The Effect of the US Safety Net on Material Hardship over Two Decades

Abstract

We address a question at the center of many policy debates: how effective is the US safety net? Many existing studies evaluate the effect of one program on economic hardship in isolation, though families typically participate in multiple programs. Using 1992–2011 data from the Survey of Income and Program Participation, our analyses examine the simultaneous effect of participation in three programs, TANF, SNAP, or Medicaid/SCHIP, on a set of outcomes of intrinsic importance—measures of material hardship. We find that a 10 percentage point increase in participation in any of these three safety net programs by low-to-moderate income families with children reduces their average number of hardships by 0.11 (−0.41 elasticity), and the incidence of food insufficiency by 1.7 percentage points (−1.27 elasticity). This analysis suggests that hardship would be even more prevalent in the United States without the existence of the current safety net programs.

1. Introduction

Over the past 85 years, the United States has developed and modified a series of programs designed to support families as they strive to achieve economic well-being. These have taken the form of both income support programs, which provide cash or in-kind support, and work incentives or work support programs such as tax credits. The income support programs are sometimes referred to as a social “safety net” established to ensure that families that meet the eligibility criteria do not fall below a certain level of well-being. The Federal government spends billions of dollars on these programs annually, which of course raises the question of just how effective these programs are in reducing poverty and hardship among the US population.

As of 2019, over 34 million Americans had incomes below the federal poverty line, representing about 10.5 percent of the US population (Census Bureau, 2020). Nearly 100 million Americans—47.5 million of them living in a working family with children in 2011—had incomes below 200 percent of the federal poverty line (Kaiser Family Foundation 2016; Population Reference Bureau 2013). Many see this widespread poverty and hardship as evidence that government programs are indeed ineffective. In 1988, Ronald Reagan jested, “My friends, some years ago, the federal government declared war on poverty—and poverty won.” But is this sentiment accurate?

The goal of our study is to analyze the effect of three safety net programs on reducing material hardship for families in the United States. More specifically, our primary research question is: How does participation in Temporary Assistance to Need Families (TANF), the Supplementary Nutrition Assistance Program (SNAP), or public health insurance (e.g., Medicaid, State Children’s Health Insurance Program [SCHIP]) affect the material hardship of low-to-moderate income families with children?¹

We answer this question using monthly data from the US Census Bureau’s Survey of Income and Program Participation (SIPP) from 1992 to 2011. We restrict our analysis to families with income below 400 percent of the poverty threshold (which we define as low-to-moderate income families). We supplement the SIPP data with state-level economic and policy data and control for other state and local safety net policies, such as the earned income tax credit (EITC) and minimum wage policies.

To answer our research question, we use an instrumental variables (IV) approach that uses the variation in state policy rules to identify the effect of program participation on material hardship. Our participation measure captures participation in any of the three programs (TANF, SNAP, or Medicaid/SCHIP) because families use multiple programs (as detailed below) and the programs likely have an interactive effect that makes them difficult to disentangle. The two instruments we use are the total SNAP outreach spending per person with income less than 150 percent of the federal poverty threshold and a simulated share of children eligible for public health insurance, given the state restrictions in place in a given year and month. We show that these instruments have high predictive power for a family’s participation in safety net programs and are not correlated with the state’s past poverty rate, general revenue, or political environment.

Our results indicate that a 10 percentage point increase in the share of families receiving TANF, SNAP, or public health insurance reduces the average number of hardships low-to-moderate income families with children experienced by 0.114, or an elasticity of −0.41. The number of hardships includes a general report of not meeting essential expenses, inability to pay rent or mortgage, eviction, unpaid utility bills, utility service cut, phone service cut, unmet medical or dental need, and food insufficiency. A 10 percentage point increase in participating in any of these three safety net programs also reduces food insufficiency by 1.66 percentage points among low-to-moderate income families with children, or an elasticity of −1.27. This effect is large but not inconsistent with earlier studies (Kreider et al. 2012; Ratcliffe, McKernan, and Zhang 2011; Shaefer and Gutierrez 2013; Schmidt, Shore-Sheppard, and Watson 2016). Finally, we find a negative but imprecise effect of program participation on unmet medical or dental need.

We build on the literature measuring the effects of safety net programs on material hardship in at least two important ways. First, unlike many previous studies, we examine the simultaneous effect of multiple programs on material hardship. One important concern with studies that measure the effect of only one program on outcomes is that the effect of participation in one program, such as TANF, will reflect participation in other programs, such as Medicaid, if people participate in multiple programs. Safety net programs also often use information they collect to help enroll people with low-income for other benefits for which they qualify (Ambegaokar, Neuberger, and Rosenbaum 2017). The only important exception is Schmidt, Shore-Sheppard, and Watson (2016), which also accounts for the interactions of multiple safety net programs and examines their effect on food insecurity.² However, while Schmidt, Shore-Sheppard, and Watson (2016) estimate the effect of potential benefits of safety net programs, our study estimates the effect of actual program participation on hardship.

Second, we focus on the effect of these programs on a broad measure of material hardship: the number of hardships experienced—as well as food insufficiency and unmet medical need. There is growing interest in measuring and understanding how people experience material hardship above and beyond income poverty.³ Hardship indicators represent direct measures of well-being, and, as such are of intrinsic importance to people. This contrasts with income, which is thought to be mainly instrumentally important for attaining other ends—such as avoiding the material hardships we measure directly in this study. While the literature has focused on the effect of safety-net participation on food insecurity (Kreider et al. 2012; Ratcliffe, McKernan, and Zhang 2011; Shaefer and Gutierrez 2013; Schmidt, Shore-Sheppard, and Watson 2016), we focus on a more general hardship measure, which includes a general report of not meeting essential expenses, inability to pay rent or mortgage, eviction, unpaid utility bills, utility service cut, phone service cut, unmet medical or dental need, and food insufficiency.

The remainder of the paper lays out the background for our study and literature, our conceptual framework and hypotheses, empirical model, data and sample, results, and conclusion.

2. Background and Literature

The design and implementation of income-tested programs have changed dramatically since the early 1990s, with important and significant changes at the federal and state levels. A primary example is the transition from the Aid to Families with Dependent Children (AFDC) program to the TANF program, which occurred with the passage of the Personal Responsibility and Work Opportunity Reconciliation Act in 1996. This legislation, along with welfare waivers from the federal government to states in the early to mid-1990s, provided states with flexibility in designing their welfare programs that resulted in significant variation in state programs. Starting in the mid- to late-1990s, the federal government also began giving states flexibility to change SNAP rules. These changes culminated in the Farm Security and Rural Investment Act of 2002 (the Farm Bill), which provided even greater flexibility to states to set their own SNAP rules. Additionally, there have been changes to health care for lower income families, such as the introduction of SCHIP in 1997 and health insurance expansions under the 2010 Affordable Care Act. All of these changes led to increased state-level variation in the rules and policies governing income-tested programs.

Other programs and policies aimed at supporting low- to-moderate income families have changed over time. The federal EITC is a refundable income tax credit, which reduces a person’s tax liability and allows refunds in excess of the income tax liability. Thus, a refundable credit can create an incentive to work even for very low-income families that have little or no tax liability. There have been large expansions to the federal EITC since the early 1990s. Some states supplement the federal EITC with a state EITC, and some of these state EITC benefits are also refundable. Another policy strategy used to boost the earnings of low-wage workers was the establishment of a federal minimum wage as well as a minimum wage in some states. Although the minimum wage has been increased periodically, its real value has eroded when it has remained unchanged over time.

A considerable literature exists on the effects of safety net programs on poverty—which indicates substantial effects of these programs on reducing poverty, though there is variability across programs considered.⁴ For example, Ben-Shalom, Moffitt, and Scholz (2012) find that the combination of the means-tested (such as TANF and SNAP) and social insurance programs (such as unemployment insurance and social security retirement) have reduced deep poverty, poverty, and near-poverty rates by about 14 percentage points in the US population. Similarly, Fox, Wimer, Garfinkel, Kaushal, and Waldfogel (2015) estimate that government antipoverty programs cuts poverty rates in the US nearly in half (from 31 percent to 16 percent). The paper also highlights the particularly crucial role for the EITC and food and nutrition programs, especially in the modern era. Finally, Hoynes and Patel (2018), focus on the EITC alone and find that a $1,000 EITC increase reduces poverty more than 8 percentage points among single parent families, after allowing for both direct and behavioral effects (e.g., increased earnings).

The literature on the effects of safety net programs on material hardship also generally indicates that these programs reduce hardship. The most consistent finding is that SNAP reduces food insecurity. Among the studies on this issue, Pilkauskas, Currie, and Garfinkel (2012), using data from the Fragile Families and Child Wellbeing Study, find food hardship during the Great Recession might have increased by twice the amount observed if not for SNAP. Nord and Golla (2009), using Current Population Survey data, likewise find that SNAP reduced the prevalence of very low food security among recent program entrants by about one-third from 2001 to 2006. Among studies that have used SIPP data, Mills and Mykerezi (2010), Ratcliffe, McKernan, and Zhang (2011), and Schaefer and Gutierrez (2013) all find that food stamps/SNAP substantially reduce food insecurity (see also Kreider et al. 2012, Bitler and Hoynes 2016a, Gundersen, Kreider, and Pepper 2017).

Related literature provides some evidence consistent with the hypothesis that safety net programs indirectly or directly reduce unmet medical need. Flores et al. (2017) find descriptive evidence in a prospective observational study of 237 children that eligible children obtaining Medicaid/CHIP coverage were significantly less likely to have unmet medical (13 percent) or dental (18 percent) needs than eligible uninsured children (48 percent unmet medical and 62 percent unmet dental). Wang, Norton, and Rozier’s (2007) results indicate that SCHIP/Medicaid-insured children were 8 percentage points less likely to report unmet dental need compared with uninsured children, based on an instrumental variables analysis using the 1997–2002 National Health Interview Survey. McManus, Chi, and Carle (2015), however, find no association between Medicaid eligibility criteria and unmet preventative dental care need among children ages 3 to 17, using the 2009–10 National Survey of Children with Special Health Care Needs. Hu et al. (2018) find that the Medicaid expansions significantly reduced the number of unpaid bills and the amount of debt sent to third-party collection agencies.

One important concern with studies that measure the effect of only one program on outcomes is that the effect of participation in one program (e.g., TANF or SNAP) will reflect participation in other programs (e.g., Medicaid/SCHIP) if participants are likely to participate in multiple programs. As described later, in our sample we find that about 99 percent of families receiving TANF and 89 percent receiving SNAP also participate in Medicaid/SCHIP. This result is not surprising given how safety-net programs interact with each other. In fact, Congress has provided options for safety net programs to use information collected by one public benefit program to help enroll low-income people for other benefits for which they qualify (Ambegaokar, Neuberger, and Rosenbaum 2017). To address the concern that the effect of participation in one program reflects participation in other programs, our study measures participation in any of the three programs.

Our review of the literature reveals that no study has directly examined how multiple income-tested programs have affected the total number of hardships experienced or instances of unmet medical or dental need, which, along with food insufficiency, are at the center of this study. Schmidt, Shore-Sheppard, and Watson (2016), have come the closest, highlighting the importance of jointly considering a full range of safety net programs in their study of the effect of potential benefits of safety net programs on food insecurity. Using Current Population Survey data from 2001–09 and an instrumental variables framework, they measure the effect of being eligible for TANF, Supplemental Security Income, EITC, food assistance, and Medicaid and find that eligibility for $1,000 in potential benefits reduces the incidence of food insecurity by 1.1 percentage points among nonimmigrant, low-income, single-parent families. We build on this literature by: i) using a direct measure of participation (instead of eligibility); ii) adding a more general measure of nine hardships as well as unmet medical or dental need; and iii) exploiting variation in two state policy instrumental variables over two decades (1992–2011) to identify our program impacts.

3. Conceptual Framework: Determinants of Material Hardship

At the macro level, material hardship (like poverty) is shaped by the economy’s performance, demographic factors, and social programs (Danziger and Gottschalk 1995; Iceland 2013). With regards to economic performance, economic growth (often measured in terms of GDP growth per capita) drives changes in average standards of living, and economic inequality affects the distribution of income. Demographic changes—the growth in single-parent families in particular—have also been thought to slow progress against poverty since the 1960s (e.g., Cancian and Reed 2009, Eggebeen and Lichter 1991). State and year fixed effects and economic variables are included in our empirical model, as discussed below, to control for macro-level variables.

At the micro level, several factors affect whether households experience hardship. Material hardship is a function of earned income, public and private transfers, and family composition, all of which are potentially endogenous variables. Because our primary focus is on the role that the social safety net and, thus, public transfers play in material hardship, we model material hardship as a function of program participation and the reduced-form determinants of earned income, private transfers, and family composition. The reduced-form determinants and their hypothesized effects are based on human capital theory (Becker, 1975) and theory of the demand for children Becker (1991) and are derived in detail in Iceland (2013) and McKernan and Ratcliffe (2005). The reduced-form determinants provide the control variables for our model and include age, race and ethnicity, gender, educational attainment, US citizenship, household structure, and metropolitan status.

The focus of this paper, however, is on income-tested program participation. Income-tested programs are hypothesized to reduce material hardship, as they are designed to provide families with additional resources to meet their basic needs. TANF provides cash and in-kind benefits to recipients, SNAP provides resources to buy food, and public health programs such as Medicaid and SCHIP help families meet medical needs and expenses.

While these income-tested programs likely directly reduce material hardship, some of the benefits could be offset through indirect effects. Previous research has generally found some small disincentive effects of welfare benefits on work (Ben-Shalom, Moffitt, and Scholz 2012). Beyond this, an analysis of the Food Stamp Program in the 1960s and early 1970s finds that the program’s introduction reduced annual hours worked and employment among single mothers by 183 annual hours (intent-to-treat estimate), 505 annual hours (treatment-on-the-treated), and 24–27 percentage points, respectively. The authors found no significant impacts on earnings, family income, or the overall sample (Hoynes and Schanzenbach 2012). Today’s more work-focused safety net (for example, the EITC encourages work) has smaller disincentives than the safety net analyzed by earlier studies (Ben-Shalom, Moffitt, and Scholz 2012). In fact, a recent study finds strong work incentives associated with EITC. Hoynes and Patel (2018) find ignoring the EITC’s behaviorally induced indirect earning effect, underestimates its antipoverty effects on single parent families by up to 50 percent.

Overall, we hypothesize that participation in TANF, SNAP, or public health insurance will reduce our material hardship measures: the number of hardships of any type experienced, food insufficiency, and unmet medical or dental need. SNAP and Medicaid/SCHIP are focused on specific types of hardship (food insufficiency and unmet medical need, respectively). However, income is fungible. Even though SNAP receipt might not directly provide funds to pay medical bills, it could still reduce unmet medical need by freeing up income that would ordinarily be used to buy food. Increases in SNAP and Medicaid/SCHIP receipt could free up cash, which could then be used to reduce other forms of hardship, such as those associated with an inability to pay rent, gas, or phone bills.

4. Empirical Model

Estimating the relationship between material hardship and program participation is complicated by the observed and unobserved differences between participants and non-participants. First, families rely more on safety net programs when they experience negative economic shocks (such as a family member being laid-off or getting sick). In addition, participants and non-participants differ on factors such as other social policies they face (e.g., state minimum wage), economic conditions (e.g., unemployment rate), unobservable individual characteristics (e.g., taste for social programs or distaste for work), and unobservable state characteristics (e.g., public sentiment toward social program participants). Measuring the causal relationship between material hardship and program participation requires disentangling the effect of program participation on material hardship from these other factors.

To address the endogeneity of program participation, we use a two-stage least squares variable model with instrumental variables. Our empirical model is specified as follows:

$${\text{Y}}_{\text{ist}}={\beta }_1{\text{BenefitReceipt}}_{\text{ist}}^{*}+X_{\text{ist}}{\gamma }_1+S_{\text{st}}{\gamma }_2+{\mu }_s+{\eta }_t+{\epsilon }_{\text{ist}}$$ (1)

where the dependent variable Y_ist is material hardship for household i in state s in month t. Material hardship is measured by our three dependent variables: number of hardships (0–9), food insufficiency (0/1), and unmet medical or dental need (0/1). ${\text{BenefitReceipt}}_{\text{ist}}^{*}$ is the likelihood that the household participates in any of the three safety net programs in period t predicted in our first stage (equation (2) below). The remaining explanatory variables in the equations are drawn from the conceptual framework. X_ist is a vector of variables controlling household-level characteristics. This includes the following characteristics of the head of the household: age, age squared, race indicators (black non-Hispanic, Hispanic, other non-Hispanic), education indicators (less than high school education, high school diploma only, associate’s degree only). It also includes characteristics of the household: indicator for single male-headed household, indicator for single female-headed household, indicator for whether some adults in the household are not US citizens, indicator for whether no adults in the household are US citizens, indicator for metropolitan area, number of adults in household, number of children in household. To control for changes in the economy, we include a vector of variables (S_st) that represent the state economic conditions (federal and state EITC, regular minimum wage, subminimum wage, state unemployment rate, state per capita income, state employment-population ratio, and quarterly GDP).⁵ Finally, μ_s is the state-fixed effect controlling for time-invariant unobservable heterogeneity (differences) across states, η_t is the year-fixed effect controlling for unobservable heterogeneity across years, and ε_ist is the error term.

The first-stage equation is as follows:

$${\text{BenefitReceipt}}_{\text{ist}}={\text{Z}}_{\text{st}}{\delta }_1+X_{\text{ist}}{\delta }_2+S_{\text{st}}{\delta }_3+{\mu }_s+{\eta }_t+e_{\text{ist}}$$ (2)

where Z_st is the vector of state policy instruments described below. The coefficient of interest in the analysis is (β₁), which captures the total effect of receiving TANF, SNAP, or public health insurance among those households directly impacted by the changes in the policy instruments (local average treatment effect). In other words, it identifies the effect of program participation among those households at the margin of receiving the program who could be affected by expansion or retraction of the generosity of these safety net programs.

Our analysis includes the effective F statistic from Olea and Pflueger (2013) for all IV regressions as well as the weak instrument threshold for the case of 10 percent potential bias (τ = 10 percent) and 5 percent level of significance. We weight the models using SIPP weights to help account for attrition, nonresponse, and complex sample design. We cluster the standard errors at the state level in all models.

Instrumental Variables for Program Participation

Our IV estimates are only as valid as the instruments used to generate them. The ideal instruments will be (1) strongly correlated with program participation and (2) not otherwise related to our outcomes, given the additional covariates for which we control. To find valid instruments, we explore state policy variables related to Medicaid/SCHIP, SNAP, and TANF. We use the variation across states and the timing of different state policies to identify the model. As we discuss below, many of these policy instruments have been shown in the literature to predict participation in their respective program. Our final instruments are drawn from an extended set of policy instruments that statistically significantly predict program participation in our sample (discussed in section 6). In addition, we provide evidence that our policy instruments are not determined by past trends in poverty rates, the state fiscal, or the political environment measures in the state (Table A1 discussed below). Our instruments are:

• SNAP outreach spending per person with income less than 150 percent of the federal poverty line, and

• Simulated share of children eligible for public health insurance.

Higher SNAP outreach spending is hypothesized to increase participation via an increase in the number of SNAP applicants (because of increased knowledge about SNAP) and SNAP recipients (because some applicants will be eligible for the program). There is evidence from the literature that outreach spending has good predictive power in the SNAP receipt equation (Ratcliffe, McKernan, and Zhang 2011).

More lenient Medicaid and SCHIP eligibility thresholds are hypothesized to increase coverage and reduce hardship. Using eligibility thresholds, we construct a simulated share of children eligible for Medicaid and SCHIP. The literature has shown that such simulated shares are a strong predictor of Medicaid and SCHIP receipt (Currie and Gruber 1996; Cutler and Gruber 1996; Gruber and Simon 2008; Miller and Wherry 2019). In the construction of our simulated share of children eligible for public health insurance, we follow the same approach used in Miller and Wherry (2019), which is derived from earlier work from Gruber and co-authors (Currie and Gruber 1996, Cutler and Gruber 1996, Gruber and Simon 2008). We begin by using income eligibility by child’s age for each state from 1992 to 2011. We then apply this state eligibility rules to a national sample of children for each year in our data. Using data from all states in the SIPP (without any income restriction), we estimate the share of children in this nationally representative sample at a given age that would be eligible for Medicaid if their families lived in each state of the country. We defined this variable as the “simulated share of children eligible for public health insurance” for each state, year, and child’s age. Finally, each household in our sample is matched to the “simulated share of children eligible for public health insurance” variable using their state of residence, year, and age of the youngest child in the household.⁶ More lenient Medicaid and SCHIP eligibility thresholds should translate to higher simulated share of children eligible for public health insurance, which are hypothesized to increase coverage and reduce hardship. This simulated eligibility instrument isolates changes in Medicaid policy by netting out any changes in state demographic or economic characteristics that influence state-level eligibility.

We also hypothesize that our Medicaid/SCHIP policy instrument and SNAP instrument impacts AFDC/TANF participation. Prior to 1996, Medicaid eligibility was automatic for AFDC recipients (Maloy et al. 1998). In addition, safety net programs use information collected by one public benefit program to help enroll low-income people for other benefits for which they qualify (Ambegaokar, Neuberger, and Rosenbaum 2017). As a result, an expansion of SNAP outreach spending and Medicaid eligibility thresholds could have a direct impact on AFDC/TANF enrollment.

The two policy instruments are hypothesized to affect material hardship only through their effect on benefit receipt. However, a concern remains if state policies depend on earlier economic outcomes; for example, if policymakers choose policies to address past material hardship or poverty trends. It is also possible that policymakers only expand their safety net program when their states are in favorable fiscal situations. Finally, states might implement more generous safety net policies in a more liberal political environment. We try to address these concerns by looking at the relationship between our two instruments measured at the state level and lagged political and economic characteristics of the state (Table A1). We test whether the instruments are associated with the political affiliation of the governor, state poverty rate, and general state revenue (all lagged by one year to reflect policy decision making) for the years 1993–2011.⁷ We do not find evidence that any of our two instruments are associated with these political and economic conditions of the state, which provide us assurance that our exclusion restrictions are satisfied.

5. Data, Measures, and Sample

To answer our research question and test the hypotheses described above, we use both household- and state-level data. The household-level data are from the Survey of Income and Program Participation (SIPP), a longitudinal dataset that follows households over time. These data are augmented with information on state-level economic and social program policies and rules from multiple sources, including the US Department of Agriculture’s SNAP Policy database and the TRIM3 model. This section also describes the hardship measures—the key outcomes—and the study sample.

Household-Level SIPP Data

Each SIPP panel contains a nationally representative (noninstitutional) sample of households whose members are interviewed at four-month intervals over an approximately two- to four-year period. The panels have sample sizes ranging from approximately 14,000 to 52,000 households. In addition to collecting monthly data, the SIPP includes “topical modules” that ask periodic questions about topics such as material hardship, child care, and wealth. The SIPP includes other economic and demographic characteristics, including age, race and ethnicity, gender, citizenship, educational attainment of the head of the household, and whether the household lives in a metropolitan area. We analyze the 1991, 1992, 1993, 1996, 2001, 2004, and 2008 SIPP panels.

The material hardship measures are based on data collected in the SIPP’s well-being topical module.⁸ The well-being topical modules consist of a series of questions on food insufficiency, ability to meet basic needs such as going to the doctor or dentist, housing problems, and other topics. With the exception of the 2008 panel, this module is administered once over the course of each panel, so we only capture well-being in select years over this period (1992, 1995, 1998, 2003, 2005, 2010, and 2011).⁹ Over these 20 years, TANF, SNAP, Medicaid/SCHIP policies were changing significantly. This period also captures both strong and weak economies.

The SIPP also provides monthly data on program participation in TANF, SNAP, or Medicaid/SCHIP—our key explanatory variable. It is well established that there is underreporting of benefit receipt in survey data, and the SIPP is no exception, although studies have found less underreporting in the SIPP than in other surveys (Bitler, Currie, and Scholz 2003; Cody and Tuttle 2002). A recent study of SNAP found that for the (24-month) period from 2009 to 2010, the SIPP captured 92 percent of SNAP receipt (Ratcliffe et al. 2016). While the SIPP presents less underreporting than in other surveys, underreporting in the SIPP is not negligible and could cause some bias in our estimate. To address this issue, we present robustness checks for our findings where we adjust program participation using a modified version of the Scholz, Moffitt and Cowan (2009) (SMC) method as described by Mittag (2019).

The SIPP also suffers from seam bias, where transitions are more likely to occur between interview waves than between months within the same interview wave. To address this seam issue, we use only the interview month (which reflects data from the preceding month) from each wave in our analysis. We conduct sensitivity analyses using alternative measures of participation described below.

In our preferred specification, all explanatory variables (including program participation) and instruments are from the interview month. In terms of the material hardship variables, as discussed below, the number of hardships and unmet medical or dental need refers to the last 12 months. As robustness checks, we estimate models where program participation is measured as (1) any program participation in the past 12 months and (2) average number of months of program participation in the past 12 months. Results, discussed below, show that our main findings are generally robust to using these alternative participation measures.

Material Hardship Measures

We examine three material hardship measures: (1) the total number of hardships experienced, (2) food insufficiency, and (3) unmet medical need. The total number of hardships experienced captures multiple components of hardship, and as such, it is a broader measure of well-being than our other two measures. Food insufficiency and unmet medical need are often considered outcomes of “intrinsic” importance, as they represent the inability to meet basic capabilities, or needs (Heflin 2017; Sen 1999). These two specific measures represent different components of well-being, and while they may be positively correlated with each other, they have different causes and arise under different social and individual circumstances. For example, food insufficiency may result from a short-term income shortfall, such as a job loss, while unmet medical need might result from a catastrophic health crisis not easily addressed with just a little extra money and is affected by other constraints, such as the availability and affordability of insurance in an area (Heflin, Sandberg, and Rafail 2009).

Our food insufficiency measure is based on a single SIPP question that asks respondents which of the statements “best describe the food eaten in your household:” (1) enough of the kinds of food we want; (2) enough but not always the kinds of food we want to eat; (3) sometimes not enough to eat; (4) often not enough to eat. We classify households as having insufficient food if a household member responds “sometimes not enough to eat” or “often not enough to eat.”¹⁰ We focus on food insufficiency instead of food insecurity because food insecurity is not available in the SIPP data before 1998.^11,12

We classify a household as having unmet medical or dental need if any household member reports needing to visit a doctor or a dentist in the past 12 months but did not go because of insufficient resources. Finally, our broader hardship measure uses information from the above items as well as additional ones. We count the number of times in the past 12 months a household faced any of the following hardships: a general report of not meeting essential expenses, inability to pay rent or mortgage, eviction, unpaid utility bills, utility service cut, phone service cut, unmet medical or dental need, and food insufficiency.¹³ See Appendix B for the survey questions. Our indicator of material hardship is the sum of nine dichotomous indicators, with each component receiving equal weight. While this approach is consistent with past work using the SIPP (Federman et al. 1996; Beverly 2001), recent work has proposed theoretically robust multidimensional aggregation methods that account for multidimensionality (Alkire and Foster 2011; Alkire et al. 2015).

State-Level Policy and Economic Data

The state-specific program policies and economic characteristics come from multiple sources and are available from 1991 through 2011. State’s SNAP outreach spending comes from the US Department of Agriculture’s (USDA) SNAP Policy Database and the USDA Food and Nutrition Service’s (FNS) National Data Bank. The SNAP Policy Database documents state SNAP program rules for each month from January 1996 through December 2011. SNAP outreach spending prior to January 1996 comes from the FNS National Data Bank.¹⁴

Medicaid and SCHIP program rules by child’s age come from multiple sources, including the TRIM3 model.¹⁵ Program rules are available by state dating back to 1988. TRIM3 provides details about states’ eligibility rules for adults and children as well as information on reporting periods and interactions with other programs (e.g., automatic Medicaid eligibility for AFDC/TANF recipients). In addition, the annual Maternal and Child Health updates from the National Governors Association provide information on age limits for Medicaid-covered children. As described above, following the same approach used in Miller and Wherry (2019), which derived from earlier work from Gruber and co-authors (Currie and Gruber 1996; Cutler and Gruber 1996; Gruber and Simon 2008), these variables are combined to create a simulated share of children eligible for public health insurance, given the state restrictions in place in a given year and month.

Our analysis also controls for the EITC, minimum wage, and state economic characteristics. The EITC measure, which comes from the Urban-Brookings Tax Policy Center, is the maximum refundable credit available for a family with two children (federal and state combined). The minimum wage, which comes from the US Department of Labor and other sources, is captured by two separate variables: the minimum wage for jobs covered by the Fair Labor Standards Act (the “regular” minimum wage) and the wage for those not covered (the “subminimum” wage). Workers in jobs not covered by the regular minimum wage include those in small businesses, in businesses not involving interstate commerce, in seasonal or recreational jobs, and in fishing operations, as well as executive, administrative, and professional employees.¹⁶ We measure state-level economic conditions—unemployment rates, employment-population ratio, annual per capita income—and quarterly GDP with data from US Bureau of Labor Statistics Local Area Unemployment Statistics and US Bureau of Economic Analysis.

Study Population and Descriptive Statistics

Our study population is low-to-moderate income households with children, defined as households with income below 400 percent of the official poverty threshold. This level of household income is higher than Medicaid/SCHIP eligibility cut-offs—some states allow children in families with income up to 300 percent of the federal poverty guideline to enroll. We choose this broader study population—400 percent of the federal poverty threshold—because there is concern that income could be endogenous to the generosity of safety net programs for low-income families.¹⁷

Defining the study population too narrowly can exclude individuals who are at the margin of eligibility and who can alter their behavior to become eligible for benefits but choose not to do so.¹⁸ To test for the endogeneity of income, we test whether variation in our policy instruments can predict the likelihood that families with children in the SIPP are below different poverty thresholds (Table A2). In fact, we find that more lenient Medicaid and SCHIP eligibility thresholds are associated with an increase in the share of households below 200–350 percent of the poverty threshold. However, we do not find evidence that these state-level variables predict the share of households under 400 percent of the poverty threshold in the state (F-test for joint significance of instrument equals 2.28). Finally, we find that among households excluded from our sample (those above 400 percent of the poverty threshold) less than 1 percent report receiving TANF, about 1 percent report receiving SNAP, and about 6 percent report receiving public health insurance (not shown).¹⁹

How prevalent is material hardship among low-to-moderate income families with children in the United States? From 1992 to 2011, 4.5 percent experienced food insufficiency and 16.8 percent reported unmet medical or dental needs. On average low-to-moderate income families with children experienced 0.97 hardships during the past 12 months (Table 1 panel A).

Table 1 –

Summary Statistics

Panel A
	Mean	Std. Deviation
Dependent Variables
Number of Hardship	0.97	1.60
Had Food Insufficiency	4.5%	20.7%
Unmet Medical or Dental Need	16.8%	37.4%
Benefit Receipt
HH Receives TANF, SNAP or Public Health Insurance	34.4%	47.5%
HH Receives TANF	7.3%	26.1%
HH Receives SNAP	18.9%	39.2%
HH Receives Public Health Insurance	32.3%	46.8%
Demographic Characteristics (Head of the Household)
Age	38.6	10.0
Race
Black non-Hispanic	16.7%	37.3%
Hispanic	13.6%	34.2%
Other non-Hispanic non-white	5.3%	22.4%
Education
Less than High School Education	15.0%	35.7%
High School Diploma Only	31.2%	46.3%
Associated Degree / two years college education	27.6%	44.7%

Panel B
	Percent that also Receives…
From HH that Receives	TANF	SNAP	Public Health Insurance
TANF	-	88.1%	98.8%
SNAP	34.2%	-	89.2%
Public Health Insurance	22.4%	52.1%	-

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: The sample is restricted to households with income below 400 percent of the federal poverty threshold at the month of interview included in the hardship regression in Table 3.

Turning to monthly benefit receipt, on average, 34.4 percent of low-to-moderate income families with children received assistance from at least one of the three programs (TANF, SNAP, or public health insurance) during our study period from 1992 to 2011. The greatest share received public health insurance (32.3 percent) followed by SNAP (18.9 percent), while only 7.3 percent received TANF (Table 1 panel A). Table 1 panel B provides detail on the combinations of benefit receipt. Specifically, 98.8 percent of TANF participant families in our sample also receive public health insurance, and 88.1 percent receive SNAP during our period of analysis. This result is not surprising given the less restrictive income eligibility limits for TANF than Medicaid/SCHIP (GAO 2017). Further, 89.2 percent of SNAP recipients also receive public health insurance.²⁰ This result highlights the importance of combining participation in any of the three programs (TANF, SNAP, or Medicaid/SCHIP) in our estimation as the programs likely have an interactive effect on families that are difficult to disentangle.

6. Results

First Stage

We first investigate the effect of each policy instrument on participation in TANF, SNAP, or Medicaid/SCHIP (Table 2). This estimation corresponds to the first stage of the two-stage least squares regression on number of material hardships presented in Table 3. The goal is to investigate whether our policy instruments are strong predictors of program participation. Both instruments have a positive effect on overall participation, although only the simulated share of children eligible for public health insurance is estimated with precision (Table 2, column 1). As we discuss in the next section, using the Olea and Pflueger effective F-stat we can reject the null hypothesis of weak instruments in this first stage.

Table 2 –

Fist Stage

	First Stage		Safety Net Participation
Dependent Variable	HH Receives TANF, SNAP or Public Health Insurance	HH Receives TANF	HH Receives SNAP	HH Receives Public Health Insurance
Simulated share of children eligible for public health insurance	0.201*** [0.023]	0117*** [0.021]	0.098*** [0.025]	0.211*** [0.025]
SNAP outreach spending per person income <150% poverty	0.105 [0.079]	0.073** [0.031]	0.125** [0.050]	0.061 [0.084]
# Observations	43,648	43,648	43,648	43,648

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: The sample is restricted to households with income below 400 percent of the federal poverty threshold at the month of interview included in the hardship regression in Table 3. Robust standard errors clustered by state in brackets. Models also included controls for demographic characteristics of the household, state economic considition state and year fixed effects (see Section 4 for the complete list of controls).

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

Table 3 –

Effect of Program Participation on Hardship

Dependent Variable	Number of Hardships		Had Food Insufficiency		Unmet Medical or Dental Need
	OLS	2SLS	OLS	2SLS	OLS	2SLS
HH Receives TANF, SNAP or Public Health Insurance	0.697*** [0.039]	−1.144** [0.472]	0.045*** [0.003]	−0.166** [0.080]	0.099*** [0.008]	−0.030 [0.108]
# Observations	43,650		44,225		44,083
Weak Instrument Tests (F-Stats)
Olea and Pflueger effective F-test		29.83		31.94		30.57
τ =10%		8.58		9.61		9.22
Implied Elasticity		−0.41		−1.27		Not Sig.

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: The sample is restricted to households with income below 400 percent of the federal poverty threshold at the month of interview. Robust standard errors clustered by state in brackets. Models also included controls for demographic characteristics of the household, state economic considition state and year fixed effects (see Section 4 for the complete list of controls). Instrumental variables in two-stage least squares models are simulated share of children eligible for public health insurance and SNAP outreach spending per person with income below 150 percent of the federal poverty line.

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

We also investigate the effect of the two policy instruments on each individual program (Table 2, columns 2–4). As expected, we find that instruments affect program take-up mostly through the program that they are intended to affect. The simulated share of children eligible for public health insurance has the greatest effect on public health insurance participation, and SNAP outreach spending has the greatest effect on SNAP participation. These results are consistent with past literature showing that our policy instruments are associated with higher participation in their respective programs (Ratcliffe, McKernan, and Zhang 2011; Miller and Wherry 2019).

In addition, we find evidence for cross-program effects of our Medicaid/SCHIP policy instrument on TANF and SNAP participation, and our SNAP instrument on TANF participation. This result is not surprising given how safety-net programs interact with each other. For example, Congress has provided options for safety net programs to use information collected by one public benefit program to help enroll low-income people for other benefits for which they qualify with the goal of making programs more efficient and reducing hassles for struggling families (Ambegaokar, Neuberger, and Rosenbaum 2017). And prior to 1996, Medicaid eligibility was automatic for AFDC recipients (Maloy et al. 1998). Our cross-program instrument findings further highlight the importance of combining the three safety-net programs in our analysis.

While our two instruments have strong predictive power for their respective programs, one might question whether other policy variables also could be used as instruments in our analysis. In Table A3, we investigate an extended set of policy instruments that could predict program participation in our sample, including additional variables from the SNAP Policy Database and the Urban Institute’s Welfare Rules Database. We search for instruments with high predictive power for their respective programs. Following this rule, we show that our two instruments—simulated share of children eligible for public health insurance and SNAP outreach spending per person—are the only policy variables that positively predict participation for their respective programs in our sample. We also note that our IV strategy would suffer from weak instrument issues if we used all the potential policy variables as instruments in our 2SLS regression (Table A3, Olea and Pflueger effective F-stat of 8.42). Finally, we explore using factor analysis to reduce the dimension of instruments used in Table A3 (not shown in the paper but available upon request). Using a principal-component factoring method, the analysis reduces the nine potential instruments to three factors. While two of these factors are positive and statistically significant, this approach provides weaker instruments than our preferred specification reported in Table 3 (Olea and Pflueger effective F-stat of 7.19). As a result, we use the more parsimonious set of two instruments in our main analysis.

The Effect of Program Participation on Material Hardship

Does the safety net help decrease material hardship among low-to-moderate income families with children? Our results suggest that receiving TANF, SNAP, or public health insurance reduces the number of material hardships – which include a general report of not meeting essential expenses, inability to pay rent or mortgage, eviction, unpaid utility bills, utility service cut, phone service cut, unmet medical or dental need, and food insufficiency. We also find that participation in these programs reduces the likelihood of being food insufficient.

Our ordinary linear least square model (OLS), finds a positive association between TANF, SNAP, and public health insurance participation and our three hardship measures (number of hardships, food insufficiency, and unmet medical and dental need), shown in Table 3. This result is not surprising given that families typically use these programs when they are in economic distress, for example after a job loss. In other words, our OLS estimates also reflect the selection of families in need to participate in safety net programs.

Using our instrumental variable strategy, which addresses selection bias, we find that a 10 percentage point increase in safety net participation (in TANF, SNAP, or public health insurance) by families reduces the average number of hardships low-to-moderate income families with children experience by 0.114 (Table 3). Translating this estimate to elasticities (using the descriptive statistics from Table 1), we find that a 10 percent increase in program participation reduces material hardship by 4.1 percent (−0.41 elasticity). This result suggests that the direct benefits of participating in the three income-tested programs out programs outweigh any potential indirect negative behavioral effects for our material hardship measures.

We also find that a 10 percentage point increase in participation in the three safety net programs by families reduces the share of food insufficient families by 1.66 percentage points. Similarly, using descriptive statistics from Table 1, we find that a 10 percent increase in program participation reduces food insufficiency by 12.7 percent (−1.27 elasticity). This result suggests that by accessing TANF, SNAP, or public health insurance, low-to-moderate-income families with children are significantly more likely to have enough food to eat.

Our finding that safety net participation reduces hardship is generally consistent with the literature. Earlier studies find that SNAP receipt alone reduces food insecurity by 12.8–16.8 percentage points (Kreider et al. 2012; Ratcliffe, McKernan, and Zhang 2011; Shaefer and Gutierrez 2013) and food insufficiency by 6–11 percentage points (Gundersen, Kreider, and Pepper 2017). We expect that participation in any of the three programs (our measure) will have a larger effect than participation in one of the programs because of interactions between the programs.²¹

In their working paper, Schmidt, Shore-Sheppard, and Watson (2013) find that an additional $1,000 in cash or food benefits actually received by families reduces the incidence of low food security by 4 percentage points.²² Converting these findings to elasticities, their estimate suggests that a 10 percent increase in actual benefit receipt reduces low food security by 4.3 percent (−0.43 elasticity), which is slightly larger than our material hardship elasticity estimate of −0.41, but still lower than our food insufficiency elasticity of −1.27. When comparing elasticities, bear in mind that a small share of households are food insufficient in our data (4.5 percent). This mechanically makes small percentage point changes translate to large percent changes.

We also estimate a negative but imprecise effect of program participation on unmet medical or dental need. In interpreting this result, it is important to keep in mind that public health insurance expansions targeted to children might be missed in our models, which capture unmet medical or dental need of both parents and children (i.e., the entire household). That is, significant declines in unmet medical or dental need for children could be obscured because parents, and thus the household, still have unmet needs.

As discussed above, the validity of our IV model depends on the quality of the instruments. Table 3 also presents effective F statistics from Olea and Pflueger (2013) for all IV regressions as well as the weak instrument threshold for the case of 10 percent potential bias (τ = 10%) and 5 percent level of significance. In all our specifications, we can reject the null of weak instruments. Thus, there is no need for weak IV instrument inference adjustments to our estimations.

Robustness Checks

We provide several robustness checks for the findings. First, in the above model specifications, our explanatory variables (including program participation) are from the interview month. As a robustness check, we examine any program participation in the past 12 months and average months of program participation in the past 12 months as our explanatory variables (Table A4).²³ Our results are generally robust to using these alternative participation measures. We estimate negative effects of safety net participation on number of material hardship (p<0.05).

Nearly two thirds (64 percent) of households report experiencing no economic hardships in the past 12 months, so we also show that our main result is generally robust to estimation methods that better handle bunching at zero. For this purpose, we estimate an IV Poisson model, which is often suggested when the dependent variable is a count variable (Mullahy 1997, Cameron and Trivedi 2013). We find a negative and marginally significant effect of program participation on material hardship using a Poisson IV model (marginal effect of −0.672, see Table A4). As an additional robustness check, we define the study population based on the household head’s educational attainment—some college or less. We again find that safety net participation reduces the number of hardships (coefficient of −0.905), although the estimated coefficient is less precise (p<0.10).²⁴

Finally, we estimate models on data that adjust for the underreporting of program participation in the SIPP. Specifically, we use a modified version of the Scholz, Moffitt and Cowan (2009) (SMC) method as described by Mittag (2019).²⁵ In the modified SMC method, we assign receipt probabilistically according to the predicted probabilities generated from a Probit model, until the number of recipients in the SIPP matches administrative aggregates in expectation.²⁶ We match administrative program participation by year at the household level for SNAP from the SNAP Data Tables (FNS 2020) and AFDC/TANF from the AFDC/TANF Caseload Data (ACF 2020). In addition, we match the individual level participation for Medicaid/CHIP from the Medicaid Enrollment (MACPAC 2020) and CHIP enrollment (from Kaiser Family Foundation 2020). Using the adjusted participation indicator for each program, we then create an indicator for participation in any of the three programs. Finally, we estimate our preferred specifications using the participation variable that has been adjusted for program underreporting (Table 4).

Table 4 –

Effect of Adjusted Program Participation on Hardship

Dependent Variable	Number of Hardships		Had Food Insufficiency		Unmet Medical or Dental Need
	OLS	2SLS	OLS	2SLS	OLS	2SLS
HH Receives TANF, SNAP or Public Health Insurance (Adjusted)	0.568*** [0.030]	−0.884*** [0.322]	0.036*** [0.002]	−0.134** [0.053]	0.092*** [0.007]	−0.024 [0.081]
# Observations	43,650		44,225		44,083
Weak Instrument Tests (F-Stats)
Olea and Pflueger effective F-test		34.65		35.79		35.66
τ =10%		9.18		11.64		8.32
Implied Elasticity		−0.44		−1.43		Not sig.

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: To address undereporting of program receipt in the SIPP, participation on TANF, SNAP or Public Health Insurance are adjusted using the modified SMC method described in section 6. The sample is restricted to households with income below 400 percent of the federal poverty threshold at the month of interview. Robust standard errors clustered by state in brackets. Models also included controls for demographic characteristics of the household, state economic considition state and year fixed effects (see Section 4 for the complete list of controls). Instrumental variables in two-stage least squares models are simulated share of children eligible for public health insurance and SNAP outreach spending per person with income below 150 percent of the federal poverty line.

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

Consistent with our preferred specification, we find negative and statistically significant effects of safety net participation on hardship (participation reduces hardship) when adjusting for underreporting using the modified SMC method. We find that a 10 percentage point increase in program participation (in TANF, SNAP, or public health insurance) reduces the average number of hardships by 0.09 and the likelihood of food insufficiency by 1.34 percentage points. While these effects are smaller than the results presented in Table 3, the higher participation in the programs from the modified SMC adjustments translates to elasticities similar to those we present in Table 3.

Individual Programs

Due to program participation multicollinearity (Table 1, Panel B), it is not feasible to separately identify the effect of each individual program using a multinomial model in our data. Nonetheless, Table 5 shows the effect of each program individually on material hardship using separate models but note that the vast majority of TANF recipients receive SNAP and public health insurance and the vast majority of SNAP recipients receive public health insurance (Table 1, Panel B). With that caveat in mind, we find that the effect of program participation on the material hardship varies from −1.11 (public health insurance) to −2.06 (SNAP), but these effects are not statistically different from each other (p-value>10%).²⁷ We conclude that there is no evidence that one of the three programs has a stronger effect on material hardship than the others.

Table 5 –

Effect of Individual Programs on Number of Hardships

Dependent Variable	Number of Hardships
HH Receives TANF	−1.944*** [0.739]
HH Receives SNAP		−2.059** [0.985]
HH Receives Public Health Insurance			−1.106** [0.454]
# of Programs HH Participates (0–3)				−0.535** [0.217]
# Observations	43,650	43,650	43,650	43,650
Weak Instrument Tests
Olea and Pflueger effective F-test	30.41	9.58	32.55	31.29
τ =10%	16.66	13.63	7.47	11.75

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: The sample is restricted to households with income below 400 percent of the federal poverty threshold at the month of interview. Robust standard errors clustered by state in brackets. Models also included controls for demographic characteristics of the household, state economic considition state and year fixed effects (see Section 4 for the complete list of controls). Instrumental variables in two-stage least squares models are simulated share of children eligible for public health insurance and SNAP outreach spending per person with income below 150 percent of the federal poverty line.

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

Finally, to identify the cumulative effect of programs on material hardship, we also estimate the effect of the number of programs a family participates in on material hardship in Table 5. We find that participation in one additional program decreases the number of hardships a low-to-moderate income family experiences by 0.54. Using this estimate, one could conclude that participation in all three programs is associated with a 1.62 reduction in average number of hardships experienced.

7. Conclusion

Exploring variation in state-level safety-net policies, our paper finds that a 10 percentage point increase in the share of low-to-moderate income families with children receiving TANF, SNAP, or public health insurance reduces the average number of hardships low-to-moderate income families with children experience by 0.11. In addition, we find that a 10 percentage point increase in participating in one of these three safety net programs also reduces food insufficiency by 1.7 percentage points. This effect is large but not inconsistent with earlier studies (Kreider et al. 2012; Ratcliffe, McKernan, and Zhang 2011; Shaefer and Gutierrez 2013; Schmidt, Shore-Sheppard, and Watson 2016). We find a negative but imprecise effect of safety net program participation on unmet medical or dental need.

Our findings suggest that the basic needs of families would be at risk should these safety net programs be cut. Food insufficiency, unmet medical and dental need, and the inability to pay basic bills still hurt families in the United States. That is not to say the programs cannot be improved. For example, see Bitler and Hoynes (2016a) for a proposal to make TANF a successful safety net program again that responds to the business cycle and helps families in the most need while preserving its emphasis on encouraging work.

Our study highlights the importance of examining the effect of programs not just on the official poverty measure, as many previous studies have done, but also on indicators of material hardship. The receipt of SNAP and public health insurance are not captured in the official poverty measure, as the indicator of resources in that measure does not include many noncash or near-cash benefits. Indeed, our empirical results confirm the substantial effects of many of these programs on the material hardships experienced by U.S. households.

Safety net programs are not just helpful for families in tough times; they might also be good for the economy. When program spending increases during a recession and provides resources to low-to-moderate income families (as well-targeted, automatically stabilizing safety net programs should), people with low incomes are more likely to spend money and stimulate the economy. Evidence from the Great Recession suggests the biggest “bang for the buck” comes first from safety net spending programs such as Medicaid, SNAP, and unemployment insurance (Blinder 2016; Schanzenbach et al. 2016), then from tax cuts to people with low incomes. Tax cuts benefiting businesses had less of an effect.

As a first look at the joint effects of multiple safety net programs on multiple measures of material hardship, this research raises additional questions for future research. How have the program effects changed over time as the programs and policies changed? Do the effects differ according to the time periods measured (e.g., 1990s, 2000s, 2010s)? What are joint programs’ effects on poverty, deep poverty, and the poverty gap?

Highlights.

• Families typically participate in multiple safety net programs

• We evaluate the simultaneous effect of TANF, SNAP and Medicaid/SCHIP on hardship

• We use an instrumental variables approach exploiting variation in state policy rules

• Participation in these safety net programs reduces the average number of hardships

• Participating in these safety net programs also reduces food insufficiency

Appendix A.

Table A1 –

Exclusion Restrictions Test

Dependent Variable	Simulated share of children eligible for public health insurance	SNAP outreach spending per person income <150% poverty
Governor is Democrat (lagged 1 year)	−0.006 [0.012]	−0.001 [0.007]
Poverty Rate (lagged 1 year)	−0.001 [0.003]	−0.001 [0.002]
Log State General Revenue (lagged 1 year)	−0.112 [0.069]	−0.150 [0.092]
# Observations	892	950

Sources: State level data from TRIM3, SIPP, USDA, SNAP Policy Database, University of Kentucky Center for Poverty Research, Urban-Brookings Tax Policy Center for 1993–2001.

Notes: Simulated share of children eligible for public health insurance is averaged across all children in the SIPP for each state across the years. Robust standard errors clustered by state in brackets.

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

Table A2 –

Sample Selection Test

Dependent Variable	HH Income below 200 poverty threshold in the current month	HH Income below 250 poverty threshold in the current month	HH Income below 300 poverty threshold in the current month	HH Income below 350 poverty threshold in the current month	HH Income below 400 poverty threshold in the current month
Simulated share of children eligible for public health insurance	0.086*** [0.028]	0.091*** [0.034]	0.077** [0.033]	0.055* [0.031]	0.034 [0.022]
SNAP outreach spending per person income <150% poverty	0.102 [0.074]	0.112 [0.074]	0.095 [0.074]	0.100 [0.081]	0.067 [0.049]
# Observations	61,013	61,013	61,013	61,013	61,013
Instruments are jointly different from zero
F-test	6.51	5.56	4.6	2.56	2.28
P-value	0.00	0.01	0.01	0.09	0.11

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: The sample is restricted to respondents of hardship questions. Robust standard errors clustered by state in brackets. Models also included controls for demographic characteristics of the household, state economic considition state and year fixed effects (see Section 4 for the complete list of controls).

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

Table A3 –

The Effect of Extended Policy Instrument in Program Participation

Dependent Variable	HH Receives TANF, SNAP or Public Health Insurance	HH Receives TANF	HH Receives SNAP	HH Receives Public Health Insurance
Simulated share of children eligible for public health insurance	0.206*** [0.024]	0.112*** [0.021]	0.098*** [0.026]	0.219*** [0.027]
SNAP outreach spending per person income <150% poverty	0.102* [0.061]	0.078*** [0.026]	0.120** [0.057]	0.078 [0.068]
SNAP all legal noncitizen adults eligible	0.024* [0.013]	−0.004 [0.010]	−0.003 [0.011]	0.025 [0.016]
SNAP some noncitizen adults eligible	−0.010 [0.012]	0.008 [0.009]	0.012 [0.009]	−0.008 [0.014]
State uses BBCE	0.010 [0.013]	−0.020*** [0.006]	0.011 [0.010]	0.006 [0.013]
Fingerprinting	0.023* [0.013]	0.008 [0.007]	−0.008 [0.013]	0.027** [0.012]
One vehicle excluded from asset test	0.008 [0.011]	−0.004 [0.007]	0.003 [0.009]	0.005 [0.012]
TANF maximum monthly benefit for family of three ($/100)	0.010 [0.008]	−0.006* [0.003]	0.003 [0.005]	0.015* [0.008]
TANF maximum number of months eligibility extended to pregnant women without other child	0.000 [0.003]	0.003 [0.002]	0.000 [0.002]	−0.001 [0.003]
# Observations	43,650	43,650	43,650	43,650
Weak Instrument Tests
Olea and Pflueger effective F-test	8.42
τ =10%	15.73

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: The sample is restricted to households with income below 400 percent of the federal poverty threshold at the month of interview. Robust standard errors clustered by state in brackets. Models also included controls for demographic characteristics of the household, state economic considition state and year fixed effects (see Section 4 for the complete list of controls).

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

Table A4 –

Robustness Check

Dependent Variable: Number of Hardship in the past 12 months
Sample	Family Income < 400 poverty			Head with some College Education or Less
Method	2SLS	2SLS	IV Poisson	2SLS
HH Receives TANF,SNAP or Public Health Insurance any time in the Past 12 months	−0.715** [0.313]
Average HH Participation on TANF,SNAP or Public Health Insurance any time in the Past 12 months		−0.628** [0.270]
HH Receives TANF,SNAP or Public Health Insurance			−0.672* [0.383]	−0.905* [0.465]
# Observations	43,650	43,650	43,650	46,249
Weak Instrument Tests
Olea and Pflueger effective F-test	25.29	46.53		23.32
tau=10%	9.96	9.04		10.04

Sources: Weighted Survey of Income and Program Participation, state policy, and economic data for 1992, 1995, 1998, 2003, 2005, 2010, and 2011.

Notes: Robust standard errors clustered by state in brackets. Models also included controls for demographic characteristics of the household, state economic considition state and year fixed effects (see Section 4 for the complete list of controls). Instrumental variables in two-stage least squares models and IV poisson model are simulated share of children eligible for public health insurance and SNAP outreach spending per person with income below 150 percent of the federal poverty line.

^*p < 0.1,

^**p < 0.05,

^***p < 0.01.

Appendix B. Material Hardship and Benefit Receipt Questions from the SIPP Survey

Material Hardship²⁸

1. During the past 12 months, has there been a time when you did not meet all of your essential expenses (e.g., mortgage or rent payments, utility bills, or important medical care)? (yes/no)

2. Was there any time in the past 12 months when you did not pay the full amount of the rent or mortgage? (yes/no)

3. In the past 12 months, were you evicted from your home or apartment for not paying the rent or mortgage? (yes/no)

4. Was there a time in the past 12 months when you did not pay the full amount of the gas, oil, or electricity bills? (yes/no)

5. In the past 12 months, did the gas or electric company turn off service or the oil company not deliver oil? (yes/no)

6. Was there a time in the past 12 months when the telephone company disconnected service because payments were not made? (yes/no)

7. In the past 12 months, was there a time you needed to see a doctor or go to the hospital but did not go? (yes/no)

8. In the past 12 months, was there a time you needed to see a dentist but did not go? (yes/no)

9. Getting enough food can also be a problem for some people. Which of these statements best describes the food eaten in your household in the last four months?

   • Enough of the kinds of food we want
   • Enough but not always the kinds of food we want to eat
   • Sometimes not enough to eat
   • Often not enough to eat

Program Receipt

1. Did you receive any public assistance payments such as AFDC or TANF in this month? (yes/no)

2. Did you receive income from food stamps in this month? (yes/no)

3. Were you covered by Medicaid in this month? (yes/no)

   • Set as no for people also covered by Medicare or military to isolate Medicaid/SCHIP.