The Impact of the Earned Income Tax Credit and Welfare Reform on Work Entry and Exit Yucong Jiao University of Illinois at Chicago.

Transcription

1 1 The Impact of the Earned Income Tax Credit and Welfare Reform on Work Entry and Exit Yucong Jiao University of Illinois at Chicago Abstract This paper examines whether the employment effects associated with the Earned Income Tax Credit (EITC) and welfare reforms are due to increased work entry among non-workers or decreased work exit among workers. Differentiating the entry and exit effects is important for improving employment policies and for understanding the low-skilled labor market more generally. My empirical analysis focuses on the 1993 EITC expansion and 1992/1996 welfare reforms. I find that the EITC increased employment of unmarried mothers by six percentage points. Approximately half of the employment increase can be attributed to fewer labor market exits and half to additional labor market entry. In contrast, welfare reform (AFDC waivers/tanf) increased employment by 5.6 percentage points, and the employment effect was entirely through increased entry. The findings raise questions about the efficacy of the EITC at inducing labor market entry among the hard-to-employ. On the other hand, the direct work requirements such as those associated with welfare reforms are effective at inducing greater entry into the labor market. 1.! Introduction During the mid- to late-1990s, work incentives for unmarried mothers were enhanced due to the expansion of the Earned Income Tax Credit (EITC) and welfare reform. The EITC is the largest cash transfer program in the United States and has the dual goal of lowering the tax burden and raising employment of low-income families. It provides a wage subsidy to lower-income families with children in the form of refundable tax. The program was expanded considerably during , offering a credit three-to-five times as large as previous subsidies for unmarried working mothers. During the same period, cash transfers to single families with children through Aid to Families with Dependent Children (AFDC) was restructured to include time limits and work requirements. A large body of evidence shows that the EITC and welfare reform succeeded at raising employment of unmarried mothers during the mid- to late-1990s (Meyer and Rosenbaum 2001, Hotz and Scholz 2001, 2006,! Eissa and Hoynes 2006, Hoynes and Patel 2015, Ellwood 2000, MacKernan et al 2000, Grogger 2004, 2003). This paper extends this literature by examining whether the employment effects from the EITC and welfare reform are due to increased work entry among non-workers or decreased work exit among workers. In the traditional, static labor supply framework, the EITC raises employment only by inducing some people to enter the labor market when they otherwise would not have. However, the benefits of the EITC may largely be unknown to those out of the labor force and thus have little effect on them. On the other hand, workers may have better knowledge of 1

2 2 EITC benefits they have received the tax credit and know the effective increase in the wage (Mead 2014). Therefore, the EITC could affect the decisions of people who are already working by giving them an increased incentive not to exit the labor market following, for example, a plant closing or the birth of a child. Identifying the entry and exit effects of the EITC is important for improving employment policies and for understanding the low-skilled labor market more generally. The entry effect is likely to be smaller (and the exit effect larger) if the hard-to-employ population are less likely to know about the EITC or if they have sufficiently high reservation wages so that wage subsidies have little effect on their decisions. Unlike the EITC, welfare reform should increase employment primarily from increased work entry due to the time limits and work requirements. These restrictions are transparent to welfare users and explicitly promote welfare-to-work transition. Indeed, welfare reform administratively forced even those with high reservation wages into the market; welfare reform did not rely on a labor supply response to higher wages like the EITC. To measure the effect of the EITC and welfare reform on work entry and exit, I use data from the Current Population Survey (CPS) and the Survey of Income Program Participation (SIPP) and a difference-in-difference (DD) research design that exploits variation in the generosity of the EITC program and welfare reform. In 1993, the EITC credit was significantly increased and the wage subsidy differed by family size. In the early 1990s, states began to alter AFDC program rules such as imposing time limits and work requirements, and in 1996, national reform along similar lines took place with the creation of TANF. I find that employment went up by six percentage points for unmarried mothers relative to unmarried childless women after the major EITC expansion. Approximately half of the employment increase can be attributed to fewer labor market exits and half to additional labor market entry. Notably, all of the reduced exit from the labor market is among women whose family income is sufficiently low that they are indeed eligible for the EITC, which provides evidence of the validity of the research design. Results also suggest that for a family with children living in a state that adopted AFDC waiver or TANF, employment increased by 5.6 percentage points. Consistent with prior expectation, I find that the employment effect of welfare reform was entirely driven by increased entry. These findings raise questions about the best way to increase employment among the hard to employ. The less strong effect of the EITC on work entry (vis-à-vis welfare reform) suggest that the hard-to-employ may either lack awareness of the EITC or maintain a high value of the non-market time that the wage subsidy fails to affect them. To motivate those largely out of the labor force, administrative mandates such as those implemented in the welfare reform may be more effective. 2.! The Earned Income Tax Credit and Welfare Reforms 2

3 3 2a. The Earned Income Tax Credit The EITC is a refundable tax credit for low- and moderate-income working people. It encourages work by offsetting payroll and income taxes. According to the 2015 IRS Statistics, almost 28 million tax filers received over $67 billion in EITC benefits for the tax year 2014, and the average amount of the EITC paid out was more than $2,400. The EITC is now the largest antipoverty program in the US. The amount of EITC benefits varies by its recipient s income and number of children. There are three benefit regions that depend on the family earned income: phase-in, flat and phase-out region. In the phase-in region, the EITC credit raises the wage rate, which means the more dollars a family earns, the larger the subsidy (tax refund they can claim). In the flat region, the credit stays fixed as the earned income increases. Finally, in the phase-out region, the credit decreases with each additional dollar earned until it becomes zero. Although similar in its general shape, the EITC benefit schedule differs by number of children and filing status. There have been several major expansions of the EITC since its start in Figure 1 provides an illustration of the program history by plotting the maximum federal EITC amount over time, and separately for the number of children. As seen from the figure, it was expanded in 1986, 1990, 1993, 2001 and The 1993 expansion (takes effect from 1994 to 1996) is the most dramatic change. As a result, the 1993 expansion increases the maximum benefit for families with children relative to no children, and also for families with two children relative to one child. This study exploits different expansions over time while focusing on the 1993 expansion. 2b. Welfare Reforms During the early 1990s, states undertook a series of reform efforts. Since 1992, states started to impose restrictions, such as time limits and work requirements on the AFDC participation through federally-approved waivers. The goal of these waivers were to reduce welfare participation and encourage welfare users to enter the workforce. In 1996, the TANF program was introduced nationwide and had many features of earlier state reforms. These changes provide work incentive for the same population affected by the EITC unmarried mothers with children. 3.! Work Entry and Exit In a traditional, static labor supply framework, the EITC increases wage rates and does so differentially by family income as income moves higher first through the phase-in region, then the flat region, and finally to the phase-out region. The credit will expand family budget constraint and alters the value of work. Figure 2a illustrates the budget constraints for a person with a $10 hourly wage and work 52 weeks a year without the EITC, with the EITC in 1993 (pre-expansion) and with the credit in 1997 (post-expansion). The picture predicts 3

4 4 that a person with a utility curve as shown in Figure 2a will enter the labor market voluntarily in response to the enlarged wage subsidy, enjoying a greater utility than before. Figure 2b demonstrates the budget constraints before and after welfare reform, specifically with the introduction of work requirements. It assumes 20 hours of weekly work, as imposed by the TANF for single mothers with children under 6. The actual average monthly benefit provided by the TANF is $490, with the tax rate on wage earnings ranging widely across states. Here I assume the monthly benefit is $500 with a 50% tax in year 1991 (pre-reform) and a lower tax rate at 33% in 1997 (post-reform). In addition, I assume all the 20 hours of work are paid. It is clear as to the effect of a 20-hour work requirement: an individual will enter the labor force reluctantly due to the administrative mandate and ends up with a lower utility than before, but they will enter since the utility derived from a welfare program with work requirements is still larger than no welfare at all. The static labor supply model illustrates the employment increase due to the EITC expansion and welfare reform. But it is the entry and exit behaviors that are cause of the change in employment and it is these behaviors that are arguably more relevant to understanding the causal effects of these programs. To help differentiate the entry and exit effects by the EITC and welfare reform, one can assume there are three distinct groups with different preferences for work/leisure: one that is largely out of the labor market; one that goes into and out of the market; and a third that is always in the market. The EITC literature attributes the entire employment effect to the first group through increased work entry. However, this conclusion relies on two conditions:1) those largely out of the labor force are aware of the EITC, and 2) their reservation wage is sufficiently low that the wage subsidy can actually affect them. It may be difficult to satisfy the first condition if the first group remain out of the labor force for a long time and never file a tax return. In fact, the Internal Revenue Service estimates about 20 percent of eligible taxpayers do not claim the benefit each year. As to why people are leaving billions of dollars uncollected, one study showed that the take-up rate is 35 percent for eligible households not required to file a tax return as compared to approximately 90 percent for those required to file (Blumenthal et al 2005). Also, some studies suggest large demographic differences in the EITC awareness (Phillips 2001, Magg 2005), and that low program awareness and/or misconstrued program incentives are major barriers for lowincome families to claim the credit (Liebman and Luttmer 2011, Bhargava and Manoli 2015). This research raises doubt about the awareness of the EITC among the first group. In addition, the second condition can be violated if people largely out of the labor force retain high reservation wage that the EITC-induced wage subsidy fails to motivate them. Thus, due to the possible disadvantage in gaining EITC information or high value of the non-market time, the EITC may have no effect on the first group. On the other hand, the EITC may reduce exits among the second group because these people may have better knowledge of EITC benefits they have received the tax benefit and 4

5 5 know the effective increase in the wage. Those with intermittent labor force attachment may reduce exits through different channels: One possible mechanism is that the EITC prevents them from leaving the labor market when their value of non-market time increases, such as the birth of a child or a family member becomes sick. In addition, laid-off workers may experience shorter unemployment spells, resulting in decreased exit when exit is measured at the annual level. For example, before the EITC, a laid-off worker may spend more than one year finding a job, consequently classified as an exit, but now with the EITC, he may settle for a lower-pay job within one year, thus considered as non-exit. In contrast to the EITC, TANF (welfare reform) is likely to exert a major impact on the group of hard-to-employ who are usually out of the labor market. Further, welfare reform may have a relatively small influence on the second because of its administrative mandates featuring work first. As explicitly stated in one of its four major goals, the TANF aims at end the dependence of needy parents on government benefits by promoting job preparation, work, and marriage. (U.S Department of Health and Human Services). Although it is reasonable to assume the deterrence effect of TANF in preventing those marginally attached to the labor force from quitting jobs, a more direct impact should be among welfare usersthose who are currently out of the labor force. As one study finds welfare exits were most significantly affected by the economy and federal welfare reform whereas welfare entries were most significantly affected by the economy, the decline in welfare benefits, and the expansion of the EITC. (Gorgger 2004) This finding not only supports that the welfare reform affects employment mainly through increased work entry but also provides evidence of the EITC s role in keeping people away from welfare and thus reducing work exits. 4a. Decomposition of Employment into Entry and Exit As noted, there is evidence that the EITC and welfare reform increased employment. In this section I show that any change in employment can be decomposed into changes in the entry and exit rate to employment. This is useful because it is changes in behavior related to entry and exit rates that cause a change in employment. Therefore, it is important to understand how policy changes such as EITC and welfare reform affect these entry and exit behavior if a full understanding of the effect of these policies is to be gained. I begin by defining unconditional employment rate at time t as probability of working in year t: Employment * = P(E * = 1) The conditional entry rate at time t is the probability of working in year t, conditional on not working in year t-1. It is the entry rate among non-workers: Entry * = P(E * = 1 E *23 = 0) 5

6 6 Similarly, the conditional exit rate at time t is the probability of not working in year t, conditional on working in year t-1. It is the exit rate among workers: Exit * = P(E * = 0 E *23 = 1) where E * is an indicator for whether being employed in year t. Next, I show that Employment * can be decomposed into three terms: employment rate at time t-1, unconditional entry at time t, and unconditional exit at time t: P(E * = 1) = P(E *23 = 1) + P E * = 1 E *23 = 0 P E *23 = 0 P E * = 0 E *23 = 1 P E *23 = 1 where the second term on the right is the unconditional entry and the third term is the unconditional exit. Take the derivative with respect to the EITC at time t, the change in employment is: Employment * EITC * = 1 α *23 Entry * EITC * α *23 Exit * > EITC * >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>(1) where α *23 is Employment *23. Equation (1) is the basis of my empirical models below. It suggests that the effect of the EITC on employment is a weighted average of entry and exit effects, with the weight being the previous employment rate. The same framework applies to the welfare reform in which case the EITC is replaced by TANF. 4b. Simulation of Entry and Exit Effects Associated with the policy change To provide an illustration of how policy-induced changes in employment are related to entry and exit, I conduct two simulation exercises focusing on the 1993 EITC expansion. In the first simulation, I ask what the upper bound of entry and exit effects are based on Equation (1). A preview of the results shows that the employment effect associated with the EITC (?@ABCDEAFG* H?@IJK H ) is 0.06, and employment in pre-policy period (α *23 ) is Assuming that all of the effects of the EITC are through changes in exit rate (?@G*LE H?@IJK H = 0). It is straight-forward to calculate the upper bound of the exit effect (?@MN* H>?@IJK H >) which in this case is 0.06/0.726= On the other hand, if all the effects are through entry (?@MN* H>?@IJK H = 0), the maximum increase in entry is 0.06 / ( ) =0.22. The pre-policy means of entry and exit are and respectively. The maximum entry and exit effects are very large relative 6

7 7 to their baseline means. The entry effect is much bigger than the exit because work entries can only be increased among the unemployed in year t-1, whereas work exits can be reduced among the entire employed population. Given the employed population is approximately three times larger than the unemployed population, it makes sense to arrive at an upper bound of entry effects 3-4 times bigger than exit effects. In the second simulation, I graph the employment effects by year, assuming a permanent increase in entry or decrease in exit influenced by the 1993 EITC expansion. Figure 3a presumes that all effects are through exits (a permanent decrease in exit since 1994; no change in entry). Figure 3b makes the opposite assumption (a permanent increase in entry and no variation in exit). Figure 3c plots the simulated employment rate under both conditions. The 1993 and 2003 employment rate are true to the data. It can be easily seen the same employment effects can be achieved from both a permanent decrease in exit and increase in entry. This simulation exercise offers another instructive implication-if entry and exit follows a step function of time, the employment rate will not be a step function but increase with time at a declining rate. The function may be polynomial, but the bottom line is the effect of the EITC on employment is time-varying. These simulated figures are important in guiding the empirical specifications in Section 6. While TANF may impact entry and exit differently than the EITC expansion, the same continuous increase of employment will result from either a permanent dip in exit rate or rise in entry rate in response to welfare reform. 5. Data 5a. Full Sample The data used in this analysis come from the March Current Population Survey (CPS) and the Survey of Program Participation (SIPP). Both data sets possess different strengths. The March CPS is the traditional data employed in the literature but not designed for longitudinal studies. The ideal data to perform entry/exit analysis would be a national longitudinal panel, and that is the main purpose of using the SIPP. The CPS is a nationally representative survey of approximately 60,000 households, providing monthly reports on employment situations. In the March interviews, detailed retrospective information including work status, weeks worked, wages and income during the previous year are collected. The SIPP is a household-based survey designed as a series of national panels. In this analysis, I use data from the , ,1996 and 2001 panels of the SIPP. Each of these panels is a longitudinal survey which lasts for two to four years. Interviews take place every four months, in which respondents report monthly employment status for each of the past four months. This feature allows me to measure employment more precisely. 7

8 8 The March CPS data are from , providing information on the years , while the SIPP data are from , revealing information in the same year. Following Meyer and Rosenbaum 1999 a, I limit the sample to single women (widowed, divorced, and never married) without a college education, who are 19 to 44 years of age, and not in school. In addition, women who were ill or disabled during the previous year, with positive earned income but zero hours of work, or with zero hours of work but positive income are also excluded. The final sample has 122,602 for the March CPS and 53,974 for the SIPP. The employment status is based on whether a woman ever worked in the previous year. Work status last year is preferred to weekly or monthly work status because the majority of EITC recipients receive the credit annually, therefore annual employment should be more relevant regarding the EITC expansion. 5b. Matched Sample To construct work flow measures, at least two years of data are needed. Although the CPS are not initially designed for linking people across years, the respondents are interviewed four months in year t-1 and then the same four months in year t. This feature will allow me to match individuals across two years. In a particular year, the best matching case is that half the people are linked to the previous year, with the other half to the next year. This is what Madrian and Lars John Lefgren (2000) refers to as a naïve match. This procedure includes some false matches and misses potential matches due to errors in the CPS identifiers. According to Madrian and Lars John Lefgren, some common causes of matching failure include sample non-responses, mortality, migration and recording errors. Finally, I drop the false positives by restricting the matched person to have the same race and gender in two consecutive years, with age difference equal to one. I find 71% of people from year t-1, who have potential matches in the next year, can be linked to year t during the time period , which is the same finding as in Madrian and Lars John Lefgren The analysis in this paper restricts sample to college or less educated single women, aged during and find a final match rate of 45.7%. Appendix Table A1 reports the match rates of each and all years. The matching process with the SIPP data is more straight-forward thanks to its longitudinal feature. First, I aggregate the monthly measures to the annual level since the majority of EITC recipients receive the credit annually. One problem arises from the aggregating procedure: if a survey starts in the latter half of a year-such as, October 1984, I will only observe a person three months in To measure annual employment using three-month information may introduce non-random errors to entry and exit rates. For example, if a person didn t work from October to December in 1984, but ever worked in other months in 1984 and 1985, I would miss-classify him as an entrant, whereas he is actually a stayer in the workforce. Due to this issue, I delete those three months of observations in 1984, and keep data for the following years. Consequently, year 1984,1988,1989,1995,1999 and 2000 cannot be matched since these are the starting or ending of a panel. Then, I create time t-1 and t variables if a person is in the survey for more than two years to prepare for the entry/exit analysis. 8

9 9 5c. The Maximum EITC, Welfare, and State Unemployment Data The maximum amount of the EITC is constructed using policy rules varying by year and number of qualified children (Tax Policy Center 2016). It has been deflated using CPIU. The EITC eligibility is determined if a working family is associated with zero federal EITC. The actual amount of the federal EITC is calculated using the TAXSIM program provided by the NBER (Feenberg and Coutts 1993). Information regarding family income, filing status, number of qualified children, etc. are collected from the March CPS. Two welfare reform variables are gathered to reflect the policy change during The indicator whether the state a family resides in receives a AFDC waiver or adopts the TANF is constructed using data from U.S. Department of Health & Human Services ( ). The variable maximum amount of welfare is provided by the University of Kentucky Center for Poverty Research ( (accessed 2016)) and is also deflated by CPIU. It reflects the total amount of AFDC/TANF and food stamp that a single parent family with children can obtain, and varies by state, year and family size. Annual state unemployment for men aged 45 to 65 is calculated directly from the March CPS. 5d. Summary Statistics of Full and Matched Sample A comparison of summary statistics in the full and matched samples are shown in Appendix Table A2. In the March CPS, the matched sample are a little older; have relatively lower fraction of high school dropouts, Hispanic population and never married women; and slightly fewer EITC benefits than the full sample. The difference between the full and matched samples is likely because those who have low socio-economic status are more likely to contribute to matching failures than those of high status. To assess whether the matched sample is representative in terms of findings, I will present the effects of the EITC and welfare reform for both full sample and matched samples. The SIPP data is balanced across all variables except for the maximum welfare benefit. The matched sample on average are associated with $90 more in the maximum welfare than the full sample. Comparing across the CPS and SIPP samples, the SIPP overall have more children, therefore enjoying higher EITC and welfare benefits than the March CPS. Other characteristics are fairly similar across both data sets. 6. Methods In the EITC and welfare reform literatures, a difference in difference approach is extensively used (Eissa and Liebman 1996, Meyer and Rosenbaum 2001, Hotz and Scholz 2006, Eissa and Hoynes 2006, Hoyens and Patel 2015). The two policy changes differentially 9

10 10 affect low-income families by number of children, therefore creating natural treatment and control groups. Before going into details about methodology, I first present the general trends of annual employment rate for unmarried women with zero, one, and two or more children in Figure 4. As seen from Figure 4, three patterns are noticeable: a) employment for unmarried women with children increased relative to those without children after 1993; b) women with two or more children experienced a sharper rise than those with only one child; and c) the increase in employment is not a one time jump but rather a gradual climb. The third feature is especially noteworthy since it is consistent with the previous simulation as how employment will change with changes in entry and exit rates. This feature of the predicted pattern of employment in response to a change in the EITC has not been previously noted. Motivated by Figure 4, I present three DD models in the following section, utilizing the 1993 expansion of the tax credit across family size, and welfare reform. The identification assumption is that the comparison group will fully capture the time trends of the treatment group in absence of the EITC policy change. Two sets of treatment and control groups are explored: a) single mothers with children vs. single women without children, and b) single mothers with two or more children vs. single mothers with only one child. The 1993 EITC expansion not only affected single mothers with children differently than single childless women but also increased the maximum credit for two children families relative to one child families by about $400. The other advantage of using 1 vs. 2+ children as control and treatment group is that these women are less likely to be affected by welfare reforms compared with 0 vs. 1+ children comparison. Therefore, one would expect to find a smaller entry and exit effects with 1 vs. 2+ children comparison, which is entirely driven by the EITC instead of welfare changes. The effect of the waivers/tanf differs from the EITC expansion in the way that it affects mothers with children. Welfare reform doesn t have differential impacts on single mothers with one child versus those with two or more children all mothers with children are affected, although perhaps those with more children are more affected by time limits. 1 Therefore, welfare reform will affect unmarried mothers with children relative to those without children, but for women with children, welfare reform is likely to have similar effects. I first present an event-history specification of the DD approach that allows the treatment group to have a different year effect in all years. Below is the specification of the model: Y N* = α + β * (Year * Treat N ) + Year * + Treat N + δx N* + ε N* > (2) 1 A mother with more children may expect to be on program longer because of the age ranges of her child. Therefore, the time limit may be more binding. 10

11 11 where Y N* includes three measures: whether worked in year t, whether exited from year t-1 to t, and whether entered from year t-1 into t. Specifically, exit is defined as the exit rate among workers, and entry is defined as the entry rate among non-workers. Year * are tax year dummies, which control for the overall time trends in unmarried female employment. Treat N >is an indicator if a woman is in the treatment group. X N* are controls of demographic and business cycle factors which includes education, race, ethnicity, marital status, and age group indicators; state fixed effects; state unemployment rate for men aged 45 to 65; and the interaction between Treat N >and unemployment rate; education; marital status. β * are the DD estimators, the outcome difference between treatment and control groups. These are key parameters indicating the overall impact of the EITC expansion and welfare reforms on unmarried mothers. Then, estimates of β are plotted and any trends away from zero in the pre-period may indicate unobserved differences in the treatment and controls which were not adequately controlled for. The event-history specification provides a direct examination of the parallel trend assumption in the DD estimation. However, given the moderate sample size of both data sets, allowing separate year effects results in somewhat noisy estimates. This is especially true for the entry model since it is defined among non-workers, which is about one fourth of the entire population. Therefore, I turn to a pre-post DD model, which summarizes the relative difference between treatment and control before and after Unlike the event-history specification, it groups all the years before and after the policy change into four data points and thus offers a less noisy estimate. The model is specified below: Y N* = α + β(post93 Treat N ) + γ(waivertanf * Treat N ) + ηmaxwelfare * + Year * + Treat N + δx N* + ε N* > (3a) Where Y N* include same outcome variables as in the multi-period DD model. The coefficient of the interaction between post93 and treatment indicator delivers the treatment effect of the 1993 EITC expansion. X N* are same sets of demographic and business cycle controls as in Equation (2). In addition to the EITC measure, two more variables separating the welfare impacts are included: the WaiverTANF dummy indicating whether the state that a family resides in has adopted the AFDC waivers or switched to the TANF, and the MaxWelfare summing up the total amount of annual maximum AFDC/TANF along with the Food Stamp benefits that a single family is eligible to claim. The WaiverTANF coefficient represents effects of welfare reforms on single mothers residing in affected states relative to single women unaffected by welfare reforms. I expect, for unmarried mothers with children relative to those without, a positive effect of the WaiverTANF on employment, big positive entry effect, and little or no exit effect; and for unmarried mothers with two or more children relative to those with only one child, a zero γ. The effect from the MaxWelfare are somewhat the opposite-a negative effect on employment and entry; and positive on exit. 11

12 12 While Equation (3a) describes a standard DD model with a one-time treatment effect, it may not be the best specification to describe Figure 2. What Figure 2 shows is not a onetime increase in employment but a time-varying treatment effect since Therefore, I add two more terms to Equation (3a) :Post93 Treat N Time>and>Time Treat N. Now Equation (3a) becomes: Y N* = α + θ>(post93 Treat N Time)> + ϕ>(time Treat N ) + >β(post93 Treat N ) + γ(waivertanf * Treat N ) + ηmaxwelfare * + Time * + Treat N + δx N* + ε N* > (3b) where the coefficient of the triple interaction term θ captures the treatment effects that vary with linear time trend. β picks up the level (time-invariant) treatment effect. The combination of those two terms provide the total effects of the EITC expansion. ϕ controls for the linear time trend of the treatment group. In general, Equation (3b) is a more generalized version of (3a), in which a zero θ suggests the treatment effect does not vary linearly with time. As in Equation (3a), there are three outcome variables: employment, entry and exit. Motivated by the previous simulation graphs, I expect a positive θ for the employment equation, but a zero θ for the entry and exit equation. The level effect β should be zero for employment, positive for entry, and negative for exit. Since there are multiple EITC expansions over time and across family size, I further replace the Post93*Year * by an intensity treatment variable, MaxEITC. The MaxEITC is the annual maximum amount of the EITC a family can receive contingent on tax year and number of children. This parameterized model is specified as follows: Y N* = α + βmaxeitc f* + γwaivertanf f* + ηmaxwelfare * + Year * + δx N* + ε N* >>>>>>(4) Where Y N* include same outcome variables as in the non-parameterized DD model. MaxEITC f* is a single policy variable that summarizes changes in the EITC by family size c and year t. The benefit of using MaxEITC f* is that it fully captures variation in the policy over multiple years and it is not endogenously determined by labor supply decision. 7. Results Figure 5 shows the event-history estimates of the employment for unmarried mothers relative to childless women (comparison 1). Both data from the March CPS and SIPP consistently display zero effects in the pre-treatment period, and a treatment effect increasing with time in the post period. This pattern validates the identification assumption and is consistent with the simulation figure of employment rates. Figure 6 plots the same estimates but for unmarried mothers with two or more children relative to those with one child (comparison 2). Both data sets predict a marginally significant treatment effect after 1994, and a marginally significant pre-trend. It indicates 12

13 13 women with one child may not be the desirable comparison group for women with two or more children. Based on Figure 4 and 5, I am more confident to rely on results from the comparison 1. I then present the event-history estimates of exit for the comparison 1 in Figure 7. As seen from Figure 7a, the March CPS estimates of exit are close to zero before 1994, and then dip a little and remain low. A similar pattern is confirmed by Figure 7b using the SIPP, with a stronger statistical significance for decreased exit in the post 1993 period. Finally, Figure 8 graphs the event-history estimates of entry for the comparison 1. Both data sets fail to show any effects on entry. But the estimates could not inform much information due to their big standard errors. Plotting the DD estimates directly in the above figures places no restrictions on the data, and it allows me to explicitly examine the common trend assumption. However, it also requires bigger sample size from the data to illustrate meaningful information. This is especially true with the entry estimates - it is derived from people out of the labor force, thus with fewer observations. To conclude from those graphs, I find results from the comparison 1 is more trustworthy. Also, there is some evidence for decreased exit associated with the policy changes, whereas no clear sign of increased entry. Next, I turn to regression results from Equation (3a) and (3b). Table 1 contains prepost DD estimates for the comparison 1. The March CPS results show, relative to unmarried childless women, the employment rate of unmarried mothers increased six percentage points in the post-93 periods, as a result of four percentage decrease in exits, and 8.3 percentage increase in entries. Put differently, the six percentage rise in the employment rate can be decomposed into a 48.4% contribution from decreased exit and a 51.6% contribution from increased entry. Findings from the SIPP is consistent with this result, indicating a 6.8 percentage increase in employment, of which 33% can be explained by decreased exit, and the rest explained by increased entry. The SIPP estimate of entry is relatively small with large standard errors indicating it is imprecisely measured. But its estimate of exit displays strong statistical significance. Therefore, the contribution from entry is calculated using one minus the contribution from exit. Although the results fail to detect an entry effect, according to the decomposition in Equation (1), the employment increase can only be attributed to increased entry if not to decreased exit. In addition to the EITC effects, the coefficients of welfare reforms meet earlier expectations. The March CPS suggest unmarried mothers living in states of AFDC or TANF increased employment by 5.6 percentage points relative to unmarried childless women in non-afdc/tanf states. This effect is mainly through significant increased entry rather than decreased exit. The SIPP data validates this finding by identifying a 4.5 percentage point exit-driven employment increase. As anticipated, an additional $1000 of maximum welfare benefit decreases employment by approximately 1.5 percentage points, which is also mostly caused by increased entry. 13

14 14 Table 2 shows the pre-post estimates with time-varying treatment effects from Equation (2b). The results from the March CPS and SIPP are consistent with each other. The March CPS suggest the employment increases 0.8 percentage points every year in the post period relative to This effect is mainly driven by a one-time permanent exit decrease of 2.3 percentage points in the post-93 periods. The SIPP results arrive at a similar conclusion. These findings line up closely with the simulation exercise. Table 3 shows the counterpart of Table 1 but for the comparison 2-unmarried mothers with two or more children relative to those with only one child. In theory, the effect of the EITC should be cleaner since the AFDC/TANF variable is less likely to differentially impact mothers with two or more children relative to one child. Therefore, a bigger exit effect is expected. The March CPS results confirm this hypothesis, suggesting a five percentage increase in employment, of which 88% is due to decreased exits. However, the SIPP results imply only 30% is due to the declined exit, and it is imprecisely estimated. Table 4 displays the time-varying treatment effect model for comparison 2. Findings from both data suggest a general time trend in employment instead of time-varying treatment effects. This feature has been predicted by Figure 5 in which a seemly pre-trends is detected. Finally, I turn to the results from the last model- the parameterized DD specification. Table 5 presents the effects of the maximum EITC, AFDC/TANF, and maximum welfare benefit on employment, exit and entry for the comparison 1. Specifically, for the March CPS, $1000 policy induced increase in the maximum EITC raises employment by 3.2 percentage points, driven mainly by 1.1 percentage decrease in exit. In addition, the AFDC/TANF increases employment by 6.2 percentage points, resulted from a 2.2 percentage decrease in exits and 7.9 percentage increase in entries. To make proper comparisons, I show that the 1993 means of the employment, exit and entry for the treatment group are 0.708, 0.089, and respectively. The SIPP data confirms these results with a bigger employment effect of 5.3 percentage points. When I limit the sample to unmarried mothers in Table 6 and conduct analysis with the comparison 2, the effect of the maximum EITC on employment becomes larger, with a greater contribution from decreased exit. However, the effect of the AFDC waiver and TANF disappears. By restricting the sample to be all mothers, thus eliminating the welfare impacts, these results are consistent with the hypothesis that exit effects are mainly from the EITC. For entry effect, the data seems to lack power of detecting it. Again, the SIPP results are consistent with the March CPS. In summary, both the March CPS and SIPP find employment of unmarried mothers increased due to the 1993 EITC expansion, equally through decreased exit and increased entry. The welfare reforms of the AFDC waiver and TANF also play an important role in raising employment, and those effects are mainly through increased entry. When comparing 14

15 15 unmarried mothers with two or more children relative to those with only one child, evidence suggests decreased exit is the major channel. 8. Sensitivity Analysis In this section, additional analyses are performed to validate 1) the consistency of employment results from the match/full samples of the March CPS and SIPP, and 2) The decreased exit effects associated with the program expansions are predominantly among the EITC eligible group. Table 7 displays the employment effects from pre-post DD model using the March CPS full/matched sample and SIPP full/matched sample. Findings from the March CPS full sample suggest a 7.7 percentage points increase in the employment for unmarried mothers relative to childless women, whereas the matched sample implies a 6 percentage points increase. The same pattern prevails with the SIPP: the full sample suggests a 7.7 percentage rise in the employment compared with a 6.8 percentage increase found by the matched sample. Despite the differences, these results are qualitatively similar based on their associated standard errors. In summary, the comparison results ensure that 1) results from the SIPP confirm the findings from the March CPS, and 2) the employment effect from the matched sample and full sample are consistent with each other. Next, I examine whether the exit effects are entirely from the EITC-eligible group, by interacting the EITC eligibility with the pre-post DD estimator. The EITC eligibility is generated using NBER TAXSIM calculator. It is an indicator based on actual family income, number of qualifying children, and tax year. I extend the Equation (3a) to include the threeway interaction- EITC_Eligibility*post93*treatment and the main effect of the EITC_Eligibility: Y N* = α + heitc_eligibiliy N Post93 Treat N + ζeitc_eligibility N + βpost93 Treat N + γwaivertanf * Treat N + ηmaxwelfare * + Year * + Treat N + δx N* + ε N* >>>>>>>>>>>>>>>>>>>(5a)>>> Similarly, I add the interaction of EITC_Eligibility with MaxEITC and expand Equation (4) in a similar manner: Y N* = α + heitc_eligibility N MaxEITC f* + ζeitc_eligibility N + βmaxeitc f* +γwaivertanf f* + ηmaxwelfare * + Year * + δx N* + ε N* >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>(5b) The test is performed on families that has ever worked in year t-1. The coefficient h+β is the DD effect for the EITC eligible families, and β shows the DD effect for those ineligible for the benefits. If the exit effects are indeed driven by the EITC eligible families, a negative h of a similar magnitude as the previous DD estimator in Equation (2a) should be found, along with a zero β. 15

16 16 Table 8 presents these results from the March CPS. First, in column (1) and (3), the findings confirm a negative and significant h and a small and insignificant β. Specifically, for the pre-post model, it shows the relative decrease in exits are 3.2 percentage higher for the EITC eligible families compared to the ineligibles, and the EITC doesn t show any impact on the latter group. For the parameterized model, it suggests a $1000 increase in the maximum amount of the credit decreases exit by one percentage points for the eligible families whereas exerts no impact on the ineligibles. In addition, comparing across models, I find the exit effects in the previous pre-post model (4 percentage points for the pre-post model; and 1.1 percentage points for the parameterized model) are entirely driven by the population who are eligible for it. This test validates my research design. 9. Conclusion In this paper, I evaluate how the EITC expansions and welfare reforms in the mid-tolate 90s affect employment, specifically, whether the employment effects are due to increased entry among non-workers or decreased exit among workers. I use a difference in difference identification strategy that exploits variation in the generosity of the EITC benefits as well as differences in the welfare eligibility. Identifying whether the policy-induced employment is due to entry or exit is important because it helps us understand how major and well-regarded cash and transfer programs have achieved their objective. In particular, it reveals whether the EITC and welfare reform have succeeded in raising employment among the hard-to-employ population. In addition, the study facilitates further discussion regarding how the EITC information is transmitted to low income families and why reduced exits occur responding to the EITC. Also, the results shed lights on the efficacy of the welfare reforms at assisting the transition from welfare to work. Using data from the March CPS and SIPP, I find the employment went up by six percentage points for unmarried mothers relative to unmarried childless women after the major policy reform. Approximately half of the employment increase can be attributed to fewer labor market exits, and half of the effect resulted from additional labor market entries. Notably, all of the reduced exits from the labor market are among families eligible for EITC benefits, which provides evidence of the validity of the research design. In contrast, a similar analysis applied to welfare reforms indicates the AFDC waiver or TANF increased employment by 5.6 percentage points. Consistent with the major features of welfare reforms, I find the Wavier/TANF-induced employment effect was entirely through increased entry. These findings raise questions about the efficacy of the EITC at inducing labor market entries, and suggest the unemployed may either lack awareness of the program or maintain a high reservation wage/inelastic labor supply. On the other hand, the adoption of the AFDC waivers and TANF is effective in facilitating the low-income families to leave welfare and enter the work force. So if future policy is to target at motivating the hard-to-employ group to 16

17 17 join the labor force, administrative mandates such as those applied in the welfare reforms may be preferred than non-administrative means featuring the EITC. Future work may extend this analysis in two directions: one is to study how the EITC information is transmitted to low-income families. Built on the literature of information and the EITC participation, one could study the role of peer effects on the dissemination of the program awareness; The other can be to identify mechanisms through which reduced exits occur responding to the EITC expansion. Specifically, one possible channel is workers may experience shorter unemployment spells with potentially lower wage associated with the program change. Overall, this paper presents evidence on how a series of well-recognized transfer policies have achieved their goal of promoting work by differentiating the entry and exit effects. It is important for improving employment policies and for understanding the lowskilled labor market more generally. In addition, the results are informative for comparable policies and provide guidance about how to design and implement the EITC and other social welfare programs. 17

18 18 Reference Blumenthal, M., B. Erard and C. Ho Participation and Compliance with the Earned Income Tax Credit. National Tax Journal 58(2): Meyer, Bruce D. and Dan T. Rosenbaum "Welfare, The Earned Income Tax Credit, and the Labor Supply of Single Mothers." Quarterly Journal of Economics 116(3): Eissa, N. and Hilary W. Hoynes Behavioral Responses to Taxes: Lessons from the EITC and Labor Supply. Tax Policy and the Economy 20: Hoynes, Hilary W. and Ankur J. Patel Effective Policy for Reducing Inequality? The Earned Income Tax Credit and the Distribution of Income. NBER Working Paper No Ellwood, David T The impact of the Earned Income Tax Credit and social policy reforms on work, marriage, and living arrangements. National Tax Journal 43 (4, part 2): Grogger, Jeffrey "The Effects of Time Limits, the EITC, and Other Policy Changes on Welfare Use, Work, and Income among Female-headed Families." Review of Economics and Statistics 85(2): Grogger, Jeffrey "Welfare Transitions in the 1990s: The Economy, Welfare Policy, and the EITC." Journal of Policy Analysis and Management 23(4): McKernan, Signe-Mary, Robert Lerman, Nancy Pindus, and Jesse Valente The Relationship between Metropolitan and Non Metropolitan Locations, Changing Welfare Policies, and the Employment of Single Mothers. Washington, D.C.: Urban Institute. Mimeograph. Mead, Lawrence M Overselling the Earned Income Tax Credit National Affairs 21: 20. Phillips, Katherin Ross The Earned Income Tax Credit: Knowledge Is Money. Political Science Quarterly 116 (3): Maag, Elaine Disparities in Knowledge of the EITC. Washington, D.C.: Urban Institute. Liebman, Jeffrey B. and Erzo F. P. Luttmer "Would People Behave Differently If They Better Understood Social Security? Evidence from a Field Experiment." American Economic Journal: Economic Policy 7(1): Bhargava, Saurabh and Dayanand Manoli Psychological Frictions and the Incomplete Take-Up of Social Benefits: Evidence from an IRS Field Experiment. American Economic Review 105(11):