identify the determinants of a household s choice to underreport, overreport, or accurately report income. Conditional on this choice, we also identify the determinants of the amount of income misreported. We find that marginal tax rates have a moderate influence on the reporting of income: the estimated elasticity of underreported income with respect to the tax rate is.27, and it is not significantly different from zero for income overstated. The credit, primarily in the phase-out range, has a moderate effect on the understatement of income, but only for proprietors, explaining less than ten percent of the amount of income understated, or less than two percent of income. The remainder of the paper is organized as follows. First, we briefly review the literature on tax evasion, describe the structure of the EITC, and take a preliminary look at the observed pattern of compliance. Then, we discuss our econometric model of tax evasion and the construction of the variables. Finally, we report the empirical results and provide some concluding comments. BACKGROUND AND DATA SOURCES Allingham and Sandmo (1972), Yitzhaki (1974), and Pencavel (1979) adapt the expected utility framework to the study of tax evasion. In these models, the amount of declared income is a function of the individual s income, income tax rate, audit rate, and penalty rate conditional upon the individual s attitude toward risk. Following the standard comparative static results, the theoretical literature shows that reported income varies positively with income, audit rate, and penalty rate, with the tax rate effect ambiguous. If we assume that declared income decreases with tax rates (Clotfelter, 1983) and credits (negative taxes) are treated symmetrically, as required by the axioms of the expected utility hypothesis, then it follows that declared income increases as the credit rate increases. 3 The reverse, of course, should be observed if declared income increases with tax rates (Feinstein, 1991; Graetz and Wilde, 1985). THE TCMP DATA In order to evaluate the influence of tax and credit rates on the reporting of income, we use data from the IRS s 1988 TCMP. The 1988 TCMP study consists of a randomly selected sample of 54,9 tax returns. Each return in this sample is subject to an extensive line-by-line audit. Thus, we observe the amount of income reported by the taxpayer and, as corrected upon audit, tax and credit rates, among other variables, for every observation. 4 Since we focus on the impact of both negative and high positive marginal rates on low-income taxpayers, we eliminate all returns with adjusted gross income (AGI), or labor income, greater than $18,575 and returns reporting nonpositive income. Furthermore, we eliminate the returns of taxpayers who are either under 18 or over 64 years old, married taxpayers filing separately, individuals claimed as dependents by others, single individuals, and taxpayers subject to the alternative minimum tax. We also exclude observations where the filing status and/or the number of dependent children changed upon audit. 5 After excluding these taxpayers, 554

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES we are left with a sample of 3,219 taxpayers of which 2,153 are eligible for the credit. The remaining 1,66 observations are low-income, childless joint filers who are ineligible for the credit. 6 THE STRUCTURE OF THE EITC In 1988, the tax code provided for a tax credit of 14 cents per dollar of earned income. The maximum credit was $874, which would be attained at $6,225 of earned income. The credit was phased out at the rate of 1 cents per dollar of the greater of earned income or AGI in excess of $9,85 and, thus, completely phased out at an income of $18,576. Unlike wage earners, whose payroll taxes are withheld by the employer, the self-employed report their payroll tax liability when filing their income tax return. Consequently, when we compute the tax liability of the self-employed, we include their Social Security tax liability, which is equal to.132 percent of income. 7 Table 1 provides a summary of the marginal tax rates in effect in 1988 for wages in the column labeled τ w and selfemployment income in the column labeled τ s. In order to simplify the exposition, the table is constructed assuming 1 a single head of household with a dependent child and AGI consisting solely of labor income, or wages, proprietorship, and farm income. 1 It is important to note that, if AGI exceeds labor income, the credit may be reduced and consequently the combined marginal tax rates may be higher for given levels of income. Column 1 shows that individuals with income under $8,35 face a statutory tax rate of zero. Columns 2 and 3 show that the phase-in range of the credit creates a negative marginal tax rate (.14) if the taxpayer s income is less than $6,225. On the other hand, if the taxpayer s income is between $6,225 and $8,3, both the statutory and credit rates are zero. When income is between $8,3 and $9,85, the statutory rate is.15, while the credit rate remains at zero. The phase-out range 2 occurs between $9,85 and $18,576, where the statutory tax rate is.15 and the phase-out rate is.1 for a 2combined rate of.25. Columns 4 and 5 show how these rates vary when Social Security taxes are included. 8 The Observed Pattern of Compliance In Tables 2 and 3, we provide a snapshot of the compliance pattern for EITC eligible and ineligible taxpayers. Table 2 shows the number and average amount of earned income overreported, accurately reported, and underreported, according to EITC status and income class for the entire sample. The three income classes are chosen so that taxpayers eligible for the EITC face either (1) the phase-in range, where Labor Income Under $4 $4 $6,225 $6,225 $8,3 $8,3 $9,85 $9,85 $18,576 TABLE 1 MARGINAL TAX RATES IN 1988 Statutory Rate Due Combined Payroll Combined Rates Rate to Credit Rates Tax Rate Plus Payroll Tax ( τ ) (τ ) c (τ ) w (τ ) p (τ ) s.15.15.14.14.1.14.14.15.25.132.132.132.132.14.98.132.274.374 555

NATIONAL TAX JOURNAL VOL. XLIX NO. 4 TABLE 2 OBSERVED COMPLIANCE BY INCOME CLASS: ALL TAXPAYERS (FREQUENCY, RELATIVE FREQUENCY, CONDITIONAL MEAN, AND STANDARD DEVIATION) EITC ELIGIBLE Adjusted Gross Income Labor Income Overreported <$6,225 $6,225 $9,85 >$9,85 All 32 21.6% 1,711 (2,282) 28 9.4% 1,26 (1,653) 85 5.% 1,175 (1,832) 145 6.8% 1,299 (1,98) Correctly reported 64 43.2% () 117 39.4% () 1,24 6.% () 1,25 55.9% () Underreported 52 35.1% 1,251 (1,224) 152 51.2% 1,795 (1,93) 599 35.1% 2,754 (2,827) 83 37.3% 2,475 (2,642) Total 148 297 1,78 2,153 EITC INELIGIBLE Adjusted Gross Income Labor Income Overreported <$6,225 $6,225 $9,85 >$9,85 All 18 17.8% 1,443 (2,74) 15 11.5% 2,581 (2,635) 4 4.8% 1,291 (1,195) 73 6.8% 1,593 (1,844) Correctly reported 2 19.8% () 38 29.% () 472 56.6% () 53 49.7% () Underreported 63 62.4% 1,471 (1,444) 78 59.5% 2,14 (2,232) 322 38.6% 2,95 (3,348) 463 43.4% 2,66 (3,33) Total 11 131 834 1,66 negative rates occur (AGI less than $6,225); (2) zero marginal rate (AGI between $6,225 and $9,85); or (3) the phase-out range (AGI greater than $9,85). Similarly, Table 3 reports this information for the subsample of sole proprietors. Several patterns emerge from examination of Tables 2 and 3. First, the proportions that overreport, accurately report, and underreport income appear to be invariant to credit eligibility. For those eligible for the EITC, shown in the upper panel of Table 2, 6.8 percent overstate income by an average of $1,299, 55.9 percent accurately report income, and 37.3 percent understate income by an average of $2,475. While, as shown in the lower panel of Table 2, 6.8 percent of the sample ineligible for the EITC overstate income by an average of 556

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES TABLE 3 OBSERVED COMPLIANCE BY INCOME CLASS: SOLE PROPRIETORS (FREQUENCY, RELATIVE FREQUENCY, CONDITIONAL MEAN, AND STANDARD DEVIATION) EITC ELIGIBLE Adjusted Gross Income Labor Income Overreported <$6,225 $6,225 $9,85 >$9,85 All 24 25.3% 1,19 (1,447) 18 9.9% 1,241 (1,74) 56 9.4% 1,25 (1,551) 98 11.3% 1,15 (1,542) Correctly reported 28 29.5% () 41 22.7% () 97 16.4% () 166 19.1% () Underreported 43 45.3% 1,269 (1,25) 122 67.4% 1,892 (1,934) 44 74.2% 3,22 (2,97) 65 69.6% 2,67 (2,761) Total 95 181 593 869 EITC INELIGIBLE Adjusted Gross Income Labor Income <$6,225 $6,225 $9,85 >$9,85 All Overreported 13 18.3% 1,367 (2,216) 1 11.% 1,972 (2,161) 27 8.5% 1,144 (1,192) 5 1.4% 1,368 (1,78) Correctly reported 11 15.5% () 18 19.8% () 61 19.1% () 9 18.7% () Underreported 47 66.2% 1,387 (1,384) 63 69.2% 2,194 (2,35) 231 72.4% 3,257 (3,522) 341 7.9% 2,83 (3,185) Total 71 91 319 481 $1,593, 49.7 percent accurately report income and 43.4 percent understate income by an average of $2,66. Second, the source of income appears to influence taxpayer compliance. For example, comparing Tables 2 and 3, sole proprietors, whether eligible or ineligible for the EITC, appear to be less compliant than the population as a whole, at least judging by the proportion that accurately report income. About 7 percent of proprietors understate income, compared to only 4 percent for the entire sample. Third, observed overreporting may simply reflect taxpayer mistakes. For example, the top panel of Table 2 shows that individuals who are eligible for the EITC and face negative tax rates are two to four times more likely to overreport 557

NATIONAL TAX JOURNAL VOL. XLIX NO. 4 income than those in the other two income groups, or 21.6 percent versus 9.4 and 5. percent, respectively. This pattern could lead one to conclude that negative tax rates induce taxpayers to overreport income. However, the bottom panel shows that individuals ineligible for the EITC exhibit a similar pattern, although they do not face negative marginal tax rates and thus do not have the same incentive to overreport income. The absence of an incentive for taxpayers who are ineligible for the EITC to overreport income suggests that overreporting should be attributed to taxpayer mistakes, and not just to the negative tax rates in the phase-in range of the EITC. 9,1 taxpayer attributes. Though we do not observe the values of the index in each state, we do observe the state chosen, denoted by I. Consider the following trichotomous choice model: 1 I * S = z S γ + η S where s = 1, 2, or 3. Define ε s as follows: 2 MODELING COMPLIANCE While the results in Tables 2 and 3 are suggestive, we turn to multivariate analysis in order to shed further light on the pattern of compliance and its correlates. The observed pattern of compliance reported in these tables raises certain econometric issues not traditionally addressed in the tax compliance literature. Econometric Issues As shown in Tables 2 and 3, we observe taxpayers underreporting, accurately reporting, and overreporting income. Accordingly, we assume taxpayers choose between these three states, denoted by 1, 2, and 3, respectively. Let us suppose an index I * represents the s utility of a taxpayer in state s. Furthermore, suppose the taxpayer chooses the state with the maximum value of the index. Finally, assume the value of the index depends upon a vector z s of ε s = max (I * j ) η s where j = 1, 2, or 3; j s. Then I = s iff fε s < z s γ. If the η values are independent and s follow identical extreme value distributions, then, as Domencich and McFadden (1975) show, 3 e Ζ s γ P (I = s) = 3 Σ e Ζ j γ j=1 where P (I = s) is the probability of event s. Consequently, we estimate the probability of making choice s using a multinomial logit model. The error term captures unobserved characteristics 558

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES (e.g., preferences, attitudes toward risk, and mistakes) that lead individuals to misreport income. For states 1 and 3, we assume the amount of income under- or overreported, y s, depends upon a vector x of personal attributes including certain tax variables: 4 y s = x s β s + u s where s = 1 or 3. We observe y s if and only if I = s. Again, we assume the error term represents unobserved characteristics that influence the amount of misreported income. The joint role of the unobserved characteristics, which influence the probability and the amount of misreported income, means the errors are likely to be correlated. This situation differs from the usual selectivity problem discussed by Heckman (1976), because the error term in the first stage is not normally distributed. Lee (1983) describes a generalization of Heckman s two-stage procedure which can accommodate this situation. In the first stage, conventional multinomial logit estimation is used to estimate the probabilities of noncompliance. In the second stage, we obtain consistent estimates of β and corrected standard errors by estimating the amount of misreported income by ordinary least squares (equation 4) augmented by the inverse Mill s ratio computed as described by Lee. We believe this procedure is dictated by the structure of the credit. It also responds to criticisms of the customary approach adopted in the empirical tax evasion literature for failing to distinguish between overreporters and those who accurately report their income. 11 Variable Construction In order to estimate equations 3 and 4 and to examine the pattern of income misreporting and its determinants, we employ a number of variables traditionally used in the literature to represent taxpayer attributes. The following is a brief description of the variables. Misreported Income: Our dependent variable is defined as the difference (±$1) between corrected labor income and reported labor income. Labor, or earned, income is defined as the sum of wage, proprietorship, and farm income. 12 Detection Rate: Due to the withholding feature of the U.S. income tax, misreported wage income is subject to nearly 1 percent probability of detection, while misreported selfemployment income has a much lower probability of detection. To account for these differences in the probability of detection, we use the share of labor income attributable to wage income (wage share) as a proxy for the probability of detection. Marginal Tax Rate: We compute last dollar marginal tax rates by adding $1 to reported income and calculate the change in tax liability net of the EITC. Since this tax rate is endogenous to the amount of income misreported, we compute a predicted tax rate using a procedure similar to that described in Burman and Randolph (1994). First, we regress the last dollar tax rate on the first dollar tax rate and a vector of taxpayer attributes: marital status, age, age-squared, AGI, and wage share. 559

NATIONAL TAX JOURNAL VOL. XLIX NO. 4 Second, the predicted tax rate is used in the multinomial logit equations to estimate the probability of over- and underreporting income. Third, we compute the predicted tax rate by including the inverse Mill s ratio in the first step. Finally, the predicted tax rate is used in the second stage regressions for the amount of income under- or overreported. 13 Income: Income is defined as AGI as corrected upon audit. We include additional variables to capture preferences and account for third party preparation of tax returns. These variables are described below. Marital Status: A dummy variable that is set equal to one when the filer is married. Age: We use the age of the primary taxpayer with a quadratic specification. Preparer: We create two dummy variables for returns prepared with (1) paid professional help (certified public accountant, attorney, or national tax service) and (2) other paid help. Table 4 provides summary statistics for the sample of tax returns used in the present study. The table also reports summary statistics for three subsamples: returns for which labor income is (1) underreported, (2) accurately reported, and (3) overreported. Table 4 shows that 1,266 taxpayers understated labor income, 218 overstated income, and the remaining 1,735 appear to have accurately reported their income, according to the TCMP audit. The average marginal tax rate for the sample is 17.1 percent, or 19.75 percent using the first dollar tax rate. About 42 percent of the observations report income from proprietorships and 12 percent from farms, while 7 percent are married. The average labor income is approximately $11,784, with an average of $898 misreported or about $1,9 if we use the absolute value of misreported income. Paid preparers are employed by 62 percent of the filers; very few rely on IRS assistance. The credit was claimed by 67 percent of the filers in the sample. EMPIRICAL RESULTS Our empirical strategy is shaped by the striking fact reported in Table 4 that most of those who misreport income are either proprietors or farmers. Accordingly, we stratify our sample into wage earners, proprietors, and farmers and examine the probability of misreporting income and the amount of income misreported, following the framework described above. 14 Distinguishing between the compliance behavior of entrepreneurs and laborers allows us to control for differences in the rates of detection and audit rates and conforms to the empirical finding in Clotfelter (1983) and the theoretical exposition of Pestieau and Possen (1991). 15 The Decision to Misreport Income Tables 5 through 7 report estimates from multinomial logit equations for wage earners, proprietors, and farmers, respectively. In each set of equations, the dependent variable assumes values of, 1, or 2 depending upon whether the filer underreported, accurately reported, or overreported earned income, respectively. For each table, column 1 examines the determinants of the odds of accurately reporting income, while column 2 examines the determi- 56

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES Variables Last $ tax rate TABLE 4 SAMPLE MEANS AND STANDARD DEVIATIONS All Observations.179 (.1129) Understated Accurately Stated Overstated.1634 (.1249) Observations with Labor Income.1747 (.12).1845 (.1216) First $ tax rate.1973 (.19).2276 (.11).182 (.11).1577 (.128) Percent married.72 (.4582).8713 (.3351).5493 (.4977).983 (.2893) Percent proprietor.4194 (.4935).7472 (.4348).1476 (.3548).6789 (.468) Percent farmer.1249 (.336).2228 (.4163).282 (.1657).3257 (.4697) Age 4.54 (12.175) 42.56 (11.78) 38.76 (12.24) 42.99 (11.72) Adjusted gross income 11,784 (4,419) 11,352 (4,348.9) 12,48 (4,23.6) 8,75.1 (4,892.2) Wages/labor income.5746 (.4552).2173 (.3225).8744 (.3126).2637 (.3547) Amount misreported 897.7 (2,265.5) 2,523.1 (2,79.6) 1,397.7 (1,887.5) Percent IRS assisted.28 (.528).46 (.678).46 (.677) Percent VITA * help.84 (.912).32 (.561).133 (.1144) Percent unpaid help.814 (.2735).316 (.175).125 (.3256).596 (.2374) Percent professional help.2355 (.4244).3168 (.4654).1689 (.3747).2936 (.4565) Percent other paid help.3834 (.4863).4652 (.499).3153 (.4648).4495 (.4986) Percent North.1497 (.3569).1414 (.3486).1562 (.3632).1468 (.3547) Percent Midwest.411 (.4919).3697 (.4829).4369 (.4961).4312 (.4964) Percent South.2432 (.4291).2938 (.4557).2 (.41).2936 (.4565) Percent with credit.6688 (.477).6343 (.4818).6945 (.467).6651 (.473) Observations * Volunteer Income Tax Assistance 3,219 1,266 1,735 218 561

562 NATIONAL TAX JOURNAL VOL. XLIX NO. 4

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES 563

564 NATIONAL TAX JOURNAL VOL. XLIX NO. 4

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES nants of the odds of overreporting income, both relative to understating income. The marginal effects are reported in columns 3 through 5. Beginning with our sample of 1,521 wage earners in Table 5, we find that 1,439 (94.6 percent) accurately report their income and 1 (.6 percent) overstate income. The estimated tax rate coefficient for the odds of accurately reporting income in column 1 is.895, with a standard error of 2.14, and is statistically insignificant. Similarly, the estimated coefficients for the remaining variables are insignificant, except for marital status. The estimated coefficients for the odds of overstating income in column 2 are insignificant as well, reflecting the limited degrees of freedom. Turning to our sample of 1,35 proprietors in Table 6, we find that income is accurately reported by 256 (19 percent) and overstated by 148 (11 percent). For observations with income accurately reported, the estimated coefficient for the tax rate in column 1 is 5.857, with a standard error of 1.515. The estimates show that the probability of accurately reporting income is positively correlated with the wage share, which serves as a proxy for the probability of detection and is negatively correlated with the use of other paid help. 16 Similarly, we find the tax rate coefficient is significant for observations with income overstated, in column 2. The estimated marginal effect for the probability of under reporting income with respect to the tax rate, as reported in column 3, is 1.232, which implies an elasticity of.3, computed at mean values. The marginal effect implies that increasing the marginal tax rate by.1 will increase the probability of under reporting income by approximately.12, while an elasticity of.3 implies that the probability of under reporting income will increase by.3 percent for each one percent increase in the marginal tax rate. In the case of overstating income, the marginal effect is.476 (column 5) with an implied elasticity of.72. Finally, Table 7 reports estimates for the sample of 42 farmers. We find that 49 farmers (12.5 percent) accurately report their income and that 71 (17.4 percent) overstate income. Table 7 shows that all the variables are insignificant, with the exception of tax rate and marital status. The tax rate coefficient is statistically significant in the odds of accurately reporting income and overstating income equations. The marital status coefficient is statistically significant for the choice to accurately report income. However, the marginal effects are insignificant for all variables including the tax rate. In an alternative specification, we pool the sample and interact the tax rate with dummy variables indicating the source of income. 17 These results are consistent with those reported in the first two columns of Tables 5 through 7; the estimated coefficients vary by source of income. For taxpayers with income accurately reported, the estimated tax rate coefficients are 2.529, 4.343, and 2.768 (standard error = 1.493, 1.1, and 1.783) and, for those with income overstated, 2.168, 5.435, and 5.786 (standard error = 2.625, 1.33, and 1.571) for wage earners, proprietors, and farmers, respectively. Using a Wald test, we reject the null hypothesis that the three coefficients are equal (x 2 = 2.68 and p =.1). Since they are receiving a tax benefit, taxpayers claiming the EITC may be less 565

NATIONAL TAX JOURNAL VOL. XLIX NO. 4 inclined to cheat because they may have a more positive attitude toward the government (see Carroll, 1992). Therefore, it would be useful to test whether EITC recipients and nonrecipients exhibit the same underlying attitude toward tax compliance. We test for this by including a dummy variable indicating EITC eligibility. The estimated coefficients for this credit indicator using the pooled data are.183 and.319 for the samples which accurately and over report income, respectively, but they are not precisely measured, having standard errors equal to.14 and.191, respectively. Using a Wald test, we fail to reject the null hypothesis that these two coefficients are equal to zero (x 2 = 3.793 and p =.15). Otherwise, the estimated coefficients for the remaining right-hand-side variables, including the tax rate, are unaffected. A similar test was extended to proprietors. The estimated coefficients are.281 and.244 with standard errors of.185 and.231, respectively (x 2 = 2.929 and p =.231). These results are consistent with the credit participation rates observed in Table 4. The Amount of Misreported Income Now we turn to the amount of income misreported. Columns 6 and 7 of Tables 5 through 7 present estimated coefficients for regressions in which the dependent variable is the amount of income misreported (under- and overstated). The independent variables are identical to those employed in Table 5, augmented by the inverse Mills ratio. Beginning with wage earners in Table 5, the conditional average amount of income understated is $718 (5 percent of income) and the average amount overstated is $1,86 (15 percent of income). 18 We fail to find a statistically significant relationship between the explanatory variables and the amount of income understated. As should be expected, given the limited degrees of freedom, the coefficients in the amount of income overstated regression are statistically insignificant. On the other hand, the average amount of income understated by proprietors is $2,718 (24 percent of income) and the average amount of income overstated is $1,194 (13 percent of income). Column 6 of Table 6 shows that the tax rate coefficient is equal to 17,497, implying an elasticity of.27, and is more than twice the size of its standard error. The results also show that the amount of income understated varies positively with income and negatively with the share of wages in earned income. Finally, the amount of income overstated (column 7) does not appear to be affected by the explanatory variables. The tax rate results for taxpayers who understate income are qualitatively consistent with Clotfelter (1983), but contradict those in Feinstein (1991). Our results, however, cannot be directly compared to those in the literature since we account for EITC and Social Security taxes. Furthermore, we limit our sample to low-income households. Turning to farmers, the average amount of income understated is $2,32 (22 percent of income) and the average amount of income overstated is $1,759 (25 percent of income). As with proprietors, the amount of income understated seems to vary inversely with wage share and positively with income. The effect of wage share on the amount of income understated by proprietors and farmers is consistent with the high probability of detection for income subject to third party withholding. As in the case of wage earners and propri- 566

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES etors, we are unable to account for the determinants of the amount of income overstated. Some Simulations Except for proprietors, misreported income seems to be invariant to marginal tax rates, including the EITC rates. Accordingly, we simulate the effect of credit rates on proprietors only. Using the estimated coefficients in Table 6, we undertake four experiments. In each experiment, we peel off a layer of the EITC, recompute the tax rate, and estimate the effect on compliance, holding all other variables constant. First, we eliminate the phase-in range of the EITC and the corresponding credit rate of.14. Using row 2 and column 5 of Table 1 and assuming zero wage income, eliminating the phase-in range increases the tax rate from -.98 to.132. This change in the credit rate increases the average probability of understating income from.696 to.774, or by 1.68 percentage points. It decreases the average probability of overstatements from.1951 to.1873, or by.78 percentage points, and increases the amount of income underreported by an average of $2. Second, we set all negative tax rates equal to zero. Again, using row 2 and column 5 of Table 1, this change increases the tax rate from -.98 to zero. This increase in the tax rate increases the probability of understating income by.7 percentage points and lowers the probability of overstatements by.34 percentage points, with negligible effects on the amount of income underreported. Third, we eliminate the phase out of the credit and the corresponding increase of.1 in the marginal tax rate. This change, which lowers the maximum tax rate from.374 in Table 1 to.274, decreases the probability of underreporting income by 5.2 percentage points and the amount underreported by an average of $18 and increases the probability of overreporting income by 1.96 percentage points. Finally, we eliminate the credit altogether. This change reflects the offsetting effects of eliminating the phase-in range and the phase-out range. As demonstrated above, eliminating the phase-in range will increase the probability of understating income, while eliminating the phase-out range will reduce the probability of understating income. The net effect of eliminating the credit decreases the probability of understatements by 3.34 percentage points and the amount of income understated by an average of $161. The probability of overstatements increases by 1.64 percentage points with an insignificant effect on the amount overstated. These simulations demonstrate that the phase-out range is the principal source of income reporting errors by proprietors caused by the credit. One may conclude, however, that misreported income is only a minor problem: overall, the credit has a negligible impact on the probability of misreporting income and increases the amount of income understated by approximately 9 percent of the amount of income understated or approximately 1.5 percent of income. Concluding Comments We examine the influence of tax and credit rates on reported labor income. Our sample consists of 3,219 lowincome filers drawn from the 1988 TCMP, with one-third ineligible for the credit. We stratify the sample accord- 567

NATIONAL TAX JOURNAL VOL. XLIX NO. 4 ing to occupational choice and obtain multinomial logit estimates of the odds of accurately reporting income and the odds of overreporting income, both relative to underreporting income. Conditional on the taxpayer s choice, we also examine the determinants of the amount of income misreported. Except for proprietors, misreported income seems to be invariant to tax and credit rates. We find that tax rates play a moderate role in shaping compliance behavior. The estimated elasticity of income understated with respect to the tax rate is.27, and not significantly different from zero for income overstated. The marginal tax rates induced by the EITC do not appear to affect the amount of income overreported, irrespective of income source, nor, except for proprietors, the amount of income underreported. Even in the case of proprietors, the effects of the EITC on the understatement of income are modest: the credit, primarily the phaseout range, explains less than ten percent of the amount of income understated, or less than two percent of income. Note, however, that our results apply to low-income taxpayers and may not apply to taxpayers who have significantly higher incomes, are subject to higher audit rates, or enjoy greater opportunities to conceal income. In addition, our findings reflect 1988 law and IRS administration, which differ significantly from current tax law and administration. ENDNOTES We would like to thank Lowell Dworin, Janet Holtzblatt, Janet McCubbin, Ann Parcell, Karl Scholz, and three anonymous referees for their comments. The views expressed in this paper are those of the authors and do not necessarily reflect those of the U.S. Treasury. 1 For studies that assess the effectiveness of the EITC by examining its labor supply effects and participation rates, see Browning (1993), Hoffman and Seidman (199), Holtzblatt, McCubbin, and Gillette (1994), Kosters (1993), and Scholz (199, 1994). Also, see Killingsworth (1983). 2 Under current law, these rates range from.4 to.49. See Munnell (1994) and Scholz (1994). We include Social Security taxes in computing these rates. Also, see the Office of Management and Budget (1994) for the estimated cost of the current EITC. 3 See Kahneman and Tversky (1979) for a discussion of the inadmissibility of preference reversals in an expected utility framework. 4 Beron, Tauchen, and Witte (1992) discuss the importance of socioeconomic variables on taxpayer compliance. Unfortunately, tax data do not provide a very rich set of demographic variables. For a critique of TCMP data, see Long (1992). It should be noted that the TCMP does not incorporate the effects of other IRS enforcement activities. 5 It is our understanding that many of the adjustments to the number of dependents of lowincome households can be attributed to their failure to satisfy the dependency requirements because these households receive government assistance. Since the data do not provide information on public assistance, such as Aid to Families with Dependent Children, we are unable to distinguish between an adjustment to the number of dependents from failing to satisfy the dependency test and one due to claiming fictitious children in order to qualify for the EITC. Although misreporting filing status and the number of dependents is an alternative mode of tax evasion induced by the structure of the EITC, modeling the simultaneity between misreporting of income, filing status, and dependents is beyond the scope of this paper. For an investigation of multiple modes of tax evasion, see Klepper and Nagin (1989) and Martinez-Vazquez and Rider (1994). 6 We choose low-income, childless joint filers because two-member households more closely resemble the family structure of eligible taxpayers. 7 Note that one-half of the self-employment tax is deductible in computing individual income tax liability. 8 We define the marginal tax rate as follows: α τ w + (1 α) τ s if eligible for the EITC, and α τ + (1 α) (τ + τ p.5 τ τ ) otherwise, where α is p the wage share, τ is the statutory income tax rate, τ w is the statutory rate plus the credit rate, τ is p the Social Security tax rate, and τ s = τ w + τ p.5 τ τ. The marginal tax rates in effect in 1988 are p provided in the corresponding columns of Table 1. Note that a more appropriate specification of the marginal tax rate would account for state tax rates. 568

TAX EVASION IN THE PRESENCE OF NEGATIVE INCOME TAX RATES 9 Alternatively, MacKie-Mason (1992) speculates that overreporting may represent the true preference of taxpayers who do so in order to reduce their probability of audit. 1 Including taxpayers who are ineligible for the EITC in the sample helps alleviate two potential problems. First, as explained above, observed noncompliance may simply reflect taxpayer mistakes. Including a control group is an alternative method to that employed in Alexander and Feinstein (1986) and Rice (1992) to control for the noise created by unintentional mistakes. Other researchers could not pursue this strategy because they lacked the necessary data. Second, individuals in the two groups with identical incomes may face different marginal tax rates. Introducing such heterogeneity into the sample may help reduce the identification problem whereby the tax rate captures the income effect (Poterba, 1987; Feenberg, 1987). 11 MacKie-Mason (1992) argues that Tobit does not allow for the possibility that taxpayers may overreport income. The credit range of the EITC is not the only feature of the tax code that creates an incentive for taxpayers to overreport certain income. For example, the tax treatment of capital gains creates incentives to overstate long-term capital gains by misclassifying short-term capital gains and other ordinary income. Therefore, overreporting income need not be entirely inadvertent. 12 A distinction should be made between actual income, income reported for tax purposes, and income detected by a tax audit. An auditor may not detect all income received which introduces a potential source of bias. Feinstein (1991) presents a partial detection model of tax evasion to control for such bias. Unfortunately, we do not have the necessary data to implement his technique. However, the difference between the amount of evasion and the amount of detected evasion should not be as great for low-income returns as for high-income returns. 13 Note that the reported standard errors may be understated because we do not account for the use of instrumental variables. 14 The sample of wage earners consists of taxpayers whose only source of labor income is wages. The sample of proprietors consists of taxpayers filing a Schedule C. The sample of farmers consists of taxpayers filing a Schedule F. There are 32 taxpayers that file both a Schedule C and a Schedule F and are included in the subsamples for proprietors and farmers. 15 However, we do not address two potential problems: partial detection and endogenous labor supply. For example, IRS auditors may not detect all the income that is misreported. If the amount detected is correlated with the right-hand-side variables, then the parameter estimates may be biased. Similarly, if labor supply is correlated with the right-hand-side variables, particularly the tax rate, then the parameter estimates may be biased. Since we use disposable AGI instead of labor income, endogenous labor supply may not represent as serious a problem as it might otherwise. 16 The reliance on paid preparers may not be exogenous to the evasion decision (Erard, 1993; Udell, 1995). Nevertheless, the estimated coefficient on the tax rate, the variable of interest, does not change by much when we omit this dummy variable. 17 These results are not reported but are available from the authors upon request. 18 For this subsample, we are unable to compute corrected standard errors because of the limited degrees of freedom. REFERENCES Alexander, Craig, and Jonathan Feinstein. A Microeconomic Analysis of Income Tax Evasion. Massachusetts Institute of Technology. Mimeo, 1986. Allingham, Michael G., and Agnar Sandmo. Income Tax Evasion: A Theoretical Analysis. Journal of Public Economics 1 No. 3 (1972): 323 38. Beron, Kurt J., Helen V. Tauchen, and Ann D. Witte. The Effect of Audits and Socioeconomic Variables on Compliance. In Why People Pay Taxes: Tax Compliance and Enforcement, edited by Joel Slemrod, 67 89. Ann Arbor: The University of Michigan Press, 1992. Browning, Edgar K. Effects of the Earned Income Tax Credit on Income and Welfare. Texas A&M University. Mimeo, 1993. Burman, Leonard E., and William C. Randolph. Measuring Permanent Responses to Capital-Gains Tax Changes in Panel Data. American Economic Review 84 No. 4 (September, 1994): 794 89. Carroll, John S. How Taxpayers Think About Their Taxes: Frames and Values. In Why People Pay Taxes: Tax Compliance and Enforcement, edited by Joel Slemrod, 43 63. Ann Arbor: The University of Michigan Press, 1992. Clotfelter, Charles T. Tax Evasion and Tax Rates: An Analysis of Individual Returns. Review of Economics and Statistics 65 No. 3 (August, 1983): 363 73. Domencich, Thomas, and Daniel McFadden. Urban Travel Demand. Amsterdam: North- Holland Publishing Company, 1975. 569

NATIONAL TAX JOURNAL VOL. XLIX NO. 4 Erard, Brian. Taxation with Representation: An Analysis of the Role of Tax Practitioners in Tax Compliance. Journal of Public Economics 52 No. 2 (September, 1993): 163 98. Feenberg, Daniel. Are Tax Price Models Really Identified: The Case of Charitable Giving. National Tax Journal 4 No. 4 (December, 1987): 629 33. Feinstein, Jonathan S. An Econometric Analysis of Income Tax Evasion and Its Detection. RAND Journal of Economics 22 No. 1 (Spring, 1991): 14 35. Graetz, Michael, and Louis Wilde. The Economics of Tax Compliance: Fact and Fantasy. National Tax Journal 38 No. 3 (June, 1985): 355 63. Heckman, James. The Common Structure of Statistical Models of Truncation, Sample Selection, and Limited Dependent Variables and a Simple Estimator for Such Models. Annals of Economic and Social Measurement 5 No. 4 (Fall, 1976): 475 92. Hoffman, Saul D., and Laurence S. Seidman. The Earned Income Tax Credit: Antipoverty Effectiveness and Labor Market Effects. Kalamazoo: W. E. Upjohn Institute for Employment Research, 199. Holtzblatt, Janet, Janet McCubbin, and Robert Gillette. Promoting Work Through the EITC. National Tax Journal 47 No. 3 (September, 1994): 591 67. Kahneman, Daniel, and Amos Tversky. Prospect Theory: An Analysis of Decision Under Risk. Econometrica 47 No. 2 (March, 1979): 263 91. Killingsworth, Mark R. Labor Supply. Cambridge: Cambridge University Press, 1983. Klepper, Steven, and Daniel Nagin. The Anatomy of Tax Evasion. Journal of Law, Economics, and Organization 5 No. 1 (Spring, 1989): 1 24. Kosters, Marvin. The Earned Income Tax Credit and the Working Poor. The American Enterprise 4 No. 3 (May June, 1993): 65 72. Lee, Lung-Fei. Generalized Econometric Models with Selectivity. Econometrica 51 No. 2 (March, 1983): 57 12. Long, Susan B. Commentary. In Why People Pay Taxes: Tax Compliance and Enforcement, edited by Joel Slemrod, 115 23. Ann Arbor: The University of Michigan Press, 1992. MacKie-Mason, Jeffrey. Commentary. In Why People Pay Taxes: Tax Compliance and Enforcement, edited by Joel Slemrod, 162 6. Ann Arbor: The University of Michigan Press, 1992. Martinez-Vazquez, Jorge, and Mark Rider. Multiple Modes of Tax Evasion: Theory and Evidence from the TCMP. Policy Research Center, Georgia State University. Mimeo, 1994. Munnell, Alicia. The Coming of Age of the Earned Income Tax Credit. NTA Forum No. 17 (Winter, 1994): 1 6. Pencavel, John H. A Note on Income Tax Evasion, Labor Supply, and Nonlinear Tax Schedules. Journal of Public Economics 12 No. 1 (August, 1979): 115 24. Pestieau, Pierre, and Uri M. Possen. Tax Evasion and Occupational Choice. Journal of Public Economics 45 No. 1 (June, 1991): 17 25. Poterba, James M. Tax Evasion and Capital Gains Taxation. American Economic Review 77 No. 2 (May, 1987): 234 9. Rice, Eric M. The Corporate Tax Gap: Evidence on Tax Compliance by Small Corporations. In Why People Pay Taxes: Tax Compliance and Enforcement, edited by Joel Slemrod, 125 61. Ann Arbor: The University of Michigan Press, 1992. Scholz, John Karl. The Participation Rate of the Earned Income Tax Credit. Institute for Research on Poverty Discussion Paper No. 928 9. Madison, WI: University of Wisconsin Madison, 199. Scholz, John Karl. The Earned Income Tax Credit: Participation, Compliance, and Antipoverty Effectiveness. National Tax Journal 47 No. 1 (March, 1994): 59 81. Udell, Michael. Tax Return Preparers and Tax Evasion. U.S. Joint Committee on Taxation. Mimeo, 1995. U.S. Office of Management and Budget. Budget of the United States Government Analytical Perspectives, Fiscal Year 1995. Washington, D.C.: Government Printing Office, 1994. Yitzhaki, Shlomo. A Note on Income Tax Evasion: A Theoretical Analysis. Journal of Public Economics 3 No 2 (May, 1974): 21 2. 57