https://orcid.org/0000-0002-7588-7625
Abstract
High levels of poverty and economic precarity in the United States relative to other countries have led to academic and policy debates about whether welfare state investments accomplish what they are intended to. Although social safety net spending clearly has antipoverty effects at the national level, there is scant evidence on the “resource pathway” presumed to underlie the effects of the local welfare state on families with children. Which types of public investments have especially contributed to the total resources of households with children? Understanding this question at the state level is important, given dramatic variation in states’ safety net spending on children and the rise of federalism, which increases state autonomy in designing and administering social programs. Using annual data from the 1997–2016 State-by-State Spending on Kids Dataset linked to data from the Census Bureau's Annual Social and Economic Supplement to the Current Population Survey, we examine the relationship between transfer spending in states and household income sources. Findings suggest that government transfers raise the total income of households with the lowest income and educational levels and that transfer income among these households is more multidimensional than among higher resource households. Further, analyses using variation within and across states demonstrate that state-level spending in each area is associated with an increase in corresponding transfer income among non-college-educated households and those in the bottom half of the income distribution; such spending is associated with no increase (or a decrease) in transfer income among college-educated households and those in the top quarter of the income distribution. These results suggest that increases in state-level spending disproportionately benefit the budgets of households with the lowest resources and might be a promising means to reduce resource gaps between households.
Introduction
Relative to other industrialized countries, the United States has a small welfare state, with a smaller infrastructure of support for children and parents in the form of health services, caregiving support, and income assistance (Bradbury et al. 2015; Garfinkel et al. 2010).1 Although public spending on children in the United States has grown over time, averaging roughly $26,000 per child in recent years (Isaacs and Edelstein 2017), it has not grown as quickly as spending on adults (Edelstein et al. 2016). High levels of poverty and economic precarity in the United States relative to other countries have led to academic and policy debates about whether welfare state investments accomplish what they are intended to. This debate has long been prominent in the comparative welfare transfers literature (e.g., Brady 2005; Burtless 1994; Cantillon 1997) and is increasingly a feature of U.S. public and scholarly debates about whether government spending on social programs should necessarily translate into direct improvements in family resources (Aizer et al. 2022; Desmond 2023; Herd and Moynihan 2019; Institute for Research on Poverty 2019).
Despite these claims, safety net spending on children—which we define as spending in states (from federal, state, and local sources) on major programs relevant to children and families—clearly has antipoverty effects at the national level. Child poverty in the United States has markedly declined, primarily because of safety net spending (Creamer et al. 2022; Wimer et al. 2016). We know far less at the subnational level, where child economic circumstances and the size of state investments in children vary dramatically. For example, child poverty levels range from 10% in Iowa to more than 20% in California (Laird et al. 2018; Renwick and Fox 2016), and the size of the child social safety net differs threefold between more and less generous states (Greenberg et al. 2021). Given this dramatic variation and the rise of federalism that increases states’ autonomy in designing and administering social welfare programs, it is important to understand how the widely variable levels and types of state investments manifest for household income amounts and sources. Scholars and policymakers continue to debate whether safety net spending is an effective use of government resources, as evidenced by congressional debates that regularly threaten to shut down the U.S. government. Moreover, growing evidence suggests that a strong child safety net improves health and development for low-income children (Bradbury et al. 2015; Corak 2013; Corak et al. 2011; Waldfogel 2009). Improvements in household resources are an implied key pathway through which public investment affects children (Reardon 2011), but empirical understanding of the relationship between state investments and household resources is limited.
Using annual data from the 1997–2016 State-by-State Spending on Kids Dataset linked to data from the Census Bureau's Annual Social and Economic Supplement to the Current Population Survey (CPS-ASEC), we examine the relationship between state-level spending and the components of household income. In particular, we decompose household income into the amount received from market and transfer sources (e.g., income support, food assistance, housing), with an emphasis on understanding how this allocation varies across state environments. Which types of public investments have especially contributed to the total resources of households with children? Is the contribution of different types of investments similar across more and less generous states? We advance the field by providing evidence on how household incomes change in response to state-level safety net investments—a subject of great theorization and public and political debate but far less empirical evidence. Moreover, we do so at the state level—a unit of analysis that is increasingly relevant for U.S. policy design and implementation, with important consequences for population well-being. We build on the large comparative welfare states literature that has examined similar questions across countries with varying degrees and types of welfare state regimes, using this pioneering body of evidence to motivate a focus on U.S. states.
Background
The Social Safety Net and Family Economic Well-being
The idea that transfer programs will necessarily increase economic resources and reduce poverty is far from a forgone conclusion in scholarly and policy debates. High levels of poverty and economic precarity in the United States relative to other countries have led scholars and policymakers to question whether welfare state investments meet their intended goals. This debate has long been prominent in the comparative welfare transfers literature, a site of robust discussion about the relative importance of welfare state spending versus other macroeconomic forces (e.g., economic growth, free market forces, the supply of jobs) for economic well-being (Biegert 2017; Brady 2005; Burtless 1994; Cantillon 1997).
These debates also permeate the U.S. literature, with some research suggesting that welfare state spending might not directly translate into family resources for several reasons. First, some programs have eligibility requirements that create so-called benefits cliffs, leading families to lose eligibility for important benefits after slight increases in income (Institute for Research on Poverty 2019). Second, many programs suffer from administrative burdens that increase the complexity of enrolling in programs and understanding eligibility for them, resulting in low participation in many social safety net programs among the eligible population (Bitler et al. 2003; Herd and Moynihan 2019; Linos et al. 2022). Third, some economic arguments suggest that welfare generosity encourages dependency and decreases labor force participation (Bane and Ellwood 1994; Lindbeck 1994; Moffitt 2015; see also Schmieder et al. 2016), although other work suggests that potential disincentives for adults might be offset by sizable long-run benefits for children (Aizer et al. 2022). Finally, although U.S. spending on safety net programs has increased over time, these increases have been disproportionately targeted at adult populations as the share of total spending devoted to children has stagnated and is projected to decline as a share of total government spending (Isaacs and Edelstein 2017). This feature of U.S. spending patterns could explain why poverty among children, in particular, remains stubbornly high in the United States compared with other nations, with 2020 and 2021 being notable exceptions (Desmond 2023; Parolin et al. 2021). Beyond academia, this debate has also permeated the public sphere, with political and popular media debates considering whether safety net spending is a worthy use of government resources.
Although transfer programs are not universally theorized to improve economic well-being for lower income families, convincing comparative and U.S. evidence demonstrates that welfare state spending reduces poverty (Brady 2005; Brady and Parolin 2020; Parolin and Filauro 2023; Wimer et al. 2020). Growing evidence also demonstrates that a strong local welfare state benefits families and children and increases equality of opportunity. Levels of child poverty in the United States remain stubbornly high, with nearly 1 in 10 children living in poverty as measured by the Census Bureau's Supplemental Poverty Measure (SPM) (Fox and Burns 2021). But investments in government policies and programs targeting families with children have led to marked declines in the prevalence of child poverty over the last several decades (Wimer et al. 2016). The potential for child-focused public investments to increase household resources is also clearly displayed by the reduction in child poverty seen after the recent temporary expansion of the Child Tax Credit (CTC) in response to the tremendous economic need caused by the COVID-19 pandemic (Burns et al. 2022; Parolin et al. 2021).
Given the means-tested nature of many tax and transfer programs, low-income families should disproportionately benefit from many forms of public spending (Vericker et al. 2012). In contrast to public education, which serves children across income groups, many U.S. policies and programs are more targeted. Medicaid and other health programs, for example, have often targeted low- and moderate-income families. The Medicaid health insurance program, jointly financed by the federal government and the states, represents the second-largest form of investment in children after K–12 education (Isaacs and Edelstein 2017). Many states have expanded Medicaid beyond federal minimums for benefits and coverage, leading to wide variation in eligibility levels, service coverage, payment mechanisms, and spending per enrollee. Children also benefit from spending on the Children's Health Insurance Program (CHIP) and public health systems, and income security and social service programs also support families with children. Some of these programs are explicitly limited to families with children (i.e., Earned Income Tax Credit [EITC], the CTC, Temporary Assistance to Needy Families [TANF], child support enforcement, the disabled child portion of Supplemental Security Income [SSI], child welfare services, and childcare assistance). In addition, some programs that serve the low-income population have a disproportionate share of child recipients. For example, two thirds of Supplemental Nutrition Assistance Program (SNAP) benefits go to households with children, and SNAP was a primary form of support for children with unemployed parents during the Great Recession (Isaacs and Healy 2014). Most of these programs are federal or joint federal–state programs, and many of them target low-income families.
National-level analyses reveal that these transfers, especially from food assistance and refundable tax credits, have led to substantial income gains near the bottom of the distribution among households with children (Wimer et al. 2020). Similarly, national declines in deep and extreme poverty among households with children have been spurred by the expansion of SNAP benefits (Brady and Parolin 2020). By providing both direct cash assistance, “near cash,” and in-kind benefits, public investments should increase the funds available to parents with lower socioeconomic status (SES) to invest in basic necessities for children and make developmental investments (Jackson and Schneider 2022; Milligan and Stabile 2011; Yeung et al. 2002). A strong safety net for children and families also provides a critical buffer against the effects of poverty and economic instability on health and well-being, and ample evidence demonstrates that public investments can improve outcomes for low-income children (Bradbury et al. 2015; Waldfogel 2009).
How Should State-Level Investments Affect Household Resources?
We argue that our understanding of the relationship between government investments and household income at the smaller level of the state is much more limited, despite substantial variation within the United States in the size and scope of state and local welfare state spending. In addition, research has focused more on poverty than on the components of income, with some important exceptions (e.g., Wimer et al. 2020). Although a focus on poverty is essential, it obscures some potentially important effects of safety net spending on households’ economic resources. For example, government spending might increase household incomes even without reducing some metrics of poverty, partly because the official poverty measure does not account for many components of transfer income.
Which types of public investments should contribute the most to the total resources of households with children? Many forms of government investment are critical for the well-being of parents and children, and research increasingly suggests that overlapping, simultaneous forms of public investment are especially beneficial for child development (Ansari and Pianta 2018; Johnson and Jackson 2019). Despite changes over time in the size and targeting of the social safety net for families and evidence that the social safety net has contracted for the lowest income families in recent decades (e.g., Hoynes and Schanzenbach 2018; Moffitt 2015), most families with low incomes are eligible for various public benefits related to public health insurance, food and nutrition, income support, and housing assistance (Edelstein et al. 2014). However, the size and generosity of public support varies drastically across U.S. states. In 2016, the highest spending states (e.g., Vermont, at $22,000 per child) spent three times more per child on a bundle of parent and child-focused programs than the lowest spending states (e.g., Utah, at $7,200 per child). Although some states are universally more or less generous across different social support types (e.g., education, health, income support), states that spend more on one type of support do not always spend more on another type. For example, the correlations among income support, health, and education spending range from .13 to .46 (Greenberg et al. 2021).
What do these patterns suggest about the amount and sources of household income from government transfers across U.S. states? First, state-level investments should increase the amount of household income disproportionately for the lowest educated households and for households in the bottom of the pretransfer income distribution, at least relative to the most educated households and those at the top of the pretransfer income distribution. Given that many public supports are means-tested, households with the fewest resources will be eligible for more benefits. However, lower resource households receive income from means-tested public support, but higher resource households also receive support from the state via cash transfer and tax credit programs, such as nonrefundable tax credits and unemployment benefits (Guner et al. 2012; Kovalski and Sheiner 2020; Rodgers and Tedin 2006).
Second, state-level investments should lead to a greater mix of transfer income sources among households with lower educational and pretransfer income levels. Although the sources of transfer income among households with more education and income should primarily come from cash support and tax credit programs, households with fewer resources should also receive support from food assistance and housing subsidy programs. If higher spending states are consistently more generous across different types of spending programs, then the amount of transfer income should be higher across all types of programs. If states selectively invest in specific types of public support, then household income from transfer programs will be higher in higher spending states, but the sources of this income will be more homogeneous.
Methods
Using annual data from the 1997–2016 State-by-State Spending on Kids Dataset linked to data from the Census Bureau's CPS-ASEC, we examine the relationship between spending in states and the amount of household income received from both market and transfer sources.
Data
Spending in States on Children and Families
Despite substantial spending variation that puts some states far below and some far above the national average (Harknett et al. 2005; Isaacs and Edelstein 2017), few data sources compile these spending measures across states and over time. We use the State-by-State Spending on Kids Dataset, a state-by-year database of public spending from federal, state, and local sources for 1997–2016, aiming to cover the longest period feasible with existing administrative data (Isaacs et al. 2021). The public spending database includes all 50 states and the District of Columbia, drawing on data from the U.S. Census Bureau's Annual Survey of State and Local Government Finances, federal agency websites, the State Funding for Children Database compiled by the Rockefeller Institute of Government (1998–2008), and other sources. We measure state-level real spending per child in 2016 dollars in several areas. We begin by measuring total spending relevant to household income, including measures of spending on cash (TANF, other cash assistance payments and services, SSI, Social Security, unemployment, and worker's compensation), tax credits (EITC, CTC, additional child credit), nutrition support (SNAP), health (Medicaid/CHIP and public health), education (Pre-K–12), housing/community development, and other spending. This measure broadly captures spending programs that strongly target children or families with children. Although these data do not include a separate measure of childcare spending, our measure of spending on other cash assistance and services captures a large portion of state-level childcare spending.
Household Income and Characteristics
To understand the potential impacts of spending in states on the components of household income, we use historical data from the CPS-ASEC for 1998–2017, covering the calendar years 1997–2016. The CPS-ASEC, mostly fielded in March of each year, collects detailed income data for each household member 16 or older and is the Census Bureau's source of income data for its annual estimates of household income, poverty, and health insurance. With some important exceptions, such as the exclusion of the homeless population, the survey captures the economic circumstances of the U.S. population and can be effectively used to understand how household-level transfer income varies across U.S. states. We use data from the 1998–2017 surveys for information about household income in 1997–2016, the same years observed in the state-level data. All analyses are weighted using ASEC sampling weights (Ruggles et al. 2020).
We use a version of the data created by researchers at Columbia University to estimate historical changes in poverty using the Census Bureau's SPM before 2009, the first year that the Census Bureau released the SPM (Fox et al. 2015). Unlike the historical SPM work, our goal is to estimate not trends in poverty but rather trends in household income with a definition of resources that is consistent with the SPM resources concept. That is, we seek a measure that includes pretax income as well as the monetary value of in-kind benefits (e.g., housing subsidies or the SNAP) net of taxes (i.e., income and payroll taxes paid and the value of federal and state tax credits).
The benefit of the Columbia University time series is that it contains all of the elements of resources back to our starting year, measured consistently with the Census Bureau's estimates in later years (for a detailed explanation of methods used to create the historical SPM datasets, see Fox et al. 2015). Because we do not aim to measure SPM poverty and we seek to understand the role of spending in states in predicting household resources, we first aggregate all resource concepts from the SPM unit to the household level. The SPM unit often matches the household exactly, but it differs in some cases. The SPM unit includes all individuals related to one another by blood, marriage, or adoption plus any cohabiting romantic partners, their children, any foster children, and any unrelated children younger than 15 in the household. The household would differ from this definition in cases where multiple such units resided in the same household or when multiple unrelated individuals shared a household residence. Using the household rather than the SPM unit as the unit of analysis is consistent with the Census Bureau's various income series (Guzman and Kollar 2023) and aligns with a recent National Academy of Sciences report (Ziliak et al. 2023) that recommended changing the unit of measurement in poverty statistics to the household level.
We use two primary measures of resources at the household level: (1) total pretax, pretransfer resources, including all income sources except those coming from the tax system or government programs; and (2) total posttax, posttransfer resources, including all income sources net of federal and state taxes (including the value of tax credits), plus the value of cash and noncash government transfers.2 All income measures are adjusted for inflation and measured in 2016 dollars using the Consumer Price Index research series (CPI-U-RS).3
In addition to examining the aggregate measure of posttax and posttransfer resources, we decompose this measure to understand the relative contribution of different types of state investments to households’ economic resources: (1) food and nutrition resources (SNAP, Special Supplemental Nutrition Program for Women, Infants and Children [WIC], school lunches), (2) housing resources (housing subsidies from public housing and rental assistance, and the Low-Income Heating and Energy Assistance Program), (3) cash transfers (TANF [and Aid to Families with Dependent Children, which preceded TANF], unemployment insurance, Social Security, and SSI), and (4) tax-based transfers minus tax liabilities (tax credits, such as EITC and CTC).4 These household resource measures might not correspond perfectly to our state-level spending domains (e.g., state-level nutrition spending is limited to SNAP, whereas household-level food and nutrition resources also include WIC and school lunches). However, they correspond very closely conceptually, and sensitivity analyses limiting the household-resource measures to narrower definitions yield highly similar patterns.
Finally, we use CPS data to measure SPM-unit demographic characteristics, including the educational attainment of the household reference person; race and ethnicity and Hispanic identity of the census “householder” or reference person;5 employment status (any employment and full-/part-time), sex, nativity, and marital status (married, cohabiting, or neither) of the census householder; household size; and the number of children in the household.
State-Level Controls
We account for state-level factors that might covary with the amount of spending in states on children and families, drawing on data from the National Welfare database of the University of Kentucky Center for Poverty Research (2024), from the CPS, and from the Bureau of Labor Statistics to construct state-year measures. Given that state-level spending increases as economic need increases during periods of economic downturn (Brown and Best 2017; Edelstein et al. 2016; Rodgers and Tedin 2006), we control for the unemployment and poverty rates. As an indicator of broader state support, we control for the prevailing minimum wage and state gross state product per capita. Using data from the Union Membership and Coverage Database drawn from the CPS (Macpherson and Hirsch 2021), we measure the state-level percentage of employed workers who are union members to account for the possibility that unionization is a pivotal state-level institution shaping poverty and income inequality (VanHeuvelen and Brady 2022). Finally, we use CPS data to account for state demographic composition with the share of the population that does not have a college degree, is non-Hispanic Black, and is Hispanic because prior research connected demographic composition with spending generosity (Alesina et al. 2001:255–272; Preuhs 2007; Rodgers and Tedin 2006; Soss et al. 2011).
Approach
Our analysis proceeds in several steps. First, we describe each measure of income and resources (pretax money income, total posttax and posttransfer resources, food/nutrition support, housing support, cash transfers, and tax-based transfers) across the distribution of spending in states. We examine patterns for all state-years as a whole and consider descriptive patterns separately by the household reference person's education and quartiles of posttax/posttransfer income (Figures 1 and 2). We focus on both household education and income because although both indicate SES, education is a cleaner measure that is less endogenous to transfer income. This approach allows us to show how both total household resources and the amount/share of resources received from different types of programs varies across the state-level spending distribution (Figures 3 and 4). It also allows us to identify how state investments are related to different parts of the income and educational distribution via different programs across spending levels in states (Figure 5).
Predicted income by state spending on nutrition × education (panel a), state spending on cash transfers × education (panel b), and state spending by tax credits × education (panel c). All estimates are significantly different relative to less than high school (HS) education.
Predicted income by state spending on nutrition × education (panel a), state spending on cash transfers × education (panel b), and state spending by tax credits × education (panel c). All estimates are significantly different relative to less than high school (HS) education.
Linear regression of household transfer income on state spending by household education
| Nutrition (1) | Cash Transfers (2) | Tax Credits (3) | |
|---|---|---|---|
| Spending Type | |||
| Nutrition | 1,076.00** | ||
| (97.91) | |||
| Cash | 55.48 | ||
| (43.37) | |||
| Tax credits | 535.20** | ||
| (97.07) | |||
| Education of Household Head (ref. = less than high school) | |||
| High school | −443.20** | −19.20 | −683.40** |
| (20.23) | (72.67) | (41.32) | |
| Some college | −646.70** | −57.80 | −1,070.00** |
| (24.96) | (79.74) | (45.34) | |
| College+ | −850.00** | −573.40** | −1,572.00** |
| (30.59) | (82.09) | (43.08) | |
| Interactions (ref. = less than high school × spending) | |||
| High school × Spending | −904.00** | 197.00** | −782.30** |
| (112.70) | (66.42) | (77.37) | |
| Some college × Spending | −1,480.00** | 258.20** | −1,144.00** |
| (114.40) | (50.97) | (83.99) | |
| College+ × Spending | −2,671.00** | 134.00** | −1,819.00** |
| (121.6) | (50.44) | (97.77) | |
| Constant | −447.10** | −4,313.00** | −378.00* |
| (129.10) | (312.60) | (168.70) | |
| Number of Observations | 424,137 | 424,137 | 424,137 |
| R2 | .220 | .054 | .320 |
| Nutrition (1) | Cash Transfers (2) | Tax Credits (3) | |
|---|---|---|---|
| Spending Type | |||
| Nutrition | 1,076.00** | ||
| (97.91) | |||
| Cash | 55.48 | ||
| (43.37) | |||
| Tax credits | 535.20** | ||
| (97.07) | |||
| Education of Household Head (ref. = less than high school) | |||
| High school | −443.20** | −19.20 | −683.40** |
| (20.23) | (72.67) | (41.32) | |
| Some college | −646.70** | −57.80 | −1,070.00** |
| (24.96) | (79.74) | (45.34) | |
| College+ | −850.00** | −573.40** | −1,572.00** |
| (30.59) | (82.09) | (43.08) | |
| Interactions (ref. = less than high school × spending) | |||
| High school × Spending | −904.00** | 197.00** | −782.30** |
| (112.70) | (66.42) | (77.37) | |
| Some college × Spending | −1,480.00** | 258.20** | −1,144.00** |
| (114.40) | (50.97) | (83.99) | |
| College+ × Spending | −2,671.00** | 134.00** | −1,819.00** |
| (121.6) | (50.44) | (97.77) | |
| Constant | −447.10** | −4,313.00** | −378.00* |
| (129.10) | (312.60) | (168.70) | |
| Number of Observations | 424,137 | 424,137 | 424,137 |
| R2 | .220 | .054 | .320 |
Notes: The model includes controls, state fixed effects, year fixed effects, and household weights. Robust standard errors at the state level are shown in parentheses.
p < .05; **p < .01
For each household h in state s in year t, Eq. (1) predicts the amount of income from specific transfer sources as a function of relevant spending in states (Spend), time-varying state and household controls (X), and state and year fixed effects. That is, we predict food support income from nutrition spending, cash transfer income from cash transfer spending, and tax credit income from tax credit spending.
In addition, we estimate models that interact state-level spending and household SES (as defined by highest educational level in the household) to understand how spending in states is associated with transfer income at different levels of the socioeconomic distribution. Building on recent work by Balli and Sørensen (2013), Giesselmann and Schmidt-Catran (2022), and others, we de-mean the items in the key interaction terms in fixed-effects models (i.e., de-meaning measures of spending and household education before creating interaction terms). All models include robust standard errors at the state level.
The inclusion of state and year fixed effects means that model identification is based on within-state variation in public spending across years and cross-state differences in public spending in a given year. Spending variation within and across states is large, as demonstrated by previous research using these data (Greenberg et al. 2021; Jackson and Schneider 2022). For example, during our study period, Vermont increased total spending per child by 132% or 4 standard deviations, whereas Georgia increased it by 1 standard deviation. Note that we are not seeking to identify causal effects in these models. The relationships identified in the models will not be causal if there are other unmeasured differences between states or if spending in states and household income are jointly determined. Nonetheless, this approach will account for many differences between and within states, affording a careful description of the relationship between subnational welfare state investment and the arrangement of household income.
Results
Spending in States and Household Income Amounts
Figure 1 shows the difference between mean pretax/pretransfer income and posttax/posttransfer income by percentiles of posttax/posttransfer income among households with children (panel a) and by parental educational level (panel b). The figure combines all study years to present aggregate patterns during 1997–2016. Both panels show that government transfers increased average total household income among households in the bottom income quartile and among households whose head had less than a high school diploma. Among households in the bottom income quartile in a given year, average posttax/posttransfer income was $5,970 higher than pretax/pretransfer income. For context, posttax/posttransfer income ranged from $0 to $34,421 across all study years. During the full study period, roughly half (51.3%) of households in the bottom quartile were classified as being below the poverty line according to the geographically adjusted SPM.
For the three higher income groups, average posttax/posttransfer resources were lower than pretax/pretransfer resources because, on average, income and payroll taxes reduced incomes more than transfers increased incomes. This net difference grew across the income distribution, from a net reduction of approximately $6,870 for the middle half to $74,890 for the top decile of households. Similarly, among households whose head had less than a high school education, average posttax/posttransfer income was $2,920 higher than pretax/pretransfer income. In contrast, among higher educated households, average pretax/pretransfer income was higher than posttax/posttransfer income, similar to patterns observed for household income. This pattern reflects the fact that many transfer programs are means-tested and increase the total income of households with the fewest resources. Although households with more resources might also receive transfer income via cash transfers and tax credits, these supports do not outweigh the net income reduction through the tax system for higher SES households. Consistent with prior research (Wimer et al. 2020), these findings suggest that the welfare state plays an important role in increasing the average resources available to households at the bottom of the socioeconomic distribution while doing less to increase the resources of those just above the bottom.
Figure 2 shows how pretax/pretransfer income and posttax/posttransfer income vary with state-level spending levels for households in the bottom income quartile (panel a) and among households whose head had less than a high school education (panel b). How do average total household resources and the gap between pretax and posttax/posttransfer resources vary across the distribution of state-level spending? We separate total spending in states (in thousands per child) into terciles to show low-, medium-, and high-spending states during our study period. Panel a shows that average posttax/posttransfer income was roughly $7,200 higher than pretax income in the highest spending state years, compared with $4,880 in the lowest spending state-years. A consistent pattern is observed for households with low educational levels (panel b), among whom average posttax/posttransfer income was $4,730 higher than pretransfer income in the highest spending state years. In contrast, posttax/posttransfer income was roughly $1,720 higher than pretransfer income in low-spending state years, on average. It is notable that average pretransfer income was lower in the highest spending state-years among both households that were the lowest educated and those in the bottom income quartile. This pattern suggests that higher spending states were responding to economic need, as expected (Edelstein et al. 2016).
Spending in States and the Sources of Household Income
The results so far demonstrate that government transfers raised the average total income of households with the lowest income and educational levels and that the difference between posttransfer income and pretransfer income was larger in high-spending state-years. How multidimensional are the income sources among households? Figure 3 shows that transfer income was more multidimensional among the bottom income quartile and lowest educated households than among other groups. Households with the fewest resources received transfer income from a mix of food assistance (SNAP), housing support, cash transfers (TANF, other cash assistance payments and services, SSI, Social Security, unemployment and worker's compensation), and tax credits (EITC, CTC, additional child credit). By contrast, households with higher income and educational levels received almost all transfer income via cash transfers. This pattern was more pronounced by income percentiles (panel a) than by educational levels (panel b), given heterogeneity in income within educational groups.
How did the income support sources vary according to levels of spending in states? Panel a of Figure 4 shows that mean income amounts from most sources (especially cash transfers, tax credits, and food assistance programs) were higher in higher spending state-years and that the share of income from all transfer sources was steady but slightly rising across the distribution of state-level spending. In other words, there is more support for the idea that high state-level spending increases transfer income broadly across most public support types than for a pattern of higher spending states selectively investing in some types of transfers. Panel b of Figure 4 shows the program shares of total transfers across the state-level spending distribution. This panel reveals that food support tended to constitute a higher share of spending in lower spending state years, likely because food assistance programs (SNAP) receive more of their funding from federal sources than some other transfer programs (e.g., EITC, CTC, unemployment) that allow states the flexibility to implement their own supplemental programs.
The results shown in Figures 1–4 suggest that these patterns—higher levels of transfer income from multidimensional sources in high-spending state-years—should apply most strongly to households with lower income and educational levels. Figure 5 reveals that the average amount of transfer income is substantially higher among all households with children in the highest spending state-years, and this pattern is especially pronounced among households with fewer resources. For example, for the highest state-level spending (in the top third of the distribution relative to the lowest), the average transfer income amount for a household whose head had less than a high school education was $692 higher from nutrition programs, $649 higher from housing subsidies, $606 higher from cash transfers, and $929 higher from tax credits (panel b). Transfer income was also higher in state-years with larger spending amounts for households whose head had a college degree. However, among this group of households, the larger amount of transfer income was driven especially by increased support from cash transfers in higher spending state-years rather than by a more varied mix of state transfers. Cash transfers received by higher educated households come from various programs but especially unemployment insurance and Social Security payments, which make up 50% of cash transfers (on average) among such households.
How Robust Is the Relationship Between Spending in States and Household Income?
The results so far suggest that a larger state-level safety net is associated with higher household income among households with the lowest income and educational levels, and pretax/pretransfer income is higher than posttax/posttransfer income among households with more resources. In addition, transfer income sources are quite multidimensional among households with the lowest income and educational levels. However, these patterns could be conflated by cross-state differences other than the size of their transfer programs, such as the level of economic need in the population or the strength of the local labor market. To account for some of the differences across states and years that could affect the allocation of household income, we estimate regression models that include state and year fixed effects to control for fixed differences across states and to separate the effects of government investment from economic need.
The panels in Figure 6 show the results of regressions predicting transfer income from nutrition, cash, and tax credit programs, respectively, as a function of state-level spending on each source of support, household and state-level control variables, and state and year fixed effects. In this part of the analysis, we focus on spending measures that are most likely to be related to short-term household income: cash, tax credits, and nutrition support. We do not analyze education spending separately because although educational investments are strongly implicated in human capital accumulation over the life course and might indirectly affect household income, they are less likely to be directly related to contemporaneous household income. In addition, health investments might affect household income by reducing out-of-pocket expenses, but it is difficult to quantify the additional household-level income associated with health investments. Finally, we do not separately examine housing because our household-level measure of transfer housing income corresponds less closely to the state-level measure of housing spending (which also includes broader community development spending) than is the case for the other components of household income. However, in earlier analyses, transfer income for housing support is included in the measure of total posttransfer income.
Because of space constraints, we show results only by educational levels, but the pattern of results by pretransfer household income is substantively identical. The results consistently show that as spending in states in each area increased, the predicted amount of transfer income in the same area increased markedly among households whose head had less than a high school education but did not increase (or decreased) among households whose head had a college degree. Table 1 shows that a $1,000 increase in SNAP spending per child was associated with an average increase of $1,076 per year in income from food and nutrition resources among a household with an average educational level.6 Average state-year spending on SNAP was $271 per child during our study period. Panel a of Figure 6 shows this result across standard deviations of state-level spending on SNAP. At high levels of spending on nutrition assistance (2 standard deviations above the mean), average predicted income from nutritional support programs was $2,037 among households whose head had less than a high school education, compared with $293 among households whose head had a college degree. This gap was substantially larger in the highest spending state-years than at the bottom of the state-spending distribution, where the gap in nutritional support transfer income between the highest and lowest educated groups was $295.
The results for transfer income (panel b, Figure 6) show that predicted cash transfer income increases with spending on cash programs, such that predicted cash transfer income among households whose head had less than a college degree was more than 25% higher in the highest spending versus the lowest spending state years. Increases in cash transfer income as spending rises were especially concentrated among households whose head had a high school or some college education relative to households whose head had less than a high school education. However, unlike the other forms of spending we examine, the benefits of state investments in cash transfers were not strongest among the lowest educated or lowest income quartile households. Panel c of Figure 6 shows the analogous figure for state-level spending on tax credits. Again, as spending on tax credits increased, household income from tax credits increased for the less educated groups (those with less than a college degree).
To increase confidence in the multivariate estimates, we conduct several supplementary analyses. First, we estimate regression models that drop cases with negative or below-mean residuals, as Jakiela (2021) suggested. Results are substantively identical. Second, we estimate exploratory difference-in-differences models (using didregress in Stata 18) that treat the spending measures as continuous measures of treatment intensity. Models estimated separately by household education yield identical results. Because our spending measures most likely do not satisfy the assumptions accompanying an exogenous treatment variable, it is important to emphasize that these models are exploratory and cannot be interpreted as causal effects. Third, we estimate models using a state-level spending measure that is lagged by one year (e.g., where spending in 1999 predicts household income received in the calendar year 2000). These results are substantively identical.
Discussion
Debates about the benefits of safety net spending for the economic well-being of U.S. households permeate academic and policy arenas. Despite convincing comparative and national-level evidence demonstrating the benefits of welfare state spending for poverty reduction, our understanding of the relationship between state-level investments and household income at the smaller level of U.S. states—which demonstrate dramatic variability in the size and scope of safety net spending—is much more limited. We advance understanding by providing a detailed description of transfer income sources across U.S. states among households with children and how these sources vary by state-level spending decisions. How do widely variable levels and types of state investments manifest at the household level in terms of both the amount and sources of income? Which types of public investments have especially contributed to the total resources of households with children, and how do these patterns vary across the socioeconomic distribution? Understanding how different types of state investments are related to household income components is important because states have considerable control over the size and generosity of public support (Montez et al. 2020) and because policymakers continue to debate how to support families via the welfare state.
Using annual data from the 1997–2016 State-by-State Spending on Kids Dataset linked to data from the Census Bureau's CPS-ASEC, we found that government transfers raise the total income of households with the lowest income and educational levels and that transfer income among these households comes from a greater variety of sources. In contrast, government transfers do not raise the total income of households with higher income and educational levels. This finding is consistent with prior research at the national level (e.g., Wimer et al. 2020) that demonstrated the role of the welfare state in increasing the resources available to households at the bottom of the socioeconomic distribution.
Transfer income also comes from a greater mix of sources among households in the bottom income quartile and in the lowest educational group, who receive income from a mix of food assistance, housing support, cash transfers, and tax credits. In contrast, households with higher income and educational levels receive almost all transfer income via cash transfers. This pattern is consistent across different spending levels in states, supporting the idea that state investments increase transfer income among low-SES households broadly across many types of public support rather than via one type.
Finally, in relying on variation within and across states, we find that higher area-specific spending in states is associated with an increase in corresponding transfer income (e.g., nutrition assistance, cash support, and tax credits) among households whose head had less than a college education and among households in the bottom of the income distribution. In contrast, college-educated households and households in the top quarter of the income distribution experience no increase (or a decrease) in transfer income as spending increases. The benefits of increased spending on cash transfer programs, although positive, are less pronounced among households with the lowest levels of education (less than high school) and income than among households with slightly higher educational and income levels. This finding is consistent with recent evidence suggesting that cash spending programs are a decreasingly prominent part of the social safety net for the lowest SES families in the United States (Hoynes and Schanzenbach 2018; Moffitt 2015). Following welfare reform in the 1990s, cash welfare is now a very small portion of the income received by low-SES families (Moffitt 2015). In contrast, as Figure 5 documents, cash transfers make up a large share of the transfer income received by higher SES households, especially from unemployment insurance and Social Security payments.
It is important to acknowledge some limitations to our analysis. First, we rely on self-reports of sources and income amounts in the CPS-ASEC data. Income is not measured perfectly in household surveys, with evidence of the over- and underreporting of income (Meyer et al. 2022) as well as the undercounting of populations with low or no incomes (Schmitt and Baker 2006). If household income were reported more fully and accurately, we might see a larger role of the local welfare state in augmenting multidimensional resources among households with low income and educational levels, but it could be offset by a large population receiving little to no transfer income (e.g., undocumented and unhoused populations). Second, although states have become increasingly important actors in shaping policy, the federal government continues to play a large role in shaping public policy, and the local level is also relevant for certain types of government investments. Our measures of spending in U.S. states also include spending from federal and local sources. Future research could build on our state-level focus by additionally examining investments at the federal and local levels. Third, given the difficulty of estimating the cash/income value of Medicaid or CHIP benefits, we do not separately consider the share of household income that comes from health-related investments. It would be useful in future research to examine how spending in states on public health insurance programs is associated with out-of-pocket medical expenses as a proxy for the household savings that might be associated with state health insurance investments.
Despite these limitations, our findings demonstrate strong relationships between spending in states and the resource pathway that is hypothesized to connect the social safety net to family well-being. State investments in families do matter for the resulting resources that households command across states. Our results suggest that increases in spending disproportionately benefit the budgets of households with the lowest resources and might be a promising means to reducing household resource gaps by education and income. An important caveat to this finding is that transfer income constitutes only one source of household income, especially among higher SES families. Although they receive less income via transfers, households with more education and income are more likely to derive income from private sources, including family wealth, savings, housing, and other assets (Pfeffer 2018).
Future research could usefully extend this analysis to examine other group differences in transfer receipt. For example, program-specific eligibility rules for foreign-born populations might affect the ability of immigrant populations to benefit from the social safety net, particularly in the post–welfare reform era (Bitler and Hoynes 2011). Second, it would be beneficial to examine measures of income inequality at the state-year level to consider whether the relationships observed here translate into lower overall resource gaps. Third, beyond our focus on the direct resource pathways by which state investments allow families to make ends meet, other sources of spending (e.g., education spending) might indirectly affect household income components by affecting human capital accumulation over the life course. Tracing those indirect, longer term pathways is also an important avenue for future research. Finally, a related body of research sought to quantify the public value of U.S. safety net investments by measuring behavioral responses to policies (e.g., savings and investment behaviors, work hours, and worker productivity) and their impact on government spending and the federal budget (e.g., Hendren 2016). Although our analysis focuses on understanding how household incomes change in response to safety net investments, this interest in the “leaky bucket” is relevant for understanding the aggregate effects associated with the tax and transfer system (Okun 1975).
Acknowledgments
We gratefully acknowledge helpful feedback from participants at Princeton University's Office of Population Research Notestein series, the Columbia University Center on Poverty and Social Policy workshop, and the annual meeting of the Population Association of America. We are grateful to the Population Studies and Training Center at Brown University, which receives funding from the NIH (P2C HD041020), for general support.
Notes
A notable exception is that the United States historically has been a world leader in the state provision of public education (Garfinkel et al. 2010).
In analyses of total household income, we also examine two additional measures of resources: (3) total pretax, pretransfer resources with medical and work and childcare expenses excluded from the resource definition; and (4) posttax, posttransfer resources with medical and work and childcare expenses excluded from the resource definition. We test the sensitivity of our results to the exclusion of nondiscretionary expenses, given that they are fully imputed for much of the time series (i.e., before 2009) and we want to offer a potentially cleaner test of the effect of state spending on household incomes alone, rather than a resources concept closer to disposable income (Zewde and Wimer 2019). Because analyses with these two additional measures yield similar patterns, we show results for only (1) and (2).
Because we are estimating relationships with total household income (or resources) instead of poverty, SPM poverty units (i.e., family-like units that are sometimes smaller than the household) and poverty thresholds do not apply to our analyses.
Taxes are estimated by the Census Bureau using a tax calculator applied to available income information from the survey data, and tax estimates generally increase in precision over time as more is known about various income components. For example, after 2009, the Census Bureau began asking respondents whether they had a home mortgage, allowing the tax calculator to begin estimating mortgage interest deductions after that year (the mortgage interest deduction is not included in estimates before 2009–2010). Other income elements in the SPM’s definition of resources are based on respondents’ self-reported receipt amounts and monetary value (e.g., SNAP) or on valuation of self-reported benefit receipt only (e.g., WIC, housing subsidies). The researchers at Columbia University use various imputation techniques to fill in missing data elements historically (for full details, see Fox et al. 2015). However, because our study period begins in 1997, most income elements underlying the SPM concept of resources are directly observed in the census data (or simulated with a tax calculator in the case of posttax income).
During our study period, the Census Bureau designated a “householder” for each household in the CPS-ASEC, typically the household member listed on the household’s lease or mortgage. For details, see https://www.census.gov/programs-surveys/cps/technical-documentation/subject-definitions.html#householder. We deviate from using the householder only in the case of educational level, given that higher education often connotes socioeconomic advantage, and we sought to classify households’ educational status according to levels of such advantage or disadvantage.
The main effects of the de-meaned interaction terms can no longer be interpreted at the zero point of the moderator and instead should be interpreted at the means of the moderator.
