Few public policy issues receive greater attention than the

Transcription

1 Impact of the Earned Income Tax Credit on Health Insurance Coverage Evaluating the Impact of the Earned Income Tax Credit on Health Insurance Coverage Abstract - The goals and design of the Earned Income Tax Credit suggest that it has the potential to affect private health insurance coverage rates in its target population through income, tax price and employment effects. Results from an analysis using data from the National Longitudinal Survey of Youth show that the EITC expansion of the mid 99s did increase the rate of employer based health insurance coverage in the low skilled population. The overall effect of EITC expansions was to increase the probability of coverage by 3.8 percentage points, or approximately 375, more individuals who were covered by employer provided policies between 992 and 998. Reagan A. Baughman Whittemore School of Business & Economics, University of New Hampshire, Durham, NH 3824 National Tax Journal Vol. LVIIl, No. 4 December 25 INTRODUCTION Few public policy issues receive greater attention than the significant number of individuals in the United States who lack health insurance. The Census Bureau (U.S. Department of Commerce, 24) estimated that in 23 more than 5 percent of all Americans were without health insurance coverage. Largely responsible for this problem is a dramatic decline in the fraction of private sector workers covered by employer sponsored insurance policies, from 7.9 percent in 979 to 64.5 percent in 997 (Farber and Levy, 2). A particularly striking finding of recent Census Bureau studies of health insurance coverage has been that nearly half of poor, full time workers have no health coverage. In fact, the working poor are less likely than the non working poor to be covered by a health insurance policy. Until now, no one has addressed the potential impact of this country s largest income support program on health insurance coverage for low income workers. The Earned Income Tax Credit (EITC) currently provides refundable tax credits of up to $4,3 per year to families with qualifying children and incomes below $35,458 per year. In 992 it surpassed Aid to Families with Dependent Children (AFDC) as the largest federal cash transfer program for the poor (Liebman, 998). Evidence suggests that the credit has been an effective policy tool along a number of dimensions, and there are It is credited with raising incomes above the poverty line (Center on Budget and Policy Priorities, 998), increasing the labor force participation of single mothers (Eissa and Liebman, 996; Meyer and Rosenbaum, 2), and even encouraging human capital investments (Smeeding, Ross and O Connor, 2). 665

2 NATIONAL TAX JOURNAL three specific reasons to believe that the EITC might encourage employment based health insurance coverage among low income workers. First, the income effect of the large annual benefit available to low income working families makes employee co premiums for insurance coverage more affordable. Second, the structure of the credit is such that workers receiving the EITC face high marginal tax rates as their benefits are phased out. Liebman (998) demonstrates that the EITC combined with Old Age, Survivors, Disability and Health Insurance (OASDHI) payroll taxes and state income taxes produces marginal tax rates in excess of 5 percent for the 65 percent of EITC recipients who fall into the phaseout region of the credit. Because of the differential treatment of wage and benefit income in the U.S. personal income tax system, workers receiving the EITC have strong incentives to demand compensation in benefit rather than wage form. Finally, it should be noted that helping low income workers to afford health care was once an explicit policy goal of the EITC program. The Omnibus Budget and Reconciliation Act (OBRA) of 99 authorized the EITC Supplemental Health Credit, which provided a refundable tax credit on top of the basic EITC for families purchasing health policies covering an EITC eligible child. The structure of the credit mirrored that of the basic credit, and the maximum amount for 992 was $45. The General Accounting Office (GAO) estimated in 994 that this credit reimbursed approximately 23 percent of the average reported cost of a private policy among families that took the credit. OBRA 993 removed the credit for several reported reasons, including fraud on the part of several insurance companies (U.S. Congress, 993) and low participation levels. A GAO report (994) attributed low participation to the small fraction of health insurance costs that the credit reimbursed and low awareness on the part of eligible tax filers. According 666 to 993 Congressional testimony, there was widespread support for ending the supplement and an understanding that the amount of the supplement removed would simply be folded into a larger basic cash EITC payment, which came about starting in 994 (U.S. Congress, 993). The expansions that began in 994 and continued through 996 were the largest in the history of the EITC. The goal of this paper is to use this source of exogenous variation in tax rates and benefit amounts to determine whether the program encouraged private health insurance coverage for its target population. My results suggest that the EITC expansions did increase the likelihood that low skill workers were employed and covered by employment based health insurance policies. The remainder of the paper is organized as follows. The second section provides background on the structure of the EITC and the third section uses this information to form hypotheses about how the credit might affect health insurance coverage. The fourth section discusses the data and empirical strategy used to test these hypotheses and the fifth section presents the results of the empirical models. The sixth section discusses the likely mechanisms by which the EITC affected health insurance coverage during the 99s and the seventh section concludes. STRUCTURE OF THE EARNED INCOME TAX CREDIT The EITC provides refundable tax credits to workers with low family incomes, regardless of marital status. In 24, it was available to individuals with incomes of up to $2,8 (no children), $3, (one child) or $35,458 (two or more children). Individuals who file for the credit have their earnings up to a certain amount matched at a phase in rate. Recipients receive maximum benefits up to another cutoff level of income, and then the maxi-

3 Impact of the Earned Income Tax Credit on Health Insurance Coverage mum benefit is reduced at a phaseout rate until the point where family income equals maximum income for EITC eligibility. Maximum benefits, income cutoffs, and phase in and phaseout ranges all depend upon the number of dependent children claimed by the tax filer. Although a small credit, with a maximum amount of only $39 in 24, has been available to childless filers since 994, the program primarily targets families with children. The period between 989 and 998, which saw the largest growth in EITC benefits and eligibility in the program s history, will be the focus of this analysis. Table provides the full set of federal EITC parameters for these years. Consider, for example, a family with two children in 997. This family could claim a credit equal to 4 cents per dollar earned over the first $9,4 in earnings. A flat credit of $3,656 would be available if the family s annual income fell between $9,4 and $,93. If the family s income exceeded $,93, the maximum credit would be reduced by 2.6 cents per dollar earned until it was entirely phased out at $29,29. As the table shows, maximum benefit levels and the income ranges for eligibility both increased over the period. The largest increase took place between 993 and 996, due to a program expansion mandated by OBRA 993 and phased in over the years 994, 995 and 996. For a family with one (two) child(ren), between 993 and 996 the maximum credit increased by $78 ($2,45) and the maximum income for eligibility rose by $2,28 ($5,445). Additionally, not only did the average benefit for all eligible workers with TABLE EARNED INCOME TAX CREDIT PARAMETERS, Year 989 Kids + Phase In Rate (%) 4. Phase In Range ($) 6,5 Maximum Credit ($) 9 Phaseout Rate (%). Phaseout Range ($),24 9, ,8 953.,73 2, ,4 7,4,92, ,25 2, ,52 7,52,324, ,84 22, ,75 7,75,434, ,2 23, , 7,75 8, ,38 2, , 9,, 23,755, 25, , 6,6 8, ,94 3, ,3 9,23,29 24,396,29 26, ,22 6,33 8, ,52 3, ,28 9,5,6 25,78,6 28, ,34 6,5 9, ,2 3, ,43 9,77,93 25,76,93 29, ,45 6,65 9, ,27 3,756 Sources: Liebman (998) and U.S. Treasury Publication 596 (various years) ,6, 2,3 26,45 2,3 3,95

4 NATIONAL TAX JOURNAL children increase over this period, but the incremental benefit for families with two or more children also increased by $,327 over these four years. The growth of both overall benefits and incremental benefits for additional children will be used in the analysis in order to identify the impact of the EITC on health insurance coverage. Additionally, as is shown in Table 2, ten states either implemented or changed the generosity of their own supplemental EITCs during the 99s. In 989 only Wisconsin and Vermont offered refundable credits, while Maryland and Rhode Island offered non refundable ones. By 998, Kansas, Massachusetts, Minnesota, New York, Iowa and Oregon had added state credits, and Maryland offered a refundable portion. State generosity for refundable credits ranges from a minimum of four percent of the federal EITC in Wisconsin for families with one child ( ) to a maximum of 75 percent of the federal EITC in Wisconsin for families with three or more children ( ). This translates to annual benefits that are State Iowa First Available 99 TABLE 2 STATE EITC PARAMETERS, Rate as Percent of Federal EITC 6.5% Refundable No Without Qualifying Children Kansas 998 % Massachusetts 997 % Maryland : 5% (NR) : 5% (NR) % (R) No Minnesota : % 998: 5% (NC) 2% (C) 999: 5% (NC) 22 46% (C) New York : 7.5% 995: % : 2% Oregon 997 5% No Rhode Island : 22.96% 99 97: 27.5% 998: 27% 999: 26.5% No Vermont : 25% 99 93: 28% : 25% Wisconsin : 5% ( C) No 25% (2 C) 75% (3+ C) : 4% ( C) 6% (2 C) 5% (3+ C) : 4% ( C) 4% (2 C) 43% (3+ C) Source: Thanks to Nick Johnson at the Center on Budget and Policy Priorities for providing this data. R = Refundable. NR = Non refundable. C = Available to filers with qualifying children. NC = Available to filers without qualifying children. 668

5 Impact of the Earned Income Tax Credit on Health Insurance Coverage anywhere between $84 and $,65 above and beyond federal EITC payments in 997. Changes in EITC at the state level during the 99s provide an additional source of exogenous variation by which to measure the impact of the credit on health insurance coverage. EFFECTS OF THE EITC ON HEALTH INSURANCE COVERAGE The goal of the EITC is to remove working families from poverty by increasing employment incentives. Therefore, the most obvious way in which the credit could affect health insurance coverage is by simply moving more individuals into the workforce. While not all low wage employers offer their employees private health insurance, some do, and so the EITC expansions, which have been documented to have increased labor force participation (Eissa and Liebman, 996; Meyer and Rosenbaum, 2), may have increased the net number of those privately insured in the United States. Changes in employment status have the potential to either move formerly uninsured non workers or Medicaid covered non workers into jobs with employer provided insurance. While this is a desirable indirect outcome of the employment incentives built into the credit, the main focus of this analysis is the income and substitution effects of the credit that theory would predict to increase health insurance coverage of EITC recipients. It will be estimates of the magnitudes of these effects, apart from increased coverage through higher labor force participation, that provide useful information to policymakers about not only the relationship between the EITC and health insurance, but about the effectiveness of tax subsidies for health insurance in the low income population more generally. Substitution Effect In a very simple model of a competitive labor market, at the margin, employee compensation (wages plus benefits) is set to equal the value of that employee s marginal product. Within that fixed dollar amount, wages and benefits are substitutable forms of compensation. In the absence of tax subsidies, the optimal mix of wages and benefits offered depends upon worker characteristics including tastes, family size and risk aversion, as well as savings from risk pooling due to firm size. The current U.S. tax system, which does not treat all forms of compensation equally, introduces distortions to the optimal mix of wages and benefits in a compensation package (Feldstein and Friedman, 977). Current federal tax law excludes the value of health insurance and other employer provided fringe benefits from the base for federal and state personal income taxes. This system creates incentives for employees to favor compensation packages with greater amounts of health insurance than would otherwise be the case. 2 Specifically, because of this tax subsidy, workers facing higher marginal tax rates may demand more benefits relative to wages. Several authors estimate the responsiveness of demand for health insurance and other employment based benefits to tax subsidies. In a time series analysis with aggregate U.S. data for 954 to 979, Turner (987) estimates that approximately five percent of the growth in fringe benefits over these 25 years was due to the increased value of the tax subsidy. More recent studies using natural experiment methodologies find even stronger ef- 2 Employer incentives resulting from tax law are likely to be relatively minor. To the firm, wages and benefit contributions are both deductible for the purposes of the federal corporate income tax. Only for the payroll tax are wages taxable while benefit contributions are not. Employers benefit from payroll tax treatment only to the extent that they do not pass the tax on to their workers. 669

6 NATIONAL TAX JOURNAL fects. Woodbury and Hamermesh (992) examine the impact of tax rates on health coverage for a panel of college faculty over the period of tax changes implemented by the Tax Reform Act of 986 (TRA86). They calculate that for their sample, the share of total compensation in benefit form was lowered by a half of a percentage point due to the drop in marginal tax rates after TRA86. Gruber and Poterba (994) also develop an identification strategy based upon variation in marginal tax rates caused by TRA86, focusing upon a different aspect of tax reform a provision that for the first time allowed self employed individuals to claim a tax deduction of 25 percent of their health costs. 3 They find a strong response to the tax subsidy. For single individuals the effect is strongest: a one percent increase in the cost of insurance decreases the probability of coverage by.8 percentage points. Royalty (2) exploits variation in state level personal income tax rates and identifies the effect of the tax subsidy on the availability employer provided health insurance by comparing it to the availability of employer provided sick leave, a non tax preferred fringe benefit. She finds that a one percentage point increase in tax rates leads to a.8 to. percentage point increase in the likelihood that an employee is eligible for employer sponsored health insurance. The findings in all of these studies support the hypothesis that tax treatment affects demand for health insurance. However, with the exception of Royalty (2), they estimate the tax effect on insurance coverage primarily for those with higher incomes because the identification in their econometric models derives almost exclusively from marginal tax rate changes in the top brackets. Individuals with high incomes are not, however, the only ones to face high marginal tax rates. The structure of the EITC creates different marginal tax rates for individuals at different points on the income schedule, which have different implications for health insurance demand. The substitution, or price, effect of the tax subsidy either encourages or discourages health insurance coverage, depending upon the recipient s income level. The substitution effect in the phase in range makes benefits costly relative to wages, because wages lead to an increase in the EITC, while benefit compensation does not. In the flat maximum benefit range of the credit, there is only an income effect. In the phaseout range, where benefits are reduced for every wage dollar earned, a price effect makes earnings in benefit rather than wage form attractive. High implicit marginal tax rates caused by the phasing out of benefits should, in theory, increase the demand of workers at these income levels for larger shares of benefits in a compensation package. How significant an impact the substitution effect might have on health insurance coverage of low income workers depends upon several factors, including how many of these workers file for the EITC, how many fall into the phaseout range, and how well these recipients understand the structure of the credit. The GAO (2) estimates that 75 percent of all eligible individuals and 86 percent of all eligible individuals with children filed for the EITC in 999. Liebman (998) finds that a large fraction of recipients, 65 percent, fall into the phaseout range of the credit; the GAO (997b) estimates this fraction to be 59 percent in 994. These estimates taken together suggest that an EITC price effect could be large enough to influence coverage rates of the entire population of low income workers. Whether EITC recipients fully understand how the credit phases out at higher levels of earnings is still unclear, but there are reasons 3 Previously, the self employed had not received any tax benefit for health care unless they took an itemized deduction for health care expenses in excess of five percent of adjusted gross income. 67

7 Impact of the Earned Income Tax Credit on Health Insurance Coverage to believe that understanding has been improving significantly as the credit has expanded. Romich and Weisner (2) find that the majority of the respondents in their ethnographic study are aware of the EITC and that understanding of marginal incentives is correlated with labor market experience. However, only one of the respondents, who had relatively strong labor force attachment, was able to describe the phase in/phaseout design of the credit in detail. The other major necessary condition for an EITC price effect to encourage employer based insurance is that recipients have the ability to affect their compensation package. One important factor in this respect is attachment to the labor force. A National Center for Health Statistics (U.S. Department of Health and Human Services, 997) study finds that most establishments offering benefits require close to full time work hours for eligibility and that firms in establishments with highest turnover (services, retail, construction) are least likely to offer benefits. Liebman (998) finds in 99 that 75 percent of EITC recipients worked more than, hours per year (or 2 hours per week) and 6 percent worked more than,5 hours per year (or 3 hours per week). My own tabulations from a Survey of Income and Program Participation (SIPP) sample show that workers with family incomes that make them eligible for the EITC in 993 work an average of 3 hours per week, compared to an average of 36 hours per week for all workers. The 8 percent of EITC eligible workers whose incomes put them into the phaseout range of the credit work an average of 33 hours per week. This evidence suggests that many EITC recipients have strong enough labor force attachment to be offered health insurance. Changes in the fraction of individuals who are offered health insurance by an employer (as opposed to the probability that workers actually sign up for coverage) 67 would reflect only the substitution effect of the tax subsidy. Therefore, an additional factor influencing the strength of the substitution effect is how EITC recipients are able to influence the probability that they are offered employment based insurance. Goldstein and Pauly (976) present a model in which employer provided fringe benefits are analogous to public goods and employees sort into jobs according the their preferences for wages versus benefits. In this sort of a model, the preferences of the median worker generally determine the wage/benefit package. Although EITC eligible workers are unlikely to be the median workers in most firms, they should be able to sort into jobs where the wage/benefit packages match their preferences. Scott, Berger and Black (989) show that the tax treatment of fringe benefits does increase sorting in the labor market. Income Effect The goal of the EITC, as a refundable tax credit, is to raise recipients annual incomes. The income effect of wages being subsidized, assuming that health care is a normal good, should consistently increase demand for insurance for all EITC recipients. Recent empirical work suggests that the income elasticity of demand for health insurance is substantial; estimates using aggregate data range between. and 2. (Phelps, 997). There are several ways in which income could increase health insurance coverage rates. First of all, higher income would make a privately purchased non group (not employer sponsored) insurance plan more affordable. Higher income would also increase the affordability of employer sponsored plans that require employees to pay a fraction of the premium out of pocket. Levy (998) estimates that 4 percent of the employers in the 993 Robert Wood Johnson Foundation Employer Survey required that employees contribute something toward

8 NATIONAL TAX JOURNAL their insurance premium; the average amount of the contribution (if required) was 35 percent of the cost of the policy. For the majority individuals who make co premium contributions after tax, there is no tax subsidy incentive affecting the take up decision once an employee has already been offered employer sponsored insurance. An increase in participation in employer sponsored health insurance plans conditional upon being offered coverage would be primarily attributable to an income effect of the EITC. It should be noted that Congress eliminated the targeted EITC Supplemental Health Credit after 993 due to low participation and concerns about fraud, but expanded the basic credit at the same time. If, in fact, some portion of the large OBRA 993 expansions was from money formerly targeted at health care, a relevant policy question is whether or not this expansion actually affected health coverage. And, in any case, asking the more general question of whether an EITC income effect exists provides some evidence as to whether tax based cash transfers to the working poor are an effective policy tool for promoting health insurance coverage. EMPIRICAL ANALYSIS To perform the empirical analysis, I use a sample of individuals from the National Longitudinal Survey of Youth (NLSY) between 989 and 998 that includes both workers and non workers. Respondents to the NLSY were originally sampled in 979, when they were between the ages of 4 and 2. They have been interviewed in every year since, until 994 when interviews moved to every two years, and each interview has collected information on demographics, family structure, employment, income, and health insurance coverage. Both employment and health insurance coverage status are coded at the time of each interview. This data set is well suited to the analysis because it follows a cohort of young workers during the period when they are most likely to be EITC eligible. The NLSY provides information on health insurance coverage and demographics that is comparable to what is collected in the Current Population Survey; additionally, the panel nature of the NLSY allows for the inclusion of individual fixed effects in the empirical models. I select a sample in which each person (i) is present for all interviews during the period I consider and (ii) has 2 or fewer years of schooling in I am able to follow 4,47 individuals over ten years and observe changes in both their employment and health insurance status. I use data from six interviews over these ten years (989, 99, 992, 994, 996 and 998). Table 3 gives descriptive statistics for the sample broken down by year and by whether or not the respondent has a child. Because the NLSY follows a single cohort, the age range of the sample is from 24 to 32 in 989 and from 33 to 4 in 998; average age changes correspondingly. Overall, approximately 6 percent of the sample is black and 5 percent is female. Individuals are more likely to be married over time (55.7 percent in 989 and 6.6 percent in 998) and to have more children (an average of. in 989 and.6 in 998). Also consistent with an aging sample, average family income rises from $23,96 in 989 to $37,866 in 998. These average incomes fall just outside the phaseout range of the EITC in each year, but there is significant heterogeneity in income, so many individuals in the sample are potential EITC recipients. Individuals with children are more likely to be mar- 4 Some individuals in the NLSY do obtain more education over the course of the panel. I also estimate models for a sample in which individuals may have no more than 2 years of education in any year. All results presented in the paper are robust to this specification change. 672

9 Impact of the Earned Income Tax Credit on Health Insurance Coverage Percent Married Mean Age Percent Black Mean Education Number of Children Percent Female Family Income Percent Married Mean Age Percent Black Mean Education Number of Children Percent Female Family Income Percent Married Mean Age Percent Black Mean Education Number of Children Percent Female Family Income Percent Married Mean Age Percent Black Mean Education Number of Children Percent Female Family Income Percent Married Mean Age Percent Black Mean Education Number of Children Percent Female Family Income Percent Married Mean Age Percent Black Mean Education Number of Children Percent Female Family Income TABLE 3 SAMPLE DESCRIPTIVE STATISTICS With Children $25, $26, $29, $32, $37, $4,382 Without Children $9, $2, $2, $22, $25, $27,99 Source: Author s tabulations from National Longitudinal Survey of Youth (NLSY) sample of individuals interviewed all years between 989 and 998 and with 2 years of schooling or less in 989. N = 26,499. ried, female, non black and have higher family incomes. For purposes of a baseline difference in difference analysis, I code individuals in each year as belonging to one of eight status groups that combine employment and health insurance coverage. Using joint employment/insurance categories accomplishes two things. First, by separating employment and non employment I allow for the fact that a greater fraction of the low education population is participating in the labor force over the course of the 99s. Because employment based insurance coverage rates are not measured as a fraction of workers, estimates of changes in these rates are not biased by changes in the composition of the work-

10 NATIONAL TAX JOURNAL force. Second, I break insurance coverage into four types within each employment status group. This allows me to separately identify changes in coverage rates for the primary source of private health insurance coverage employer based policies as well as changes in Medicaid coverage, which may be important for reasons discussed below. Following are the eight groups: Employed with Employer Provided Insurance (ESI); Employed and Uninsured; Employed with Medicaid; Employed with Other Insurance; Not Employed with Employer Provided Insurance (ESI); Not Employed and Uninsured; Not Employed with Medicaid; Not Employed with Other Insurance. An individual may have employer provided insurance through either a current or previous employer and the policy may be in his/her own name or that of a spouse. The Other Insurance categories include both private non group coverage and military plans (CHAMPUS, CHAMPVA). The Not Employed with Employer Provided Insurance category includes those who are not working and are either (i) covered by a spouse s employer based insurance, (ii) covered by a previous employer s insurance plan or (iii) not working but covered by a current employer s insurance plan (presumably on disability leave). I include Medicaid coverage as an explicit category for both workers and non workers because another important policy change occurred in the 99s that increased health insurance availability for low income workers state Medicaid expansions. In the early 99s, at the state level but by federal mandate, coverage under Medicaid expanded greatly, primarily to cover pregnant women and children not receiving AFDC at income levels as high as 85 percent of the federal poverty line. Although most states legislated their Medicaid expansions before 994, certain phased in expansions continued over the rest of the decade. In the baseline model, I estimate the following linear probability regression for each of the eight (k =...8) Employment/Insurance status groups: [] Status kit = α i + β X it + β 2 P st + γ YEAR + γ 2 CHILD it + γ 3 YEAR*CHILD it + ε it, where STATUS kit is a dummy variable for one of the status groups. CHILD it is a dummy variable for having at least one dependent child and YEAR is a vector of dummy variables for the year of observation 989, 99, 994, 996 and 998 (992 is omitted as a reference year). The γ 3 coefficients on the YEAR*CHILD interactions measure the differential time trend in health insurance coverage for individuals with children. Assuming that other relevant policies that affect parents have been explicitly controlled for, this is a measure of the impact of the EITC expansion on health insurance coverage. All models are estimated using individual fixed effects. The fixed effects control for two major things: () unobservable person specific taste for health insurance coverage and (2) other unobservable factors (such as attitudes toward childbearing) that affect the movements of individuals between control and treatment groups and may also be correlated with employment decisions. 5 The vector X contains time varying personal character- 5 Because individuals in these data move from control to treatment group during the panel as they have children, the role of the fixed effect is particularly important. The fixed effects allow for difference in difference analysis in panel data by controlling for unobservables that affect both insurance demand and childbearing. The only 674

11 Impact of the Earned Income Tax Credit on Health Insurance Coverage istics including age, age squared, education, marital status and family income. The other vector of control variables, P, measures the generosity of state level policies that could affect health insurance coverage. It includes a set of variables that capture Medicaid eligibility changes for children and pregnant women during the 99s. In order to construct these variables, I first calculate the fraction of children in a 99 national sample from the Survey of Income and Program Participation who would have been eligible for Medicaid by state s rules in year t. 6 This measure varies only by exogenous differences in the generosity of state rules over time and is a simplified version of a measure proposed by Currie and Gruber (996). I break this fraction eligible measure down into four quartiles (.6% eligible,.7 6.% eligible, % eligible, 26%+ eligible) and include dummy variables for these quartiles in the regression models. These quartile variables will not only capture any effect of Medicaid expansion on adult health insurance coverage, but allow for non linearity in the effect. The Medicaid variables are interacted with a CHILD dummy variable because the expansions in coverage were focused on children and parents. There are two control variables for welfare policy in the model. The first is the log of the maximum AFDC/Temporary Aid to Needy Families (TANF) benefit for a single mother with two children in a given state and year. I also include a dummy variable for whether or not a state had implemented a welfare program waiver in a given year. The welfare variables are also interacted with the CHILD variable. RESULTS Results for the baseline set of linear probability models are presented in Tables 4A and 4B. These results demonstrate that within the low skill population likely to be affected by the EITC expansions, there were differential trends in employment and health insurance coverage depending upon whether or not individuals had children. Overall, after 992, those with children were more likely to be employed with employment based insurance and less likely not to be employed with either employment based or Medicaid coverage. The net effect of the EITC, as measured by this child specific time trend, is a 3.8 percentage point increase in employer sponsored coverage between 992 and 998. It is made up of a 7.8 percentage point increase in the number of people employed with employer based insurance (the coefficient on CHILD*998 in Column of Table 4A) and a 4. percentage point decrease in the number of people not employed with employer based insurance (the coefficient on CHILD*998 in Column of Table 4B). 7 Applying population weights to these estimates, this represents a net gain of instance in which the fixed effect will not remove bias is when the unobservables that influence insurance demand and timing of childbearing are systematically correlated. However, in the most likely case, in which individuals have higher demand for health insurance and bear children later in life (a set of traits both correlated with higher incomes and education), the results presented here will actually be biased downward. Another concern is that the EITC itself might encourage childbearing; Baughman and Dickert Conlin (23) find that the increase in fertility caused by the EITC expansions is not likely to be large enough to bias the results of the model presented above. 6 This also includes children eligible for state Child Health Insurance Programs (CHIPs) in 998. CHIP programs, authorized by the Balanced Budget Act of 997, allow states to set up public health insurance programs for children that may be separate from Medicaid. In expanding public health insurance for low income children, some states chose to establish CHIP programs, while some simply worked within existing Medicaid programs. 7 An alternate measure of the program s effect would be based upon the averages of the Child*994, Child*996 and Child*998 coefficients. This average estimate would be a 3.5 percentage point increase in coverage. 675

12 NATIONAL TAX JOURNAL Child*989 TABLE 4A LINEAR PROBABILITY MODELS, FULL SAMPLE Employed with ESI.6 (3.92) 2 Employed and Uninsured.72 (.48) 3 Employed with Medicaid.336 (3.28) 4 Employed with Other Insurance.86 (4.58) Child*99.27 (.49).84 (.96).55 (.7).5 (.6) Child* (.97).4 (.48).7 (.2).5 (.8) Child* (.4).264 (.4).33 (.36).37 (.55) Child* (2.25).66 (.6).69 (.75).58 (.44) Child.856 (.6).26 (.4).42 (2.3).24 (3.2) 2 Children.647 (.26).28 (.3).367 (2.8).233 (3.5) 3 Children.368 (.78).278 (.42).45 (2.5).249 (3.2) 4+ Children.365 (.69).484 (.69).25 (.22).23 (3.6) Medicaid Quartile (3.29).58 (.89).286 (3.63).2447 (.7) Medicaid Quartile (3.36).48 (.39).29 (2.6).4 (.69) Medicaid Quartile 4.46 (.5).774 (.86).2 (2.29).923 (.49) MedicaidQuartile2*Child.38 (5.6).35 (.6).429 (4.84).5 (4.37) MedicaidQuartile3*Child.64 (2.2).6 (.43).276 (2.6).83 (3.56) MedicaidQuartile4*Child.49 (2.67).483 (.8).294 (2.68).923 (3.56) Ln Max AFDC/TANF.386 (.).26 (.5).74 (.97).433 (3.33) AFDC*Child.73 (.64).339 (.28).7 (.84).3 (.87) Welfare Waiver.82 (.35).62 (.76).47 (.76).6 (.6) Waiver*Child.34 (.5) (.4).4 (.37).39 (.45) Sample of individuals from NLSY with 2 or fewer years of schooling in 989. Absolute value of t statistics in parentheses. Huber White robust standard errors are used. All models include individual and year fixed effects as well as controls for age, age squared, < years education, years education, family income and marital status. Child*992 is the omitted interaction variable. Medicaid Quartiles are based upon the fraction of children in a national sample who would be eligible for Medicaid by state/year rules.

13 Impact of the Earned Income Tax Credit on Health Insurance Coverage Child*989 TABLE 4B LINEAR PROBABILITY MODELS, FULL SAMPLE Not Employed with ESI.489 (2.4) 2 Not Employed and Uninsured.9 (.3) 3 Not Employed with Medicaid.5 (.5) 4 Not Employed with Other Insurance.343 (2.78) Child*99.3 (.3).258 (2.7).3 (.37).52 (.3) Child* (.3).58 (.37).34 (3.72). (.22) Child* (2.29).34 (.7).47 (4.76).36 (.54) Child* (3.4).247 (.2).84 (5.5).72 (.) Child.477 (.49).8 (.4).582 (.65).349 (2.7) 2 Children.429 (.39).48 (.).874 (2.52).387 (2.28) 3 Children.269 (.92).8 (.5).984 (2.96).389 (2.32) 4+ Children.365 (.6).3 (.2).279 (3.32).35 (2.9) Medicaid Quartile 2.64 (.4).527 (2.3).2 (.2).4 (3.4) Medicaid Quartile 3.25 (.65).427 (.42).4 (.23).429 (3.2) Medicaid Quartile 4.77 (.42).47 (.44).47 (.24).47 (3.3) MedicaidQuartile2*Child.246 (.45).234 (.8).62 (.35).397 (3.25) MedicaidQuartile3*Child.277 (.2).88 (.24).4 (.63).49 (2.98) MedicaidQuartile4*Child.389 (.5).87 (.5).33 (.7).463 (2.85) Ln Max AFDC/TANF.89 (.82).26 (.96).44 (.82).94 (.4) AFDC*Child.273 (2.33).5 (.68).36 (.85).4 (.37) Welfare Waiver.38 (.5).46 (.3).59 (.6). (.22) Waiver*Child.3 (.).27 (.72) (.56).27 (.5) Sample of individuals from NLSY with 2 or fewer years of schooling in 989. Absolute value of t statistics in parentheses. Huber White robust standard errors are used. All models include individual & year fixed effects and controls for age, age squared, < years education, years education, family income & marital status. Child*992 is the omitted interaction variable. Medicaid Quartiles are based upon the fraction of children in a national sample who would be eligible for Medicaid by state/year rules.

14 NATIONAL TAX JOURNAL ,424 more individuals covered by an employer based health insurance policy after the EITC expansions. As discussed earlier, through an income effect, the EITC could also encourage coverage through other private individual health insurance policies. In the models for which estimates are given in Tables 4A and 4B, I do not focus on this type of coverage. Private non group coverage is included in the Other Insurance categories, along with military insurance, because few people have these types of policies. I do not find any trend in Other Insurance coverage that is specific to those with children. The coefficients on policy control variables in the models indicate that, as one might expect, Medicaid expansions for children had a significant effect on health insurance coverage for low skilled workers, while welfare reform did not. Medicaid coverage (measured as the fraction of children eligible for Medicaid in a given state and year) is divided into quartiles and the lowest quartile is the omitted category. The significant negative effects of higher levels of Medicaid coverage on employer sponsored insurance in Column of Table 4A indicate that crowding out of private insurance by public insurance is more of an issue the higher the income threshold for eligibility is raised. As the Medicaid results in the third columns of Tables 4A and 4B illustrate, more generous Medicaid expansions were likely to increase Medicaid coverage for low skill working parents, but not low skill parents who did not work. Finally, with the exception of a statistically significant positive effect of AFDC/TANF benefits on the probability of being unemployed with ESI in Column of Table 4B that is admittedly difficult to explain, welfare programs (as measured by waivers and benefit levels) do not seem to have had a significant impact on health insurance coverage patterns. Before concluding that the EITC did increase employer sponsored health insurance coverage for its target population, it is important to make sure that these results capture changes in employer sponsored health insurance coverage that came about due to the EITC expansion, rather than another time trend that was specific to workers with children. Although I have controlled explicitly for the two most likely factors to have differentially affected low skill families welfare reform and Medicaid expansions there may be other determinants of insurance demand that changed significantly for those with children during the 99s. The most likely possibility is that the cost of family coverage rose more quickly than the cost of employee only coverage during this period (GAO, 997a). If this were an important factor for the low skill population in this sample, it would bias the estimated net increase in employer sponsored coverage downward. In order to check the sensitivity of these results to the choice of a baseline specification that relies upon differential trends in coverage for individuals with and without children, I estimate two additional specifications. Both of these specifications rely on variation in EITC benefits within a sample that contains only individuals with children, and so their estimates will not be biased by any child specific national trend in insurance coverage for which I am not able to include adequate controls. In the first column of Table 5, I present results for a simplified linear probability model in which the dependent variable is any employer sponsored health insurance coverage (ESI). This model exploits the growth in incremental EITC benefit offered to families with more than one child during the 99s and tests to see if adults with two or more children are more likely to gain coverage than adults with only one child. [2] ESI it = α i + β X it + β 2 P st + γ YEAR + γ 2 YEAR*2 + KIDS it + γ KIDS + ε it.

15 Impact of the Earned Income Tax Credit on Health Insurance Coverage TABLE 5 SPECIFICATION CHECKS DEPENDENT VARIABLE = ESI COVERAGE SAMPLE OF INDIVIDUALS WITH CHILDREN () Number of Children (2) State EITC 2+ Children 2+Child*989 2+Child*99 2+Child*994 2+Child*996 2+Child*998 State EITC (,s) 2 Children 3+ Children Medicaid Quartile 2 Medicaid Quartile 3 Medicaid Quartile 4 Max AFDC/TANF Welfare Waiver.568 (.43).4 (.6).6 (.5).27 (.2).26 (.49).57 (2.22).354 (.8).9 (.75).352 (7.7).236 (5.4).85 (4.78).243 (.77).63 (.44).6 (2.25).8 (.87).254 (.6).387 (7.44).25 (5.26).9 (4.77).23 (.66).6 (.43) Sample of individuals from NLSY with children and with 2 or fewer years of schooling in 989. Absolute value of t statistics in parentheses. Huber White robust standard errors, clustered at the state level, are used. Model includes individual and year fixed effects, as well as controls for age, age squared, < years education, years education, family income and marital status. Medicaid Quartiles are based upon the fraction of children in a national sample who would be eligible for Medicaid by state/year rules. The coefficient on dummy variable 2+KIDS is negative and marginally significant, indicating that, if anything, adults with more children were less likely to have employer sponsored health insurance coverage on average between 989 and 998. However, the coefficient on 998*2+KIDS is positive and statistically significant and indicates that individuals with two or more children were significantly more likely than individuals with only one child to gain insurance coverage between 992 and 998. This is consistent with the effect of an incremental EITC benefit for additional children. In the second column of Table 5, I present results for a model that exploits a completely different source of variation in EITC benefits for individuals with children. By 998, ten states offered Earned Income Credits to supplement the federal credit. Most of the growth in these credits occurred during the 99s; in 989 only four states offered credits, and all states that initially offered them changed the size of the maximum benefit between 989 and 998. The model estimated is a linear probability model in which the dependent variable is employer sponsored health insurance coverage. The variable used to measure EITC benefits is the maximum real federal + state benefit available to a family with two children in an individual s state of residence. [3] ESI it = α i + α t + β X it + β 2 P st + γ STATEBEN st + ε it. Because fixed effects are used, identification of the effect of the EITC in this model depends upon variation in state credit amounts within states over time. The coefficient on EITC in Column (2) of Table 5 indicates that a $, increase in the maximum combined state and federal credit is associated with a statistically significant six percentage point increase in the fraction of individuals covered by employer sponsored insurance. Because the results reported in the baseline model appear to be robust to these specification checks, I perform one final test of their plausibility. I calculate 679

16 NATIONAL TAX JOURNAL the fraction of the total dollar increase in EITC benefits that my estimates imply would have been directed toward health insurance coverage. Using Gruber s (2) estimate that an employment based insurance policy costs $,86 per year, 375,424 newly insured workers would have resulted in a $698 million increase in health insurance spending. At the same time, between 992 and 998 total federal spending on the EITC increased by $9.3 billion. This implies that 3.6 percent of the EITC spending increase in the mid 99s was directed to health insurance coverage. 8 This figure seems quite reasonable, especially given that Smeeding et al. (2) find that EITC recipients in their survey report payment of any medical and health related costs as a relatively infrequent use of refunded credit payments. IDENTIFYING THE SOURCE OF INCREASED COVERAGE For the purposes of interpreting the baseline results, the final question that remains to be answered is whether the impact of the EITC on private health insurance coverage was a direct one operating either through the income and substitution effects of the credit or whether it was simply a result of EITC expansions increasing labor force participation. The panel nature of the NLSY provides some advantages in trying to answer this question. One strategy is to look only at individuals who were working at the beginning of the panel, and were not drawn into the labor force during 99s, and to see whether employer based health insurance coverage rates changed for them. I estimated a set of models identical to ones reported in Tables 4A, 4B and 5, but estimated only on individuals who reported working in In the baseline federal EITC models, the coefficients on the CHILD*Year interactions are positive in 994, 996 and 998, but the estimates are not significant at conventional levels. State EITCs had a significant and positive effect on the probability of employer sponsored insurance (coefficient =.867, s.e. =.37). These results provide some evidence that a portion of the increase in coverage associated with the EITC expansions came about as a result of the income and substitution effects of the credit, rather than simply increased employment. Another way in which to get a sense of what portion of the increase in ESI coverage is attributable to employment increases is to compare these results to the employment increases suggested by previous studies in the EITC labor supply literature. Eissa and Liebman (996) estimate a 2.8 percentage point increase in the employment of single mothers between the 984 to 986 and 988 to 99 periods. Meyer and Rosenbaum (2) estimate a six percentage point overall increase in the employment of single mothers between 984 and 996, 6 percent (or 3.6 percentage points) of which they attribute to EITC expansions. Using Meyer and Rosenbaum s (2) estimate that 73.2 percent of single mothers worked in 984, the estimates given by these two studies suggest increases in labor force participation of single mothers of 3.8 percent and 4.9 percent, respectively. The Eissa and Liebman (996) and Meyer and Rosenbaum (2) estimates are measured against increases of $22 (86 percent) and $724 (282 percent) in the average real value of the credit per family. During the period I consider This is a lower bound estimate if the EITC also increased health insurance coverage of individuals outside the age cohort represented in my data. 9 Full results of these models have been omitted for brevity, but are available from the author upon request. This is their estimate of changes in current week employment, which is the measure I use in the NLSY. Changes in the average credit per family are calculated from Table 3 4 in the 2 Green Book. 68

17 Impact of the Earned Income Tax Credit on Health Insurance Coverage to 998 the average real value of the credit increased by $449 (49 percent), and so existing labor force participation estimates imply an increase of between.85 and 2.2 percent over the mid 99s. 2 The 57 percent of single mothers in my sample who work in 992 represent.4 million women, and so we would estimate that between 9,69 and 25,8 additional single women in this age cohort went to work because of the EITC during the period under consideration. Obviously, not all of the women who entered the labor force would have been offered health insurance by their employers. I estimate using the 997 Community Tracking Survey that 6 percent of working single mothers with a high school education or less are offered health insurance on the job. Assuming that all women who are offered health insurance at work sign up for coverage and multiplying this by the labor force participation changes, I estimate that between 5,9 and 5,299 single mothers automatically gained employer sponsored health insurance when they entered the labor force because of the EITC. This represents at most four percent of the total increase in employer sponsored coverage between 992 and 998. It seems clear that a substantial portion of the increase in employer sponsored health insurance coverage is a result of the design of the EITC, and not just its employment effects. The model in Table 6 presents further evidence as to whether the credit s substitution effect affected coverage rates. 3 As has been discussed earlier in the paper, the way in which the EITC phases in and then phases out as earnings increase creates different tax incentives for employer sponsored health coverage at different earnings levels. Specifically, individuals in the phaseout range face incentives to demand health insurance in lieu of additional wages, while individuals in the phase in range face incentives to demand wages over benefits. By estimating a model that includes an additional set of interactions for EITC eligibility range, I can exploit variation in earned income to test whether health insurance demand responds to the specific tax subsidy generated by the credit. The linear probability model for which results Maximum Federal EITC ($,s) TABLE 6 FEDERAL CREDIT INTERACTIONS Dependent Variable Is Employer Sponsored Coverage () (2).79 (.49).227 (.92) Phase Out*Maximum Benefit Flat Range* Maximum Benefit. (.32).437 (3.85) Phase In Range* Maximum Benefit.43 (5.57) Sample of individuals from NLSY with 2 or fewer years of schooling in 989. Absolute value of t statistics in parentheses. Huber White robust standard errors are used. Models include individual and year fixed effects, as well as controls for Mediciad, AFDC/TANF, age, age squared, < years education, years education, family income, family income squared and marital status. 2 These may be lower bound estimates if awareness of and participation in the program increased during the 99s. 3 In a separate analysis (results not reported, but available from the author upon request), I test to see whether take up of insurance among individuals offered coverage by an employer increased in response to the EITC. This would be indicative of an income effect, but the results of these models are not statistically significant. 68