Abstract

Refundable tax credits and food assistance are the largest transfer programs available to able-bodied working poor and near-poor families in the United States, and simultaneous participation in these programs has more than doubled since the early 2000s. To understand this growth, we construct a series of two-year panels from the 1981–2013 waves of the Current Population Survey Annual Social and Economic Supplement to estimate the effect of state labor-market conditions, federal and state transfer program policy choices, and household demographics governing joint participation in food and refundable tax credit programs. Overall, changing policy drives much of the increase in the simultaneous, biennial use of food assistance and refundable tax credits. This stands in stark contrast from the factors accounting for the growth in food assistance alone, where cyclical and structural labor market factors account for at least one-half of the growth, and demographics play a more prominent role. Moreover, since 2000, the business cycle factors as the leading determinant in biennial participation decisions in food programs and refundable tax credits, suggesting a recent strengthening in the relationship between economic conditions and transfer programs.

Introduction

Expenditures on means-tested transfer programs in the United States have risen dramatically in recent decades (CBO 2013a; Moffitt 2015b; Ziliak 2015a). However, this growth in program outlays is not universal. For example, inflation-adjusted spending on the Supplemental Nutrition Assistance Program (SNAP), the Earned Income Tax Credit (EITC), and the Additional Child Tax Credit (ACTC) grew by 290 %, 59 %, and more than 2,200 %, respectively, from 2000 to 2012 (see Fig. S1, Online Resource 1). In contrast, real spending on cash assistance from the Temporary Assistance for Needy Families (TANF) program declined, reflecting the transformation of the safety net away from cash-based assistance to one more reliant on refundable tax credits and in-kind food assistance (Bitler and Hoynes 2010; Hardy 2016; Moffitt 2013, 2015b; Ziliak 2015a). This spending pattern predated the Great Recession, suggesting that factors beyond the business cycle are potentially driving transfer program participation. These factors may include policy reforms affecting program eligibility and generosity, such as the 1990s welfare reform and expansion of EITC and creation of the ACTC, the decline in full-time work, stagnant earnings, the rise in disability, and changing demographic trends leaving families vulnerable to economic risk (Autor 2014; Autor and Duggan 2007; Autor et al. 2008; Cancian and Reed 2009; Gundersen and Ziliak 2004; Meyer and Rosenbaum 2001; Piketty and Saez 2003). Working alone or together, these demographic, macroeconomic, and policy forces might push the safety net toward a longer-term role as an income supplement for the working poor.

In this article, we estimate the determinants associated with the growth in longer-term reliance on the safety net over the past 30 years. Specifically, we construct a series of two-year (biennial) panels from the 1981–2013 waves of the Current Population Survey’s (CPS) Annual Social and Economic Supplement (ASEC) to estimate the effect of state labor-market conditions, federal and state transfer program policy choices, and household demographics governing joint participation of SNAP, the EITC, and the ACTC. With the model estimates, we then conduct a number of counterfactual simulations to assess the relative contributions of the economy, policy, and demographics on the dramatic rise in multiple transfer-program participation. We emphasize SNAP, the EITC, and the ACTC both because of the extraordinary growth in outlays and because of their ties to a more work-based safety net (Ben-Shalom et al. 2012). SNAP—known colloquially as “food stamps”—provides monthly benefits to support household purchases of food for preparation and consumption in the home. Eligibility for the program is income-conditioned, and although it is open to workers and nonworkers alike, there has been strong secular growth in the fraction of the SNAP caseload combining benefits with work (Hoynes and Schanzenbach 2016). The EITC and Additional CTC are refundable tax credits available annually if qualified when filing a federal personal income tax return. Like SNAP, the EITC and ACTC are income-conditioned; however, the tax credits are available only to workers. Also distinguishing these three programs is the fact that they are highly liquid; comparatively, SNAP is formally an in-kind program, but benefits are treated as near-cash by participants (Hoynes and Schanzenbach 2009). This stands in contrast to Medicare and Medicaid, two highly illiquid programs, or Social Security retirement and disability insofar as they are not conditional on sickness, age, or disability status. Moreover, unlike housing assistance, SNAP, the EITC, and the ACTC do not have waiting lists, so that all who are eligible should receive benefits after application.

Our empirical framework extends the prior literature in three substantive directions. First, previous research on SNAP and the EITC has relied on annual repeated cross-sectional data that provide only a snapshot of program use (Bitler et al. 2014; Meyer and Rosenbaum 2001; Ziliak 2015b). Although cross-sectional research designs capture important shorter-term conditions facing individuals and families (such as temporary job loss), our use of a panel permits us to study the factors driving less transitory, longer-term usage of transfer programs. We are necessarily limited to two-year panels based on the design of the CPS, but we demonstrate churn in program use and secular biennial growth in program participation missing from the cross-sectional case.

Second, to our knowledge, we are the first to examine the factors affecting biennial, multiple program participation in SNAP, the EITC, and the ACTC. Ziliak (2015b) focused solely on the factors driving the rise in SNAP participation in the cross-section, and Bitler et al. (2014) focused on cross-sectional analysis of the EITC. Research on multiple program participation in the cross-section, let alone over time, is scarce (Moffitt 2015a). Exceptions to this include recent work by Cancian et al. (2014) examining multiple program participation in Wisconsin through the lens of subsequent program disconnectedness, and Slack et al. (2014) examining how low-income families combine benefits in the post–welfare reform era. Slack et al. confirmed that many households combine benefits and thus treat assistance as a package and not necessarily independent. Similarly, Moffitt (2015a) showed that of those SNAP recipients captured in the 2008–2009 Survey of Income and Program Participation (SIPP), 38 % also received the EITC, and 28 % received the ACTC. These percentages rise to 53 % and 40 %, respectively, among the non-elderly nondisabled SNAP population, and to 89 % and 72 % among those with incomes between 50 % and 100 % of the poverty line. By contrast, only 10 % to 15 % of SNAP families in any one of these samples received TANF, again suggesting a shift to multiple-program participation that coincides with, or fosters, work—such as SNAP, the EITC, and the ACTC. For completeness, we also report briefly on models inclusive of TANF and Supplemental Security Income (SSI), a disability program not conditioned on work (results shown in Tables S15S15, Online Resource 1).

Third, we examine a wider array of family structures at risk of transfer use, and we focus more directly on the role of stagnant wages and the shift away from full-time work on participation. Our focus on families with children is motivated by the design of the American social welfare system, which is largely geared toward assisting adults with dependent children (Currie 2012, 2006). The key advantage of the ASEC is the comparatively large sample size, permitting us to focus on demographic groups most likely at risk of participation in the safety net from 1980 onward, especially families with low incomes and/or skills, as well as families headed by a single mother—the focal demographic group targeted by the 1990s welfare reforms. Moreover, the ASEC is the official data set used to calculate poverty and income inequality—in effect offering a useful baseline from which to assess and contextualize the predictors of participation in major social welfare programs.

We find evidence of significant growth in longer-term joint use of SNAP, the EITC, and the ACTC, increasing by 104 % across all families with children since 2000: by 2012, three years after the end of the Great Recession, nearly 1 in 16 American families with children relied on both programs across two years. This joint participation rate jumps to almost 1 in 5 families living below twice the federal poverty line, to more than 1 in 8 among single-mother families (regardless of income level), and 1 in 9 among family heads with a high school diploma or less. The model estimates suggest that SNAP operates as an unambiguous countercyclical policy with respect to state unemployment rates. Increases in the share of workers out of the labor force as well as the state median wage have led to decreases in joint use of SNAP, EITC, and ACTC, demonstrating the unique margins from which the EITC operates relative to traditional cash welfare: individuals out of work receive no benefit, and lower earners are supplemented. We also find that increases in real maximum SNAP benefits are associated with the growth in joint program participation; the generosity of the EITC phase-in rate is positively associated with EITC and ACTC program participation. This suggests that changes in the benefit structure of the programs have fostered the coupling of work with work-based assistance while strengthening the safety net for vulnerable low-skill families with children.

Combining the factors, the simulations suggest that the majority of the increase in joint use of SNAP and EITC/ACTC from 2000 to 2012 is associated neither with the cyclical nor structural aspects of the economy, nor changing demographics, but instead is attributed to changing policy in the SNAP and EITC programs. The primacy of policy holds for all subsamples and stands in stark contrast from the factors accounting for the post-2000 growth in SNAP alone, where cyclical and structural labor market factors account for at least one-half of the growth, and demographics play a more prominent role. Importantly, the former result of the importance of policy over the economy for the joint use of SNAP and EITC/ACTC since 2000 is sensitive to whether we use parameters estimated over the entire three-decade period or from the post-2000 period alone. In the latter case, the state macroeconomy becomes the leading factor for joint program participation, suggesting a strengthening in the relationship between economic conditions and participation in SNAP, the EITC, and the ACTC in recent years.

Setting the Context: SNAP, EITC, ACTC, and the Work-Based Safety Net

Fundamental changes in the U.S. social policy landscape throughout the 1980s and 1990s significantly altered the economic rewards to work and to participation in transfer programs, affecting all segments of the low-income population—especially single-mother headed families. During his first campaign for president, then Governor Clinton pledged to “end welfare as we know it,” and upon his election to the presidency in 1992, states aggressively pursued waivers from federal rules governing their main cash welfare program, Aid to Families with Dependent Children (AFDC). These changes were codified into federal law with passage of the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (“welfare reform”), which eliminated the AFDC program and replaced it with TANF. TANF policies varied widely across states but generally included a set of “carrots” and “sticks.” The former included both expanded liquid asset limits for eligibility and earnings disregards that permitted mothers to retain a higher fraction of benefits as their earnings increased; the latter included (among others) work requirements and time limits for benefit receipt. Importantly, funding was converted from an open-ended entitlement financed by a federal-state matching grant under AFDC to a fixed (in nominal dollars) $16.5 billion federal block grant under TANF. States reconfigured their programs from primarily providing cash assistance to predominantly providing in-kind assistance, such as childcare vouchers, workforce training, and marriage counseling; the reconfigured programs became much less target-efficient for assisting the poor (Bitler and Hoynes 2016).

Concurrent to the new cash welfare law were enhanced incentives for low-income individuals to work via expansions in the EITC (Eissa and Hoynes 2004), the creation of the Additional CTC, and liberalization of eligibility for participation in SNAP. The EITC was established in 1975 as the first refundable credit in the federal tax code in a bid to increase the incentive to work by offsetting the regressive Social Security payroll tax among low-wage workers (Hotz and Scholz 2003). The size of the credit first rises with earnings, is then constant over a certain range, and finally is gradually phased out as earnings continue to increase. Provided that the credit amount exceeds taxes owed, the balance gets refunded. Over time, the implicit scope of the program expanded to not only spur work but also combat poverty. This occurred through a series of increases, first as part of the Tax Reform Act of 1986 and then the Omnibus Budget Reconciliation Acts of 1990 and 1993, the latter of which increased the generosity of the credit, especially for families with two or more qualifying children. Finally, the American Recovery and Reinvestment Act (ARRA) of 2009 created a new, higher credit for families with three or more qualifying children (Bitler and Hoynes 2016; Steuerle 2015). In tax year 2015, the maximum credit for a single-headed household with two qualifying children was $5,548, or 40 % of earnings. An estimated 5 to 6 million people are lifted out of poverty each year by the EITC, and more than 9 million when the alternative poverty line from the U.S. Census Bureau’s Supplemental Poverty Measure is used (e.g. Fox et al. 2015; Nichols and Rothstein 2016).

Simultaneous with the rollout of the TANF program, Congress passed the Taxpayer Relief Act of 1997 that introduced the Additional Child Tax Credit. Initially the credit was worth up to $400 for each qualifying child under age 17, but through a series of legislative changes, the credit today is worth up to $1,000 for each qualifying child. It also has a refundable component if the amount of the ACTC exceeds the amount of tax owed and (1) the household has three or more dependent children or (2) earned income exceeds $3,000 (Congressional Budget Office 2013b). As shown in Fig. S1 in Online Resource 1, spending on the Additional CTC is now in excess of $30 billion annually and larger than TANF. The December 2015 federal budget agreement made all these ARRA expansions in the EITC and ACTC a permanent part of the income tax structure.

The 1996 welfare reform also affected SNAP rules. SNAP (originally known as the Food Stamp Program) was established in 1964 to provide food assistance to low-income and low-asset households. Benefits are federally funded, with the maximum benefit varying by household size but constant across the 48 contiguous states and the District of Columbia. Initially, households were required to meet two income tests—gross income less than 130 % of the poverty guideline for the household size, and net income below 100 % of the line—and two asset tests, one pertaining to holdings of liquid wealth (e.g., cash, checking, savings) and one to vehicle wealth. The gross income limit is waived for households containing a disabled person or with individuals aged 60 or older, and the liquid asset limit is higher for these households. No work requirement was associated with benefit receipt, and benefit was available to all legal residents who qualified. This changed with welfare reform, wherein restrictions on benefit receipt were imposed on legal immigrants and so-called ABAWDs—able-bodied adults without dependents—working less than 20 hours per week. The reform also reduced the maximum benefit and froze many deductions used in calculating net income; it allowed states to sanction individuals and households for noncompliance with TANF requirements or child support payments; and it mandated that states replace paper voucher coupons with the electronic benefit transfer (EBT) debit cards. Take-up rates (i.e., participation rates among those eligible) plummeted from nearly 75 % before reform to just more than 50 % five years after (Leftin et al. 2011). In response, rule changes implemented first by U.S. Department of Agriculture, and then within the 2002 Farm Bill passed by Congress, restored eligibility for most of the legal residents removed by the 1996 reform, liberalized financial eligibility rules (notably asset tests), and allowed states to utilize broad-based categorical eligibility that gave flexibility to apply more generous TANF asset and gross-income tests to determine SNAP eligibility. Many states also shifted to electronic applications, reducing the potential stigma associated with SNAP use. Furthermore, the 2008 Farm Bill gave states the option of increasing or removing both the vehicle and liquid asset tests. Although all states have removed the value of at least one vehicle from the test, upward of 38 states have also eliminated liquid asset tests at some point over the past decade.1

Combined, the restrictions placed on cash welfare, alongside new incentives to work with EITC expansions and the introduction of the ACTC, point to expected increases in program take-up. The retrenchment of SNAP eligibility in the 1990s and the liberalization of program rules in the 2000s point to decadal-specific shifts in policy-induced program participation. For example, with respect to SNAP, Ganong and Liebman (2013) found that longer recertification periods—whereby recipients update information on income and eligibility—and simplified reporting positively impact program take-up. Likewise, for the EITC, workers eligible for lower credits in the phase-in range are less likely to participate, as are those with less education (Jones 2014).

These policy reforms did not occur in isolation from broader macroeconomic and demographic trends. On the macroeconomic front, four recessions have occurred since 1980: 1981–1982, 1991, 2001, and the Great Recession (officially spanning from December 2007 to June 2009). Three of the four longest economic expansions since World War II followed these downturns. All else equal, participation in SNAP is expected to be countercyclical—rising when the economy is falling—as households experience declining earnings and other forms of economic hardship.

However, participation in the EITC and ACTC is less clear over the business cycle. Falling incomes from earnings and/or hours reductions could bring some workers and families into eligibility as low earners. This may be especially true for formerly EITC-ineligible married families who are buffered against economic shocks by the EITC when one worker faces economic hardship or job loss. On the other hand, if job loss occurs and results in a total loss of earnings, economic shocks associated with unemployment could result in lowered EITC and ACTC participation. During economic expansions, low-skilled individuals out of the labor force could be enticed to enter, potentially increasing EITC and ACTC participation, and those working could receive wage and/or hours boosts that lift them out of eligibility. These potentially heterogeneous participation effects suggest that we should estimate our empirical model for different skill and income groups.

On top of the business-cycle shocks are important secular economic trends that could affect program participation. Inflation-adjusted wages have been stagnant or declining in the lower half of the wage distribution for the better part of four decades (Autor et al. 2008), which could lead to greater long-term attachment to programs such as SNAP, the EITC, and the ACTC. Likewise, detachment from the labor force for extended periods (continuous two years) rose more than fourfold, from approximately 8 % to 38 % among less-skilled men of prime working age (Ziliak et al. 2011), which would at once lead to increases in SNAP and decreases in the EITC and ACTC. This has been buttressed by strong secular growth in disability, again pulling toward higher SNAP and lower EITC and ACTC, with spending on Social Security Disability Insurance (SSDI) and Supplemental Security Income (SSI) programs more than doubling in real terms to nearly $200 billion annually since 2000 (Autor and Duggan 2007).

The changing demographic makeup of the American household could also have potentially countervailing influences on trends in SNAP, EITC, and ACTC participation. Population aging and delays in childbirth each likely put downward pressure on program participation because take-up rates in means-tested programs are lower among older adults and those without children. Likewise, secular growth in high school completion and college attendance since 1980 also put downward pressure on SNAP program use, primarily because incomes and education are positively correlated, but may have resulted in greater take-up of EITC and ACTC. On the other hand, the rise of out-of-wedlock childbirth and delay of marriage point toward increased reliance on public assistance because of greater economic need (Tach and Eads 2015). One would also expect an interaction between family structure and social policy reforms, such as welfare reform and the EITC. Namely, as the 1996 welfare reform was aimed at reducing welfare use and increasing work among single-mother families, this demographic group should especially influence the growth in the EITC, ACTC, and SNAP. As such, we estimate the empirical model below separately for single mothers, the low-income, and the low-skilled.

Empirical Model and Data

Based on the preceding discussion, our objective is to model multiyear participation in SNAP, EITC, and ACTC as a function of household demographic characteristics, the state macroeconomy, and state-level policy choices as
yijkt,t+1=αk+Xitγk+Zjtδk+πjk+φtk+uijtk,
1
where yijkt,t+1 is an indicator equal to 1 if anyone in household i residing in state j receives program k (k = joint SNAP, EITC, and ACTC) at both time t and t + 1; Xit is a vector of demographic characteristics of the household head in the initial period t; Zjt is a vector of state (or federal) by year economic and policy variables from period t; πjk is a set of indicators for each state to control for fixed but unobserved state-specific factors affecting participation; φtk is a set of indicators for each year to control for macroeconomic and policy factors that affect all households the same in a given year but differ over time; and uijtk is a random error term. Although our ultimate focus is on joint program participation, we also estimate Eq. (1) separately for SNAP alone and for EITC plus ACTC.

The primary data used in our study come from the CPS ASEC for calendar years 1980 to 2012 (interview years 1981 to 2013). The CPS is a monthly survey of the U.S. labor force based on a stratified random sample of 60,000 households, with the ASEC fielded in March of each year (and with some portion from the February and April samples) to collect information on household income, family structure, and health insurance in the prior calendar year. The CPS has a rotating sample design whereby respondents are in-sample for four months, out-of-sample for eight months, and then in-sample for four more months. This makes it possible to match up to one-half of the sample from one ASEC interview to the next, creating a series of two-year (biennial) panels that form our measure of longer-term participation in welfare restricted to 20- to 55-year-olds. Our final data set consists of 694,278 matches across the 30-year sample, with a match rate of 55 %. More detail on sample selection and longitudinal matching of the ASEC is provided in Online Resource 1.2

In the ASEC, SNAP participation is asked at the household level: specifically, whether anyone in the household received SNAP in the last calendar year. A household can contain more than one SNAP unit or only a subset of members to be on assistance. With our focus on the family, we make the implicit assumption of resource sharing within the household such that all members benefit from SNAP even if they are not direct recipients. Ziliak (2015b) showed that from 1980 to 2000, population-weighted participation rates were broadly comparable with administrative data. However, the past decade has seen a divergence in the levels, although not in trends. (The levels gap was previously highlighted in Wheaton 2007 and Meyer et al. 2014.) Information on the EITC and the refundable portion of the ACTC are not collected in the ASEC, and thus we rely on simulated eligibility based on the National Bureau of Economic Research (NBER) TAXSIM model.3 The assumption in TAXSIM is that take-up rates are 100 % when in fact they are closer to 80 %, although take-up in the EITC is lowest for those who receive a lower tax bill and not a refundable credit (Jones 2014). That is, for the EITC and ACTC, we focus on eligibility rates rather than actual participation; but for convenience, we refer to these as EITC/ACTC participation rates. Because we are interested in the persistence of participation over time, we construct an indicator of whether the family is receiving SNAP for both years they are in the sample. Because the ACTC was not in effect until 1998, we combine these two refundable credits to construct an indicator for whether a family receives either or both the EITC or the ACTC in both years; we refer to this combined program as EITC/ACTC. Our broader measure of the safety net then is an indicator variable for whether the family receives SNAP and the EITC/ACTC in both years.

We consider four subsamples for our analysis of safety net participation: (1) all families; (2) low-income families (having family income-to-needs below 200 % of the federal poverty line in each of the two years); (3) low-education families (those whose head has a high school diploma or less); and (4) single-mother families (mothers who head the family in both years). The longer-term low-income sample is of interest because federal gross income eligibility for SNAP for the non-elderly is capped at 130 % of the household-specific poverty line; however, since 2000, many states implemented broad-based categorical eligibility that lifted gross income tests to 150 % to 200 % of the poverty line. Many low-income families fall within this category; roughly one in four sample respondents lie within this threshold for both years (see Table S3, Online Resource 1). Moreover, the subpopulation of those under 200 % of poverty is the group that generally qualifies for the EITC. However, we recognize that this low-income sample may be endogenous with SNAP and EITC/ACTC participation in that the programs are means-tested. Thus, we also present the low-education sample on the standard assumption that educational attainment is a proxy for permanent income, with high school or less signaling high risk for program participation (Bloome and Western 2011; Corcoran 2001; Hoynes et al. 2006). Last, we present the subsample of single-mother families because this was the group most affected by the 1990s welfare reforms and expansion of the EITC/ACTC.

The demographic controls in Eq. (1) include indicators for the household head’s age (with ages 20–34 as the omitted group), educational attainment (relative to high school dropout), race (relative to white), Hispanic ethnicity, female headship, and marital status (relative to widowed/separated/divorced/never married); the number of individuals in the household (including nonfamily members); the number of related children under age 18; residency within a metropolitan area; and within a metro area, residence in the central city to capture within-metro heterogeneity.4 All these measures are based on Year 1 of the match.

The measures of state labor market conditions include the contemporaneous unemployment rate, along with one- and two-year lags in order to capture potential business cycle dynamics; the fraction of individuals working full-time, part-time, and out of the labor force; and median wages as the focal measures of cyclical and structural changes in the macroeconomy. The key policy variables at the state and federal level include the larger of the real state or federal minimum wage rate, the maximum subsidy rate for the EITC based on year and number of qualifying children, whether the state offers a refundable state EITC, and the family size–specific maximum SNAP benefit. Even though it is set nationally, the EITC subsidy rate is identified by its variation over time and by the number of qualifying children (Hotz and Scholz 2003). For the SNAP policy variables, we assign the real maximum benefit guarantee for a one-, two-, three-, or four-person household based on family size (the four-person guarantee is assigned to households with four or more people) to measure the financial generosity of the program. Like the EITC subsidy rate, the SNAP benefit is identified in the model because it varies over time and by household size.5 We also include indicators for (1) whether a state implemented a waiver from its AFDC program; (2) when its TANF program was implemented; (3) a host of SNAP policy variables, such as the fraction of SNAP dollars redeemed via the EBT, indicators for whether the state allows broad-based categorical eligibility, noncitizen SNAP eligibility, whether it imposes short recertification periods of three months or less for households with a working member, whether the household must be fingerprinted (either statewide or partial state), whether the household is disqualified for being sanctioned by another program such as TANF, whether the state adopted simplified reporting, whether it excludes the full value of a vehicle for eligibility, and the real value of spending on outreach; and (4) whether the governor was a Democrat.6 Our choice of SNAP policy variables is consistent with those employed throughout the literature (e.g., Ganong and Liebman 2013; Ratcliffe et al. 2008) to account for both positive and negative administrative and remunerative policy incentives driving program participation. Basic summary statistics on the variables used in the regression analysis for each sample are presented in Table S3, Online Resource 1.

Results

Biennial SNAP, EITC, and ACTC Participation

We begin our results section with a descriptive analysis of two-year trends in SNAP, the EITC, and the ACTC, and joint program participation. Because analysis of biennial program participation is a key contribution of our study, an examination of trends in participation is of interest on its own.

Figure 1 depicts trends in SNAP, EITC, and ACTC, and combined SNAP and EITC/ACTC participation for each of our samples across the two-year matches in the ASEC. Each panel in the figure shows strong secular increases in biennial use of the EITC/ACTC starting after the first expansion in the Tax Reform Act of 1986, with a subsequent surge after the 1993 expansions (note the different scales for each panel). Over the whole sample period, biennial participation in the EITC/ACTC more than doubled for the sample overall and for low-income families, more than tripled for the low-education group, and rose a substantial 48 % among single-mother families. There are substantive differences in the level of participation, with rates nearly twice as high among the low-income sample compared with the low-skilled sample, and with rates among single-mother families falling in between (although closer to the low-skilled rates). For the low-income sample, biennial participation in the EITC and ACTC peaked in 2006; for single-mother families, biennial participation in the EITC and ACTC peaked in 2008 and fell in the subsequent years. For all families, EITC and ACTC participation stabilized toward the end of the sample period. Although tax credit participation appears to be more secular than cyclical, we test this formally in our subsequent empirical analysis.

The biennial trends in SNAP in Fig. 1 are considerably different than the EITC and ACTC. First, much like one finds in annual participation rates in Ziliak (2015b), there is evidence that biennial SNAP participation responds countercyclically with the health of the macroeconomy, with increases evident in the years surrounding recessions. This holds across all four samples. Second, changes in SNAP participation from 1981 to 2012 are substantial, but at a 50 % increase, growth in participation lags behind the overall rise in EITC and ACTC participation. In fact, among single-mother families, biennial SNAP participation rates fell 3 percentage points over the 30-year period. Third, biennial SNAP participation increased aggressively in response to the Great Recession and remains at elevated levels.

Rates of joint participation in SNAP, EITC, and ACTC are lower than either program in isolation. However, joint participation increased dramatically, by 180 %, over the past three decades across all families; this increase was led in part by the 250 % increase (from 3.5 % to 12.1 %) among low-skilled family heads. Among single-mother families, the 90 % increase in joint program participation meant that by 2012, 16 % relied on both programs for consecutive years. Figure 1 suggests that joint participation moves with the business cycle more like SNAP than the EITC, and again like SNAP alone, joint participation increased substantially with the onset of the Great Recession.

To further assess and decompose the prevalence of biennial program use, we construct a transition matrix of program participation, shown in Table S4, Online Resource 1. The rows in the table sum to 100 %, subject to rounding error. We find that during the 1980s, 47 % of families receiving SNAP and EITC/ACTC in Year 1 subsequently received both programs in Year 2. Year 2 dual-program participation, conditional on Year 1 participation, subsequently increased to 52 % of families in the 1990s and 54 % of families in the 2000s. We also see a fair amount of churn in program use across years. For example, SNAP receipt in Year 1 seems to be more of a gateway to joint program use in Year 2 (because individuals combine work and SNAP) than Year 1 EITC participation alone, and joint program use in Year 1 is more likely to result in EITC alone in Year 2 than SNAP alone.

Employment and Wages

Because eligibility for participation in the EITC and ACTC is work-conditioned and income-tested, and SNAP is means-tested, an obvious place to look for evidence of changes in participation in those programs is changes in macroeconomic labor market conditions. The top panel of Fig. 2 presents trends in state unemployment rates (right axis) along with state-level averages in the fraction of the sample with continuous (i.e., biennial) full-time employment, continuous part-time employment, and continuous status of not in the labor force (NILF). The unemployment rate series is that estimated by the U.S. Bureau of Labor Statistics (2014) from the CPS, current employment statistics, and state unemployment insurance claims data.7 The full-time, part-time, and NILF series are estimated from our CPS sample of matched individuals (i.e., 465,091 pairs) and aggregated up to capture state differences over time in employment opportunities as reflected in the intensity of labor-market attachment.

Figure 2 shows substantial swings in state unemployment rates with the business cycle, and these swings coincide with changes in biennial SNAP participation in the previous figure. Importantly, although the peak unemployment rate in the Great Recession actually was slightly below that of the 1981–1982 recession, the run-up of unemployment in the most recent downturn was much larger, and recovery was much smaller, each of which could account for the sustained levels of transfer program participation in the 2000s. Likewise, there are substantive changes in state labor market opportunities in terms of full-time and part-time employment as well as complete labor force exit. Although a cyclical component of continuous full-time employment is evident in the figure, a change in secular trends also appears, with the pre-2000 period exhibiting increases in biennial full-time work and then a reversal in the post-2000 period. The latter seems to have been met more by a secular increase in labor force withdrawal than in part-time work. This could help account for the more rapid growth in SNAP compared with the EITC and ACTC because SNAP has no explicit work requirement (except for ABAWDs).

The second panel of Fig. 2 depicts trends in state-level compensation for low- and middle-skilled workers. Specifically, we present inflation-adjusted state minimum hourly wage rates along with the inflation-adjusted state median biennial average hourly wage.8 The figure shows secular growth in median wages during the first two decades but a flattening out and then a decline in hourly wages of the typical worker after 2000: the median real wage in 2012 was the same as in 1998. Wage opportunities among the least skilled also declined sharply in real terms in the 1980s when the nominal federal minimum remained fixed at $3.35 per hour, and few states’ minimum wages deviated from the federal rate. Minimum wage held steady for the next 18 years until the Great Recession, when both the federal government and states increased it. However, the average real state minimum wage was actually 10 cents per hour lower in 2012 than in 1981. The stagnation of wages in the bottom half of the income distribution points both to increased eligibility and need for assistance from SNAP and the EITC and ACTC.

To gauge the importance of this growth in biennial joint program participation for family budgets, the bottom panel of Fig. 2 presents trends in the ratio of biennial average benefits from SNAP, the EITC, and the ACTC to biennial average family earnings among those with positive labor earnings in both years. The figure makes clear that the programs are increasingly “filling the gap” for low-income and low-skilled families. Assistance from these programs had the effect of raising family earnings by 25 % in 2012 among low-income families—a ninefold increase from three decades earlier.

Regression Results

We next turn to our regression results from estimating Eq. (1) via linear probability, where we correct the standard errors for heteroskedasticity and within-state autocorrelation arising from the fact that multiple households are present in each state. All models control for fixed state and year effects and are weighted using person-level sampling weights adjusted for the probability of Year 2 selection on observable demographics, conditional on Year 1 selection.

Table 1 presents estimates for the biennial SNAP participation models for each of the four samples. Comparisons across columns reveal considerable consistency in the marginal effects of the family-level variables on the probability of participating in SNAP for two consecutive years. Namely, participation in SNAP is lower for older heads, those with higher education attainment, whites and non-Hispanics, larger households, and married heads. Participation is higher for female-headed families and families with more related children under age 18. The magnitude of the coefficients differs because the baseline probabilities vary substantially across samples: average SNAP participation is 0.084, 0.302, 0.139, and 0.287 for the sample overall, for low-income, low-education, and single-mother-headed families, respectively (see Table S3, Online Resource 1).

The estimates in Table 1 provide strong evidence that biennial SNAP participation is countercyclical with respect to the state unemployment rate, which is consistent with annual estimates from pooled cross-sectional data (Ganong and Liebman 2013; Ziliak 2015b). For example, in the full sample, a 1 percentage point increase in the unemployment rate that persists two years leads to a 0.63 percentage point increase in biennial SNAP participation (the sum of the three coefficients on the unemployment rate).9 It is often more convenient to express this as an elasticity.10 The elasticity of continuous SNAP participation with respect to the two-year lagged unemployment rate is 0.47 for the full sample of family heads (see bottom of Table 1 for elasticities): that is, a 10 % increase in unemployment results in a 4.7 % increase in biennial SNAP participation. This business cycle relationship persists for the remaining categories, with an elasticity range of 0.44 (low-income heads) to 0.47 (single-mother heads). The positive relationship between the proportion of the state’s labor market that is out of the labor force and continuous SNAP participation rates is statistically significant for all but single-mother-headed families, and their SNAP out-of-the-labor-force elasticity is 0.20, which is similar to that of all heads (0.20) and less-educated heads (0.24) and less than low-income heads (0.38). These results are broadly consistent with those in Ziliak (2015b) and Bitler and Hoynes (2016), showing that SNAP performs countercyclically—with unemployment as the business cycle indicator—during the Great Recession. We find this result holds for unemployment as well as the proportion out of the labor force, and that this occurs over a 30-year period inclusive of but not limited to the 2007–2009 Great Recession. In these and remaining empirical models, we depart from Ziliak (2015b) by estimating a broader set of structural economic, policy, and demographic variables for dual program participation over two years, including additional controls for part-time employment, labor force status, and an indicator for state EITC—stratified by the adult head’s income level, educational attainment, and marital status.

Across all samples except low-income, increased generosity of the EITC leads to lower continuous SNAP use, with elasticities ranging from –1.27 in the full sample to –0.48 among single mothers. Because the generosity of the recipient’s SNAP benefit is not reduced by the size of the EITC, this reflects a behavioral response to increasing work and reducing food stamps rather than a mechanical relationship between the EITC and SNAP allotments. On the other hand, increases in the generosity of the SNAP benefit lead to increases in SNAP participation for all samples, consistent with demand theory. The elasticity of continuous participation with respect to the SNAP benefit generosity ranges from 0.30 for single mothers and 0.84 for low-skilled heads to 1.62 for the sample overall. The greater responsiveness of the full sample compared with the more disadvantaged subsamples stems in part from more permanence in SNAP use among disadvantaged groups. Most of the other SNAP policy variables have no statistically significant effect on continuous participation, with a few exceptions, such as simplified reporting and outreach spending.11 Across all samples, residing in a state with a Democrat as governor is associated with higher odds of continuous SNAP use, which may reflect overall state climate governing program access.

Table 2 contains the parallel set of results for biennial participation in the EITC and ACTC programs. In general, the results of the family-level demographic factors are similar between the SNAP models of Table 1 and the EITC/ACTC models of Table 2, with some notable exceptions where low-income and single-mother families differ from the low-skilled and families in general. For example, among these families, there is no significant difference between older and younger heads, but participation is increasing in education attainment and is higher among married heads. The positive association between education and EITC/ACTC use among the poor is consistent both with greater program knowledge and labor force attachment (Chetty et al. 2013; Moffitt 2015b).

Among the state and federal economic and policy variables, there are also several key differences from the SNAP models. For example, across all samples, the quantitative effect of state unemployment rates on EITC and ACTC participation is considerably smaller in absolute value than the SNAP-alone models (see the elasticities in the lower panel) and generally statistically insignificant. This implies that SNAP functions more as an automatic stabilizer than refundable tax credits. This is further underscored by the different qualitative associations: EITC and ACTC participation is generally acyclical with respect to the state unemployment rate, as demonstrated by the elasticities.12,13 Another key difference in Table 2 from the SNAP models is the consistently strong effect of the share of part-time workers on the odds of continuous EITC and ACTC participation, buttressing the result that the program serves as a longer-term work support for those in longer-term part-time employment.

The estimates in Table 2 show economically important own and cross-price effects with respect to program generosity. That is, continuous participation in the EITC/ACTC is strongly positively associated with the generosity of the EITC phase-in subsidy rate and negatively associated with the generosity of SNAP (although smaller in absolute value). Table S3 in Online Resource 1 shows large differences in biennial participation rates in the EITC and ACTC across samples, ranging from 16 % among all families to 53 % among low-income families. The associated elasticities of continuous EITC/ACTC participation with respect to the phase-in rate range from 1.15 among single-mother families to 0.76 among low-income families. The estimate for single-mother families is equivalent to the estimate of 1.1 in Meyer and Rosenbaum (2001), even though our study differs in that we are examining both the EITC and ACTC for two consecutive years as opposed to single-year use of the EITC alone, and our study covers an additional 18 years of data when single mothers’ employment rates were much higher than in the 1980s and early 1990s. To our knowledge, our study is the first to estimate such elasticities for the wider low-skilled and low-income populations. The corresponding elasticities of EITC/ACTC with respect to the SNAP maximum benefit are –0.19 for low-income families, –0.35 for single-mother families, –1.30 for low-skilled families, and –1.35 for all families.

We find positive associations of TANF implementation with continuous EITC/ACTC participation among single-mother family heads, which aligns with the wider cross-sectional welfare literature finding higher labor force participation in response to TANF implementation (Blank 2009). SNAP policy has no additional consistent association with EITC participation across groups when examining the remaining policies because neither program benefit is counted as a resource when determining the other program’s benefit.

Table 3 examines the determinants of joint participation in SNAP, EITC, and the ACTC across two years. Here we see an interesting mix of estimates where in some cases, the coefficients align with the SNAP-alone models and others the EITC/ACTC models. For example, qualitatively all the coefficients on the demographic factors align with the continuous SNAP models, although the magnitudes are attenuated for age, education, race/ethnicity, household size, gender, and marital status. Also notable is that joint participation in SNAP, EITC, and ACTC is countercyclical, similar to our findings in the SNAP-alone models and consistent with the income smoothing effects of the tax and transfer system found in Larrimore et al. (2013). And although the coefficients are smaller in magnitude, the corresponding elasticities for biennial lagged unemployment at the mean are comparable, ranging from 0.31 to 0.39.14 The proportion of the state’s labor market that is out of the labor force is now negatively related to joint participation, thus reflecting the direct relationship between EITC and work. This is consistent with recent work examining EITC eligibility by Jones (2015), who found that less-educated, single females are likely to experience reduced EITC due to job loss during the Great Recession.

Related to this, higher state median wages are now associated with a lowered likelihood of joint program participation, with sizable elasticities: higher-earning workers are eventually disqualified from both EITC and ACTC and SNAP receipt. The EITC subsidy rate has a much more muted effect on joint participation. The positive coefficient on the maximum SNAP benefit guarantee, like that observed in Table 1, suggests that more generous SNAP benefits are associated with higher rates of persistent use of joint programs, underscoring the importance of these programs to the work-based safety net.

Although most of the other state SNAP policies have a qualitative positive effect on joint participation, they are not individually statistically significant, with few exceptions. Vehicle assets stand out as a formerly unimportant policy in the SNAP-alone models that now positively predict SNAP and EITC/ACTC participation for all groups except the full sample, and outreach spending has the expected positive sign. Again as in Table 1, residing in a state with a Democrat as governor is positively associated with biennial participation in multiple programs.

Counterfactual Simulations: What Accounts for the Post-2000 Growth?

In Tables 4, 5, and 6, we return to the issue of whether changes in longer-term SNAP, EITC, and ACTC participation since 2000—when dramatic spending increases in both programs begin to appear—are largely associated with cyclical and structural changes in the macroeconomy, in policy reforms, or in other factors aligned with changing demographics of the American family. We do so by conducting a series of counterfactual simulations based on the parameter estimates from the models in Tables 1 (for SNAP alone), 2 (for EITC and ACTC alone), and 3 (for joint SNAP, EITC, and ACTC). Specifically, we ask, What would biennial participation in each program alone or in combination be if (1) state economic factors (unemployment rates, rates of full-time/part-time/NILF, median wages) had remained fixed at their 2000 values, (2) state and federal policies remained fixed at their 2000 values (state minimum wage, EITC subsidy, SNAP benefit, welfare, and SNAP reforms), or (3) average demographics of the family remained fixed at their 2000 values? Each of the three experiments allow all other factors to change over the 2000–2012 period except for the set of variables being held constant at the 2000 values. We perform these simulations for each of the four samples. As a robustness check, Tables 7 and 8 then repeat the analysis for SNAP alone and joint SNAP and EITC/ACTC but instead use within-period parameter estimates: that is, the models and simulations both use data only from 2000–2012.15

In Table 4, biennial participation from 2000 to 2012 in SNAP among all families increased 104 %. If we fixed the state economic factors at their 2000 values, we would have predicted a 37 % increase in SNAP participation. This implies that changes in the state business cycle accounted for 70 %

100×1-ΔSNAP^ΔSNAP
of the change over the biennial period. (Numbers in the table are rounded to the nearest whole number, and shares do not sum to 1 because some factors, such as year and state effects, are not reported.) Policy changes are equally important, with a 71 % share, but demographics have a more modest role of 15 %. (Again, recall that shares do not necessarily sum to 100 % because of omitted factors.) A similar pattern holds for low-income and low-skilled families, although for low-income families, the economy is more important (89 %) than policy (68 %); for low-skilled families, the economy-policy relationship is again evenly split, with 54 % of SNAP participation explained by changes in the economy, 56 % explained by policy, and a more substantive 36 % explained by demographics. Among single-mother families, policy changes account for 79 % of the observed 38 % growth in biennial SNAP participation, but cyclical and structural labor market factors are associated with 154 % of the growth. Demographic shifts account for a negative share of the growth, meaning that changes in the demographic composition of single-mother families alone actually slowed the growth of longer-term SNAP participation.

In Table 5, we examine the predictors of biennial EITC and ACTC participation alone based on the parameter estimates in Table 2. In comparison with biennial SNAP participation, the EITC and ACTC simulations illustrate that different groups exhibit unique responses. For the full sample, economic and demographic factors are equally important, and this roughly holds for low-skilled family heads. For the low-income sample, policy changes are the most important predictors of EITC and ACTC participation. Less-educated family heads are roughly evenly affected by the economy, policy, and demographics. Families headed by a single mother are most sensitive to the economy as it relates to EITC and ACTC use; had the economy remained at its 2000 level, our models would have predicted 2 % growth in EITC and ACTC, which instead emerged as a 9 % decline in forecasted participation.

In Table 6, we repeat the exercise but now predict changes in joint biennial participation in SNAP, EITC, and the ACTC based on parameter estimates in Table 3. With the joint participation models, a different picture emerges compared with SNAP or EITC/ACTC alone: policy changes over the past decade are associated with the majority of the growth in joint program participation for all group samples. The implication of the simulations is that continued weakness in the labor market, coupled with secular declines in full-time work, will likely lead to continued persistence in SNAP participation; for the subset of families using SNAP, EITC, and ACTC over time, policy choices will loom large.16

We next use regression parameter estimates with data restricted to 2000–2012 to implement the same simulations as described in Tables 4, 5, and 6 as a robustness check. If there were structural changes in the relationships between the business cycle and work with welfare use over time, our results could potentially differ. As reported in Tables S5 and S6 (Online Resource 1), this appears to be the case. Comparing the parameter estimates in those tables with Tables 1 and 3 suggests that the effects of state unemployment are much stronger in the 2000s compared with the 1980s and 1990s.

The implication for the counterfactual simulations reported in Tables 7 and 8 is that the state macroeconomy takes an even more prominent role in accounting for the growth in longer-term program participation. We highlight this effect within the joint SNAP, EITC, and ACTC simulations of Table 8 (based on parameter estimates in Table S6, Online Resource 1). The macroeconomy is the primary driver when the estimation sample is restricted to the 2000–2012 period, which perhaps is not surprising because most of the policy changes to the EITC and welfare took place in the 1980s and 1990s and the economy withstood a historic contraction after 2007.

Policy choices do explain a substantial portion of joint SNAP and EITC/ACTC participation among single-mother and low-income families, although still less than the structural economic factors. In comparison with the baseline simulations, the biennial joint SNAP and EITC/ACTC participation models using 2000–2012 regressions may simply pick up the intensity of the Great Recession relative to limited, concurrent policy changes between 2000 and 2012. Finally, we conduct an exhaustive range of sensitivity checks on our main results, which can be found in the Online Resource 1.

Conclusion

Our findings lend support to the thesis that longer-term attachment to SNAP, EITC, and the ACTC is driven historically by a range of factors beyond short-run cyclicality, and that socioeconomically disadvantaged groups participate in these programs primarily as a result of structural economic factors—relatively stagnant wage growth for low-skilled workers and employment—as well as deliberate federal and state policies that improve the generosity and availability of benefits. Demographic change—notably the rise in female headship and the delay in marriage—is relatively less important. Still, demographics add to the demand for both programs and are consistently important for low-education families, for low-income families, and in models using parameters from the 2000s. At the same time, in the post-2000 period, the business cycle gained in importance relative to policy and demographics and was the dominant factor underlying program growth across family structures.

The lack of recent private sector wage growth in the United States is troubling, and work alone without SNAP, the EITC, and ACTC would leave low-income working families worse off. A host of recent studies have shown that refundable tax credits improve child outcomes in health, including birth outcomes for mothers and the learning of young children (Currie 2012; Dahl and Lochner 2012; Evans and Garthwaite 2014; Hoynes et al. 2016; Milligan and Stabile 2009). On the opposite side, the lack of resources for teen parents helps explain their accumulating developmental and health disparities (Mollborn et al. 2014). SNAP receipt while in childhood is increasingly shown to improve child health (e.g. Kreider et al. 2012) and outcomes as well as significantly reduce the incidence of metabolic syndrome (obesity, high blood pressure, and diabetes). For women, SNAP has been linked to an increase in economic self-sufficiency (Almond et al. 2011; Hoynes et al. 2016; Tiehen et al. 2015). Finally, higher income during childhood for low-income families has been associated with a range of positive socioeconomic and behavioral outcomes (Akee et al. 2015; Cooper and Stewart 2013; Duncan et al. 2011, 2014; Hardy 2014).

The biennial panel design of our study yields predictors of joint program participation that are less transitory than in annual cross-sectional analysis. This, in turn, has different policy implications and suggests more limited career mobility. Among the most important means-tested safety net programs are SNAP, EITC, ACTC, and TANF. With the decline in TANF, the focus on SNAP, EITC, and ACTC—three highly liquid programs—is unique within the literature. To the degree that wages remain stagnant and some workers remain marginally attached to the labor force, reliance and spending on these programs could remain elevated even in the presence of a more robust macroeconomic recovery. If so, the work-based safety net may become an even more permanent fixture for many of America’s vulnerable families, especially those with children. Indeed, outlays on SNAP, EITC, and ACTC are forecasted by the Congressional Budget Office (2015) to remain high—roughly $75 billion and $94 billion annually in nominal dollars between 2015 and 2025, respectively.17 Many states and local governments have enacted or expanded subnational EITC programs (Hardy et al. 2015) and have raised minimum wages. There are new proposals to expand child tax credits and to make the ACTC permanent and available to all as a child allowance regardless of work (Garfinkel et al. 2016; Ziliak 2014). Thus, continuing to assess the role of refundable tax credits and SNAP in the work-based safety net (Ben-Shalom et al. 2012; Sawhill and Karpilow 2014a, b) is important.

Acknowledgments

We thank the Smith Richardson Foundation for providing financial support. We also thank Susan Mayer, Dan Puskin, Darrick Hamilton, and Robin Lumsdaine, as well as participants at the 2013 APPAM conference, 2014 ASSA meetings, the 2014 Building Human Capital Conference, and seminar participants at Washington and Lee University for helpful feedback on earlier versions. We thank Robert Hartley for timely and valuable research assistance. Finally, this project was completed while Hardy served as the 2017–2018 Okun-Model fellow in Economic Studies at The Brookings Institution, and we acknowledge their support.

Notes

1

A recent study by Heflin et al. (2015) highlighted that although 35 states in 2014 had reportedly removed liquid asset tests, 28 of them still listed the test in their prescreening web-based tools for potential SNAP eligibility, which could have the effect of deterring some from applying.

2

Online Resource 1 details sample construction and characteristics of our CPS data set, including matching procedures. It also contains numerous specification checks on the baseline estimates presented in the main text.

3

The focus on federal EITC is informative for understanding state EITC participation and eligibility as well given that, by construction, almost all state EITC programs allocate refunds using federal rules and as a fixed proportion of federal EITC received (Johnson and Williams 2011).

4

With the 1994 survey, the CPS also asked about country of birth. Because the variable is available for only two-thirds of the sample period, we do not include it in our main analyses. We do, however, report the counterfactual simulations for 2000–2012 inclusive of nativity status in Tables S9 and S10 (Online Resource 1), with no substantive change in results reported in this article.

5

Because most research has found little evidence that marriage or fertility responds endogenously to the generosity of welfare benefits (Bitler et al. 2004; Hoynes 1997; Lopoo and Raissan 2014; Moffitt et al. 1998), the family size–specific EITC and SNAP parameters are treated as exogenous in the model.

6

The data on the state economic and policy environment are obtained from the University of Kentucky Center for Poverty Research, and the SNAP policy variables come from the Economic Research Service in the U.S. Department of Agriculture (http://www.ers.usda.gov/data-products/snap-policy-database.aspx#.UhQQ-ZLVC3I). All income and spending data are deflated by the 2010 Personal Consumption Expenditure Deflator (http://www.whitehouse.gov/sites/default/files/docs/erp2013/ERP2013_Appendix_B.pdf).

7

See http://www.bls.gov/lau/lauov.htm for details on construction of state unemployment rates.

8

For the minimum wage, we use the maximum of the state and federal minimum in each state and year; for the median wage, we compute the average annual hourly wage across the two years for each matched individual in the CPS and then compute the median in each state and year. We use the personal consumption expenditure deflator with 2010 base year to adjust for inflation.

9

A Wald test of the null hypothesis that the three unemployment rate coefficients are jointly zero is rejected at the p < .001 level.

10

Extending the notation in Eq. (1), for any given continuous regressor zk, the elasticity of participation in program yk (SNAP, EITC, and ACTC) equals δkzk¯yk¯, where zk¯ and yk¯ reflect mean values of the policy variable and dependent variable, respectively.

11

Outreach spending has an unexpected negative sign. Ziliak (2015b) found a similar result in the study of cross-sectional SNAP participation, attributing this to the federal response to SNAP during the Great Recession as the coefficient is the expected positive sign if the sample period stops in 2006.

12

Wald tests of the joint hypothesis that the three unemployment rate coefficients are 0 (zero) is rejected at the 0.07, 0.10, 0.002, and 0.08 level for all, low-income, low-skilled, and single-mother families, respectively.

13

In an independent analysis conducted concurrent to this initial draft of this paper, Bitler et al. (2014) used annual cross-sections of IRS Statistics of Income data and found that single-year EITC participation was acyclical for single-mother families and countercyclical among married-couple families.

14

The Wald test of the three coefficients on unemployment being jointly 0 is rejected at p < .001 level for all families, low-income, and low skilled; and at p < .034 for single mothers.

15

A limitation of the counterfactual simulation is that the observable and unobservable characteristics of the examined subgroups could, themselves, change over time.

16

Our baseline models require that the head of household remain the same across the two survey waves, which could depress the influence of demographic factors in our simulations. As a robustness check, we reestimated the models in Tables 1 and 3 that relaxed the constant headship requirement. We report the counterfactual simulations in Tables S7 and S8 (Online Resource 1). As shown in those tables, the results are little changed from the baseline models.

17

The forecasted spending on the EITC and ACTC was before the ARRA expansions of the ACTC, which were set to expire in 2017, were extended indefinitely as part of the federal budget agreement reached in December 2015 (Rachidi 2015).

References

Akee, R., Simeonova, E., Costello, E. J., & Copeland, W. (
2015
).
How does household income affect child personality and behaviors?
(NBER Working Paper No. 21562).
Cambridge, MA
:
National Bureau of Economic Research
.
Almond, D., Hoynes, H. W., & Schanzenbach, D. W. (
2011
).
Inside the war on poverty: The impact of food stamps on birth outcomes
.
Review of Economics and Statistics
,
93
,
387
403
.
Autor, D. H. (
2014
).
Skills, education, and the rise of earnings inequality among the “other 99 percent.”
Science
,
344
,
843
851
.
Autor, D. H., & Duggan, M. G. (
2007
).
Distinguishing income from substitution effects in disability insurance
.
American Economic Review
,
97
,
119
124
.
Autor, D. H., Katz, L. F., & Kearney, M. S. (
2008
).
Trends in U.S. wage inequality: Revising the revisionists
.
Review of Economics and Statistics
,
90
,
300
323
.
Ben-Shalom, Y., Moffitt, R., & Scholz, J. K. (
2012
).
An assessment of the effectiveness of antipoverty programs in the United States
. In P. N. Jefferson (Ed.),
Oxford handbook of the economics of poverty
(pp.
709
749
).
New York, NY
:
Oxford University Press
.
Bitler, M. P., & Hoynes, H. W. (
2010
).
The state of the social safety net in the post-welfare reform era
.
Brookings Papers on Economic Activity
,
2010
(
2
),
71
127
.
Bitler, M., & Hoynes, H. (
2016
).
The more things change the more they stay the same? The safety net and poverty in the Great Recession
.
Journal of Labor Economics
,
34
,
S403
S444
. 10.1086/683096.
Bitler, M., Hoynes, H., & Kuka, E. (
2014
).
Do in-work tax credits serve as a safety net?
(NBER Working Paper No. 19785).
Cambridge, MA
:
National Bureau of Economic Research
.
Bitler, M. P., Gelbach, J. B., Hoynes, H. W., & Zavodny, M. (
2004
).
The impact of welfare reform on marriage and divorce
.
Demography
,
41
,
213
236
.
Blank, R. M. (
2009
).
What we know, what we don’t know, and what we need to know about welfare reform
. In J. Ziliak (Ed.),
Welfare reform and its long-term consequences for America’s poor
(pp.
22
58
).
New York, NY
:
Cambridge University Press
.
Bloome, D., & Western, B. (
2011
).
Cohort change and racial differences in educational and income mobility
.
Social Forces
,
90
,
375
395
. 10.1093/sf/sor002.
Cancian, M., Han, E., & Noyes, J. L. (
2014
).
From multiple program participation to disconnection: Changing trajectories of TANF and SNAP beneficiaries in Wisconsin
.
Children and Youth Services Review
,
42
,
91
102
. 10.1016/j.childyouth.2014.04.003.
Cancian, M., & Reed, D. (
2009
).
Family structure, childbearing, and parental employment: Implications for the level and trend in poverty
. In M. Cancian & S. Danziger (Eds.),
Changing poverty, changing policies
(pp.
92
121
).
New York, NY
:
Russell Sage Foundation
.
Chetty, R., Friedman, J. N., & Saez, E. (
2013
).
Using differences in knowledge across neighborhoods to uncover the impacts of the EITC on earnings
.
American Economic Review
,
103
,
2683
2721
. 10.1257/aer.103.7.2683.
Congressional Budget Office (CBO)
. (
2013a
).
Growth in means-tested programs and tax credits for low-income households
(CBO Publication No. 4504).
Washington, DC
:
CBO
.
Congressional Budget Office (CBO)
. (
2013b
).
Refundable tax credits
(CBO Publication No. 4152).
Washington, DC
:
CBO
.
Congressional Budget Office (CBO)
. (
2015
).
Supplemental Nutrition Assistance Program: CBO baseline, January 2015
. Retrieved from https://www.cbo.gov/sites/default/files/recurringdata/51312-2015-01-snap.pdf
Cooper, K., & Stewart, K. (
2013
).
Does money affect children’s outcomes? A systematic review
(Report).
York, UK
:
Joseph Rowntree Foundation
. Retrieved from http://sticerd.lse.ac.uk/dps/case/cr/casereport80.pdf
Corcoran, M. (
2001
).
Mobility, persistence, and the consequences of poverty for children: Child and adult outcomes
. In S. H. Danziger, & R. H. Haveman (Eds.),
Understanding poverty
(pp.
127
161
).
Cambridge, MA
:
Russell Sage Foundation
.
Currie, J. M. (
2006
).
Welfare vs. “making work pay.”
In J. M. Currie,
The invisible safety net: Protecting the nation’s poor children and families
(pp.
11
31
).
Princeton, NJ
:
Princeton University Press
.
Currie, J. M. (
2012
).
Programs to address child poverty and their effects
. In P. N. Jefferson (Ed.),
Oxford handbook of the economics of poverty
(pp.
277
315
).
New York, NY
:
Oxford University Press
.
Dahl, G. B., & Lochner, L. (
2012
).
The impact of family income on child achievement: Evidence from the earned income tax credit
.
American Economic Review
,
102
,
1927
1956
. 10.1257/aer.102.5.1927.
Duncan, G. J., Magnuson, K., & Votruba-Drzal, E. (
2014
).
Boosting family income to promote child development
.
Future of Children
,
24
(
1
),
99
120
.
Duncan, G. J., Morris, P., & Rodrigues, C. (
2011
).
Does money really matter? Estimating impacts of family income on young children’s achievement with data from random-assignment experiments
.
Developmental Psychology
,
47
,
1263
1279
. 10.1037/a0023875.
Eissa, N., & Hoynes, H. (
2004
).
Taxes and the labor market participation of married couples: The earned income tax credit
.
Journal of Public Economics
,
88
,
1931
1958
. 10.1016/j.jpubeco.2003.09.005.
Evans, W. N., & Garthwaite, C. L. (
2014
).
Giving mom a break: The impact of higher EITC payments on maternal health
.
American Economic Journal: Economic Policy
,
6
(
2
),
258
290
.
Feng, S. (
2008
).
Longitudinal matching of recent current population surveys: Methods, non-matches and mismatches
.
Journal of Economic and Social Measurement
,
33
,
241
252
.
Fox, L., Wimer, C., Garfinkel, I., Kaushal, N., & Waldfogel, J. (
2015
).
Waging war on poverty: Poverty trends using a historical supplemental poverty measure
.
Journal of Policy Analysis and Management
,
34
,
567
592
. 10.1002/pam.21833.
Ganong, P., & Liebman, J. B. (
2013
).
The decline, rebound, and further rise in SNAP enrollment: Disentangling business cycle fluctuations and policy changes
(NBER Working Paper No. 19363).
Cambridge, MA
:
National Bureau of Economic Research
.
Garfinkel, I., Harris, D., Waldfogel, J., & Wimer, C. (
2016
).
Doing more for our children: Modeling a universal child allowance or more generous child tax credit
(Report).
New York, NY
:
The Century Foundation
. Retrieved from http://tcf.org/content/report/doing-more-for-our-children
Gundersen, C., & Ziliak, J. P. (
2004
).
Poverty and macroeconomic performance across space, race, and family structure
.
Demography
,
41
,
61
86
.
Hardy, B. L. (
2014
).
Childhood income volatility and adult outcomes
.
Demography
,
51
,
1641
1665
.
Hardy, B. L. (
2016
).
Income instability and the response of the safety net
.
Contemporary Economic Policy
,
35
,
312
330
.
Hardy, B. L., Muhammad, D., & Samudra, R. (
2015
).
The effect of the earned income tax credit in the District of Columbia on poverty and income dynamics
(Upjohn Institute Working Paper No. 15–230).
Kalamazoo, MI
:
W.E. Upjohn Institute for Employment Research
.
Heflin, C. M., Mueser, P., & Cronin, J. M. (
2015
).
The Supplemental Nutrition Assistance Program asset limit: Reports of its death may be exaggerated
(Working Paper No. 15–06).
Columbia
:
University of Missouri
.
Hotz, V. J., & Scholz, J. K. (
2003
).
The earned income tax credit
. In R. Moffitt (Ed.),
Means-tested transfer programs in the United States
(pp.
141
198
).
Chicago, IL
:
University of Chicago Press
.
Hoynes, H. W. (
1997
).
Work, welfare, and family structure: What have we learned?
In A. Auerbach (Ed.),
Fiscal policy: Lessons from economic research
(pp.
101
146
).
Cambridge, MA
:
MIT Press
.
Hoynes, H., & Schanzenbach, D. W. (
2016
).
U.S. food and nutrition programs
. In R. A. Moffitt (Ed.),
Economics of means-tested transfer programs in the United States
(Vol.
1
,
219
302
).
Chicago, IL
:
University of Chicago Press
.
Hoynes, H. W., Page, M. E., & Huff Stevens, A. (
2006
).
Poverty in America: Trends and explanations
.
Journal of Economic Perspectives
,
20
(
1
),
47
68
. 10.1257/089533006776526102.
Hoynes, H. W., & Schanzenbach, D. W. (
2009
).
Consumption responses to in-kind transfers: Evidence from the introduction of the Food Stamp Program
.
American Economic Journal: Applied Economics
,
1
(
4
),
109
139
.
Hoynes, H., Schanzenbach, D. W., & Almond, D. (
2016
).
Long-run impacts of childhood access to the safety net
.
American Economic Review
,
106
,
903
934
.
Johnson, N., & Williams, E. (
2011
).
A hand up: How state earned income tax credits help working families escape poverty in 2011
(Report).
Washington, DC
:
Center on Budget and Policy Priorities
.
Jones, M. R. (
2014
).
Changes in EITC eligibility and participation, 2005–2009
(CARRA Working Paper No. 2014–04).
Washington, DC
:
Center for Administrative Records Research and Applications, U.S. Census Bureau
.
Jones, M. R. (
2015
).
The EITC over the Great Recession: Who benefited?
(Report).
Washington, DC
:
U.S. Census Bureau
.
Kreider, B., Pepper, J. V., Gundersen, C., & Jolliffe, D. (
2012
).
Identifying the effects of SNAP (food stamps) on child health outcomes when participation is endogenous and misreported
.
Journal of the American Statistical Association
,
107
,
958
975
. 10.1080/01621459.2012.682828.
Larrimore, J., Burkhauser, R. V., & Armour, P. (
2013
).
Accounting for income changes over the Great Recession (2007–2010) relative to previous recessions: The importance of taxes and transfers
(NBER Working Paper No. 19699).
Cambridge, MA
:
National Bureau of Economic Research
.
Leftin, J., Eslami, E., & Strayer, M. (
2011
).
Trends in Supplemental Nutrition Assistance Program participation rates: Fiscal year 2002 to fiscal year 2009
(Mathematica Policy Research Report).
Washington, DC
:
U.S. Department of Agriculture, Food and Nutrition Service
.
Lopoo, L. M., & Raissan, K. M. (
2014
).
U.S. social policy and family complexity
.
Annals of the American Academy of Political and Social Science
,
654
,
213
230
.
Madrian, B. C., & Lefgren, L. J. (
1999
).
A note on longitudinally matching Current Population Survey (CPS) respondents
(NBER Working Paper No. 247).
Cambridge, MA
:
National Bureau of Economic Research
.
Meyer, B. D., Goerge, R. M., & Mittag, N. (
2014
).
Errors in survey reporting and imputation and their effects on estimates of Food Stamp Program participation
(CES Report No. 11-14).
Washington, DC
:
Center for Economic Studies, U.S. Census Bureau
.
Meyer, B. D., Mok, W. K. C., & Sullivan, J. X. (
2015
).
Household surveys in crisis
(NBER Working Paper No. 21399).
Cambridge, MA
:
National Bureau of Economic Research
.
Meyer, B. D., & Rosenbaum, D. T. (
2001
).
Welfare, the earned income tax credit, and the labor supply of single mothers
.
Quarterly Journal of Economics
,
116
,
1063
1114
.
Milligan, K., & Stabile, M. (
2009
).
Child benefits, maternal employment, and children’s health: Evidence from Canadian child benefit expansions
.
American Economic Review
,
99
(
2
),
128
132
.
Moffitt, R. A. (
2013
).
The Great Recession and the social safety net
.
Annals of the American Academy of Political and Social Science
,
650
,
143
166
.
Moffitt, R. A. (
2015a
).
Multiple program participation and the SNAP program
. In J. Bartfeld, C. Gundersen, T. M. Smeeding, & J. P. Ziliak (Eds.),
SNAP matters: How food stamps affect health and well being
(pp.
213
242
).
Stanford, CA
:
Stanford University Press
.
Moffitt, R. A. (
2015b
).
The deserving poor, the family, and the U.S. welfare system
.
Demography
,
52
,
729
749
.
Moffitt, R. A., Reville, R., & Winkler, A. E. (
1998
).
Beyond single mothers: Cohabitation and marriage in the AFDC program
.
Demography
,
35
,
259
278
.
Mollborn, S., Lawrence, E., James-Hawkins, L., & Fomby, P. (
2014
).
How resource dynamics explain accumulating developmental and health disparities for teen parents’ children
.
Demography
,
51
,
1199
1224
. 10.1007/s13524-014-0301-1.
Nichols, A., & Rothstein, J. (
2016
).
The Earned Income Tax Credit
. In R. Moffitt (Ed.),
Economics of means-tested transfer programs in the United States
(Vol.
1
, pp.
137
218
).
Chicago, IL
:
The University of Chicago Press
.
Piketty, T., & Saez, E. (
2003
).
Income inequality in the United States, 1913–1998
.
Quarterly Journal of Economics
,
118
,
1
39
. 10.1162/00335530360535135.
Rachidi, A. (
2015
,
December
16
).
Budget deal is good for low-income working families
[Blog post].
Washington, DC
:
American Enterprise Institute
. Retrieved from http://www.aei.org/publication/budget-deal-is-good-for-low-income-working-families/
Ratcliffe, C., McKernan, S.-M., & Finegold, K. (
2008
).
Effects of food stamp and TANF policies on food stamp receipt
.
Social Service Review
,
82
,
291
334
.
Sawhill, I. V., & Karpilow, Q. (
2014a
).
Raising the minimum wage and redesigning the EITC
(Center on Children and Families report).
Washington, DC
:
Brookings Institution
. Retrieved from http://www.brookings.edu/research/papers/2014/01/30-raising-minimum-wage-redesigning-eitc-sawhill
Sawhill, I.V., & Karpilow, Q. (
2014b
,
June
10
).
Building a better EITC
[Commentary post]. Retrieved from http://spotlightonpoverty.org/spotlight-exclusives/building-a-better-eitc/
Slack, K. S., Kim, B., Yang, M.-Y., & Berger, L. M. (
2014
).
The economic safety net for low-income families with children
.
Children and Youth Services Review
,
46
,
213
219
.
Steuerle, E. (
2015
,
June
25
).
Marginal tax rates and 21st century social welfare reform
(Statement before the Joint Hearing of the Subcommittee on Human Resources, Committee on Ways and Means, and Subcommittee on Nutrition, Committee on Agriculture).
Washington, DC
:
Urban Institute
.
Tach, L. M., & Eads, A. (
2015
).
Trends in the economic consequences of marital and cohabitation dissolution in the United States
.
Demography
,
52
,
401
432
.
Tiehen, L., Jolliffe, D., & Smeeding, T. M. (
2015
).
The effect of SNAP on poverty
. In J. Bartfeld, C. Gundersen, T. M. Smeeding, & J. P. Ziliak (Eds.),
SNAP matters: How food stamps affect health and well being
(pp.
49
73
).
Stanford, CA
:
Stanford University Press
.
U.S. Bureau of Labor Statistics
. (
2014
).
Employment situation—May 2014
[News release No. USDL–14–0987]. Retrieved from http://www.bls.gov/news.release/archives/empsit_06062014.pdf
Wheaton, L. (
2007
).
Underreporting of means-tested transfer programs in the CPS and SIPP
. In
Proceedings of the American Statistical Association, Social Statistics Section
[CD-ROM].
Alexandria, VA
:
American Statistical Association
.
Wooldridge, J. (
2007
).
Inverse probability weighted estimation for general missing data problems
.
Journal of Econometrics
,
141
,
1281
1301
. 10.1016/j.jeconom.2007.02.002.
Ziliak, J. P. (
2014
).
Supporting low-income workers through refundable child-care credits
. In M. Kearney, & B. Harris (Eds.),
Policies to address poverty in America
(pp.
109
118
).
Washington, DC
:
Brookings Institution Press
.
Ziliak, J. P. (
2015a
).
Recent developments in antipoverty policies in the United States
. In J. K. Scholz, H. Moon, & S.-H. Lee (Eds.),
Social policies in an age of austerity: A comparative analysis of the US and Korea
(pp.
235
262
).
Cheltenham, UK
:
Edward Elgar Publishing
.
Ziliak, J. P. (
2015b
).
Why are so many Americans on food stamps? The role of the economy, policy, and demographics
. In J. Bartfeld, C. Gundersen, T. M. Smeeding, & J. P. Ziliak (Eds.),
SNAP matters: How food stamps affect health and well being
(pp.
18
48
).
Stanford, CA
:
Stanford University Press
.
Ziliak, J. P., Hardy, B., & Bollinger, C. (
2011
).
Earnings volatility in America: Evidence from matched CPS
.
Labour Economics
,
18
,
742
754
. 10.1016/j.labeco.2011.06.015.

Supplementary data