Abstract
The efficacy of social citizenship is determined by the generosity and the inclusiveness of assistance, and by the terms under which such assistance is offered. In the United States, social policies have historically been characterized by deference to private markets, categorical tiers of eligibility, and marked variation across local and state jurisdictions. In this article, we describe and characterize the reconfiguration of American social provision over a three-decade span (1994–2022). Using the State Safety Net Policy (SNNP) dataset, we employ comparable measures of the generosity of assistance and the inclusiveness of receipt across nine safety net programs for low-income families with children. These measures allow us to detail program trends, assess the extent of cross-state variation in provision, and tease out the implications of both for substantive social citizenship in the United States.
Social citizenship, as an aspirational analogue to political or civil rights, denotes the rights of citizens to an unconditional and minimum standard of civilized living (Marshall 1950). Social policies, in this respect, ensure a rights-based claim to well-being and support “for everybody and by everybody” (Titmuss 1968, 42) based on collective membership, solidarity, and recognition (Handler 2009; Somers 2008; Gordon and Fraser 1992; Dean 2015). They feature terms of receipt that are inclusive and unconditional, conferring both autonomy and recognition (Phillips 2022; Segall 2005; Fraser 1995), and they provide benefits sufficient to sustain a “modicum of security” without imposing any reciprocal obligations, such as work (Marshall 1950, 11) or other behavioral conditions (Mead 2008). Conceptualized in this way, the efficacy of social citizenship is determined by its boundaries (the scope or inclusiveness of assistance), its terms (the conditions under which assistance is offered), and its benefits (the generosity of assistance). The pursuit and protection of such claims are especially important in modern capitalist democracies, in which commitments to civil or political equality are often in tension with the inequality of market outcomes.
In the US, the commitment to social citizenship has always been thin and ambivalent. American social policy—measured against either those of its peers (Alper et al. 2021; Hacker 2004; Brady et al. 2017) or its own history (Fox et al. 2015)—is characterized by a weak commitment to social protection. The benefits of American social policy—especially in means-tested programs—have always been residual responses to market failures, calibrated to low-wage labor markets, often conditional on labor force participation (Piven and Cloward 1993; Bahle and Wendt 2021; Esping-Andersen 1990) and structured on narrowly contractual terms (Gordon and Fraser 1992). The boundaries of social citizenship have always been narrowly categorical and stratified by private, social insurance, and means-tested forms of provision (Schneider and Ingram 1993; Mettler 1998; Hacker 2002; Meyers 2007). The historical exclusion of Black Americans on all dimensions of citizenship consistently undermined any pretense of universalism in state or national policies (Lieberman 2003; Miller 2021).
Fragmentation, inadequacy, and inequity, in turn, have been compounded by decentralization. In most means-tested programs, important decisions shaping the boundaries, benefits, and terms of social provision are determined by subnational policymakers and administrators (Moynihan et al. 2022; Bruch et al. 2018; Soss et al. 2001), yielding wide state-to-state variation in policy design and administration and, consequently, in the substantive meaning of social citizenship. The ragged boundaries of social citizenship are both categorical and jurisdictional. Not only are programs not available to all citizens on equal terms or conditions, but the form and function of that categorical fragmentation also varies by region, state, and locality (Banting 2006; Keating 2009). The United States, in this respect, represents not just one “world of welfare capitalism” (Esping-Anderson 1990) but fifty worlds, or jurisdictional settings, in which social policy both challenges and abets the structure of social inequality.
This article examines the patterns and consequences of “ending welfare as we knew it,” underscoring the persistent weaknesses of American social provision, the fraught and fragile trade-off between work and welfare (Edin and Lein 1997), and the ways these have been compounded and transformed over the last three decades. In this respect, we make two distinct and interrelated contributions. First, we describe and characterize changing patterns of American social provision across several programs aimed at low-income families with children and over a three-decade span (1994–2022) punctuated by welfare reform (the passage of the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 [PRWORA]), the Great Recession (2007–2009), and the COVID-19 pandemic (2020–2023). Using the State Safety Net Policy (SNNP) data, we employ comparable measures of two key dimensions of social provision—the generosity of assistance and the inclusiveness of receipt—that capture the benefits and boundaries of social citizenship and document their expansion and contraction over time. Toward this end, we examine program-specific trends in nine safety net programs for low-income families with children and detail the policy and program changes that explain these trends. Social provision after 1996 meant not just a retreat from the already limited means-tested cash assistance but a reconfiguration that rippled across a wide range of policies and programs. Taken together, these changes diminish the benefits of social citizenship, erode the autonomy of its recipients, and narrow the categorical boundaries of receipt.
Second, to understand the “devolution revolution” associated with PRWORA in a broader context, we examine the extent of cross-state variation in provision, its trajectories (decreasing or increasing, converging or diverging), and its regional patterns. We describe these cross-state patterns in relation to the structure, logic, and mechanisms of state discretion in these programs, and the ways in which states use that discretion in administration, financing, and rulemaking. State discretion in federal programs, as we explore in more detail later, yields social policies that vary on the key dimensions of social citizenship (inclusion and generosity) across jurisdictions. We conclude by discussing the implications of policy design and policy variation for program inclusion and generosity and—more broadly—for substantive social citizenship in the US.
THE MORE THINGS CHANGE …
The recent history of American social policy is marked by continuity and change. This reconfiguration of American social policy was galvanized by passage of the PRWORA in 1996, which both confirmed and disrupted long-standing patterns of unequal protection (Moffitt 2015; Ziliak 2016; Blank 1997). The result is an array of social policies marked by three fundamental shifts in the form and function of American social citizenship: first, the terms of receipt have become increasingly conditional and coercive; second, the boundaries of inclusion have narrowed (with a particular focus on children); and third, the adequacy of benefits has been eroded by a retreat from cash assistance in favor of in-kind or work-incentivizing benefits, work supports, and temporary provisions. Each of these shifts, in turn, has been both enabled and exaggerated by the discretion afforded to state governments in setting the boundaries, terms, and benefits of social citizenship.
Conditioning the Terms of Receipt
Since the 1990s, the persistent weakness of American social provision has been sustained and intensified by a new regime of coercive and conditional poverty governance. Conditional assistance not only chokes access to social protection, but it also corrodes personal and familial autonomy by attaching strings to receipt—narrowing options, imposing paternalist expectations, and subjecting recipients to intrusive patterns of surveillance and enforcement (Soss et al. 2011). This loss of autonomy is no accidental consequence of policy retreat or diminished support. Conditions are explicitly designed (from various perspectives) to influence, nudge, regulate, or discipline recipients, and to match social provision with civic obligations (Dean 2007) or behavioral expectations. The beneficence of public programs is accompanied by coercive and punitive supervision such that—as T. H. Marshall famously observed of the English Poor Laws—claims to assistance “could only be met if the claimants ceased to be citizens in any true sense of the word” (Marshall 1950, 15). Autonomy (and with it civic recognition) is threatened by the increasingly coercive and conditional logic of assistance, by the disjuncture between work expectations and work opportunities, and by the preference for in-kind benefits and services.
The policy shifts of the mid-1990s compromised autonomy in myriad ways. While labor market participation and various behavioral expectations have long been central to American social provision (Bertram 2015; Bell 1967), these were now harnessed to a much more punitive and coercive structure of social provision—designed both to limit direct assistance and to shape the behavior of its recipients. “Welfare-to-work” commitments, across modern welfare states, are premised on both a shifting ethic of responsibility or reciprocity (Segall 2005; Anderson 2004) and on the “activation” policies necessary to support labor market participation (Dean 2007). PRWORA, for its part, was animated by heightened attention to the labor force participation of low-income parents, to the potential work disincentives of receiving cash assistance (for reviews, see Moffitt 1992; Ziliak 2016), to the broader behavioral norms of low-income families, and to the alleged role of government assistance in facilitating dependency (Aizer et al. 2022; Handler 2009; Bertram 2015). Deference to market relations now imbues not just the logic of social provision, but its administration as well, and the expectation of labor market participation is currently enforced with punitive conditions and sanctions (Soss et al. 2011).
At the same time, these work expectations have not been accompanied by sufficient federal commitments to the work supports (childcare, transportation assistance, paid sick and family leave, minimum wage laws, predictive work schedules) that would allow low-income parents to make ends meet (Gornick and Meyers 2009; Gornick et al. 2022). They devalue parenting as a social contribution (Anderson 2004) yet offer no guarantee of access to alternative childcare arrangements (Herbst 2023; Meyers et al. 2002; Morrissey et al. 2023), especially those demanded by the temporal vagaries (shiftwork, uneven scheduling) of low-wage labor markets (Kwon et al. 2026, this issue). The “work-first” approach privileges labor market participation over training or education (Lafer 2002), while weak labor policies and collapsing job quality (Kalleberg 2009; Rothstein and Zipperer 2020) ensure that the resulting employment offers little security. State and local policies have filled some of this gap, especially with respect to minimum wage, paid leave, and predictive scheduling, but the unevenness of these policies—compounded by their outright preemption in many states—has only widened state-to-state disparities in social protection or provision (Briffault 2018; Gerken and Tyler 2022). Taken together, the expectation of labor and the weakness of work supports dramatically constrain the autonomy of low-income parents and their families, narrowing their options for social assistance or self-support. Indeed, the invocation of personal responsibility in this era ran alongside a marked deterioration in the ability of working parents (custodial or noncustodial) to sustain job-based economic security (Gonalons-Pons et al. 2026, this issue).
The threat to autonomy was much more expansive than TANF’s new work requirements. Work expectations bled into the wider logic of social policy, recasting support around programs or policies that rewarded (and subsidized) labor force participation such as the Earned Income Tax Credit (Halpern-Meekin et al. 2015) and pushing the expectation of labor force participation into other arenas of social policy (Bauer et al. 2018). Other significant features of the post-PRWORA welfare system—including temporal limits and family caps on assistance, sanctions, and the machinery for enforcing those sanctions (Headworth 2021)—narrowed options for self-support and exposed recipients to an intrusive surveillance reminiscent of the “man in the house” rules that prevailed before the 1970s (Kohn 1970; Gordon and Batlan 2011). Growth in child-focused programs included efforts to bolster private responsibility for children’s income support through increased enforcement of child support orders and the use of child protective services as a conduit for assistance (Fong and McCarthy 2026, this issue). Coercion, punishment, and the mantra of personal responsibility animated child support enforcement, which is a private transfer between parents and relies heavily on disclosures or garnishments mandated by other social policies (Meyer et al. 2020; Pate 2016).
Narrowing the Categorical Boundaries
Alongside these shifts in the terms of receipt, the reconfiguration of social policy also narrowed the boundaries of social citizenship. American social citizenship is already stratified into private, social insurance, and means-tested tiers, with the latter increasingly narrow in scope. In addition to conditional constraints, eligibility is increasingly connected to labor force participation or parenthood. The increasing focus on supporting children after the mid-1990s (Aizer et al. 2022) marked not just an effort to target those long considered the most vulnerable and deserving, but also a pointed neglect or deflection of the claims made by others. Childless adults not “categorized into assistance” by virtue of disability, unemployment, or military service (Bruch et al. 2023; Stone 2005; Moffitt 2015) are by and large left behind by most forms of social protection (Parolin et al. 2023; Brady and Parolin 2020; Edin and Shaefer 2015).
As low-income parents experienced the retraction of unconditioned assistance, support targeted at children, particularly child health insurance and public pre-K, became relatively more inclusive and generous. The relative (if uneven) generosity of programs like child health insurance or social security income (SSI) for disabled children underscored both a renewed attention to the implications of social provision for child development (Hahn et al. 2016) and the stability of claims made on behalf of children. But it also underscored a devaluation of the very obligation—parenting—that justified assistance for low-income families in the first place (Anderson 2004). This reflects the persistence of a long-standing tension between the unquestioned importance of providing for children and the reluctance to reward parents whose perceived failures (in the market or in the family) made such provision necessary (Gordon 1994). US social policies, in this respect, continue to exhibit “an official honoring of motherhood combined with a distrust, disdain, even contempt, for women who do it” (Gordon 2001, 23).
Inadequacy of Benefits
The inadequacy of social assistance reflects, in large part, its persistent and long-standing calibration to labor markets: expanding support when labor markets are slack, as Frances Fox Piven and Richard A. Cloward (1993) and others have argued, and restricting it when low-wage labor is in demand. Since 1996, regulating the poor has become a much blunter strategy, vested less in the administration of social programs like Aid to Families with Dependent Children (AFDC) and more in one-off infusions of generosity or expanded eligibility at moments of economic crisis. This approach underscores the inherent weakness of regular social programs and the limits and fragility of social citizenship.
The new limits on American social citizenship are also evident in the increased reliance, by both state programs and those in need, on temporary program adjustments or programs cobbled together at moments of economic crisis but abandoned thereafter (Jackson et al. 2022; Congdon and Vroman 2022; Wimer et al. 2022). During the Great Recession, temporary provisions of the American Recovery and Reinvestment Act (ARRA) of 2009 mandated higher monthly benefits for families participating in Supplemental Nutrition Assistance Program (SNAP) (Gunderson 2015; Bitler et al. 2020; Wiseman 2019) and increased the generosity of unemployment insurance (UI) benefits primarily by increasing the length of time a person could continue to receive benefits and making these benefits more widely accessible for unemployed workers (Congdon and Vroman 2021, 2022). Similarly, the COVID pandemic was met largely with temporary federal programs, including stimulus payments, tax credits, and an eviction moratorium (Jackson et al. 2022; Bitler et al. 2023). In UI, federal infusions were extensive and encompassed not just conventional recessionary extensions but also a bump in benefits for all recipients and benefits for those not normally eligible (Congdon and Vroman 2021; Bell et al. 2023). Food assistance, including SNAP and school-based meal programs, was again employed as a countercyclical response. Suspension of reviews for SNAP and public health insurance during the pandemic yielded steady growth in inclusion. The uptick in generosity and inclusion was stemmed by the determination of many states to withdraw early from the pandemic UI programs (Coombs et al. 2022) and the unwinding of presumptive continued eligibility for SNAP and health insurance (Ku et al. 2022; Arbogast et al. 2024). Such protections—however fleetingly generous—suggested not a commitment to adequacy, universality, equity, or autonomy but rather a slimmer stopgap interest in social order and economic recovery. Such policies concede the inadequacy and unresponsiveness of our social policies but do little to address them.
Decentralization of Social Provision
The reconfiguration of American social policy occurred in a context in which state and local governments had varying degrees of discretion in financing, rulemaking, and administration in many of the safety net programs that serve as critical supports for low-income families with children (Bruch et al. 2018). American social policies have always been premised on deference to economic and social interests in state and local jurisdictions. As a concession to both constitutional concerns and the racially exploitive labor markets of the South, the means-tested titles of the 1935 Social Security Act (Old Age Assistance; Aid to Dependent Children; Aid to the Blind) and the UI program had few substantive conditions on state plans and participation (Gordon 1994; Lieberman 1998; Karch and Rose 2019). After 1935, such discretion ensured wide variation—in both policy design and distributional outcomes—across the states. Going forward, the inequity and unevenness of decentralized and discretionary social policies remained central concerns (Tani 2016). This was evident in several historical developments: the “War on Poverty” programs that bypassed state governments altogether (Quadagno 1994), the “welfare rights” movement of the 1960s and 1970s (Chappell 2010), and the punitive “backlash” that gained force in the 1980s and 1990s (Kohler-Hausmann 2017). Across this history, state and local policy choices, which relied on the police power, were at once less invested in redistribution or equal protection and more invested in local (often punitive) regulation and market relations (Cashin 1999; Weir 2005; Mettler 1998; Miller 2021). In turn, regionally uneven commitments to social rights have been irretrievably racial in their logic, pattern, and impact (Lieberman 1998; Fox 2012). The passage of PRWORA embraced and formalized this commitment to decentralization, granting states broader discretion in the form and level of basic assistance, while simultaneously narrowing state discretion regarding the terms of receipt (Bruch et al. 2018).
Key policy choices, even in putatively federal social programs, are increasingly made by state and local jurisdictions (Bruch et al. 2018). State discretion varies across programs, across program dimensions, and across policy indicators (see table 1).1 Discretion in financing reflects both the federal-state share of program costs and any accompanying incentives or thresholds: state discretion is high in programs where there is no federal role (state taxes) or where federal block grants come with few strings attached (TANF); state discretion is medium where programs rely on state and federal revenues (UI) and where federal block grants are more closely proscribed (childcare); state discretion is low where the spending (and any accompanying rules) are largely federal (SNAP and SSI). Discretion in rulemaking reflects federal standards for eligibility, benefit levels, and other program elements, and the room afforded to states to define their own rules or standards through legislation or administrative practice (Dardanelli et al. 2019): discretion is high (or medium-high) where states largely control eligibility and benefit levels (TANF, childcare); medium where federal rules or standards set narrower parameters around state choices (UI, child health insurance); and low where the states have little influence on the terms of receipt (SNAP, SSI). Discretion in administration is shaped by both the direct jurisdictional responsibility (federal, state, local) for interpreting or carrying out the provision of assistance, and by the degree to which state or local agencies are held accountable to program expectations and rules: discretion is high if program responsibility is vested largely in state agencies subject to broad federal guidelines (TANF, childcare); medium if those federal standards include tight compliance or performance standards (as in child health insurance); and low if state programs are administered through federal agencies (SSI) or are subject to regular federal oversight.
State Discretion in Safety Net Programs
The implications and impacts of this discretion are profound. Where federal programs or high federal standards are absent, social policies and labor standards are left to the uneven willingness and capacity of state and local governments (Bruch and Gordon 2022; Freeman and Rogers 2007; Briffault 2018). The range and mechanisms of state discretion vary over time and across programs, yielding different constellations of social protection or social rights. Financing discretion (encompassing the relative shares met by federal and state dollars, state fiscal capacities, and the terms or rules under which federal dollars can be spent) yields cross-state variation in generosity and inclusion. Depending on how such financing is structured, it can discourage innovation and expansion, as in the case of unemployment insurance (Karch and Rose 2019), or encourage states (or at least some states) to be more generous and inclusive, as in the case of child health insurance. State discretion in rulemaking shapes both inclusion (eligibility) and generosity (benefit levels). Decentralization is, in and of itself, an engine of inequality: under the right conditions, it might enable more robust social protection, but such conditions—including high federal standards, robust protections against discrimination, and equalization of state fiscal capacities—are rare and hard to sustain (Bruch and Gordon 2022; Freeman and Rogers 2007; Miller 2021; Gerken and Tyler 2022). Higher levels of administrative discretion generate wider variation in inclusion, evident not only in TANF, child support, childcare and early childhood education but in otherwise low-discretion programs (like SNAP), which allow states to determine administrative rules, guidance, and practices in relation to processing applications and determining eligibility (Herd et al. 2023). In our polarized federalism, some states forgo even the temporary programs implemented in extraordinarily challenging times.
Taken together, the increasingly conditional nature of social assistance, its narrowing categorical boundaries, and its renewed decentralization have reshaped the inadequacy and fragmentation of American social policy. They undermined the responsiveness of some programs (Bitler and Hoynes 2016; Bitler et al. 2020) while bolstering others, adjusted the targets of social protection, and tethered receipt increasingly to labor force participation. As welfare reform recast policy goals from poverty alleviation to behavior-focused outcomes (Ziliak 2016; Blank 2018), social provision remained inadequate, deepening the precarity of those no longer eligible for assistance and allowing high rates of poverty and food and housing insecurity even among those receiving assistance (see also Brady and Parolin 2020; Edin and Shaefer 2015; Bruch et al. 2023). The use of paternalistic conditions and punitive sanctions, coupled with the meagerness of work supports, effectively compelled and subsidized low-wage employment (Paz-Fuchs 2008; Heinrich and Scholz 2009; Piven and Cloward 1993; Soss et al. 2011). More assistance for some meant less for others: the boundaries of social citizenship remained categorical and conditional; the benefits increasingly focused on in-kind services and tax credits; and both fragmented across jurisdictions with disparate motives and capacities.
All of this yields a social citizenship that is more conditional in its terms of receipt, more narrowly categorical in its boundaries of coverage, and less adequate in its reliance on conditional, in-kind, and temporary benefits. Decentralized policy and provision compounded each of these shifts, inviting variation in policy design, administration, and outcomes—and ensuring that even efforts to bolster social provision or economic security in some state or local settings would come at the expense of equity in social citizenship across those settings.
DATA AND MEASURES
We employ the SSNP dataset to measure two key dimensions of social provision—generosity and inclusion—across nine safety net programs that influence the economic resources of economically marginalized working-age adults and their dependents either directly (by providing cash) or indirectly (by providing other goods or services), and in which states have some degree of discretion in financing, rulemaking, or administration (Bruch et al. 2018). The SSNP includes yearly state-level estimates of generosity and inclusion from 1994 through 2022 for cash assistance (AFDC/TANF), food assistance (Food Stamps–SNAP), child health insurance (Medicaid and Children’s Health Insurance Program [CHIP]), child support enforcement, childcare subsidies (Child Care and Development Block Grant-Child Care and Development Fund [CCDF] and TANF), early childhood education (Head Start and state pre-K programs), Unemployment Insurance (UI), child disability assistance (SSI), and state income taxes.2
The generosity of benefits (or amount spent or received per case or recipient) is calculated as the average benefits reported by program or calculated by dividing total benefit spending by a state’s caseload or number of recipients (table 2 provides a description of the construction of each policy indicator, including data sources). The generosity indicators are adjusted for cost-of-living differences over time (that is, inflation) using the Bureau of Labor Statistics Consumer Price Index research series using current methods. For the 2024 release, the generosity measures are adjusted to 2022 constant dollars.3 The generosity measure is designed to capture the actual receipt or experience of social assistance. Unlike expenditure or “welfare effort” measures (Moffitt 2015), we employ a population denominator to calculate a per-person or per-case benefit level. Unlike measures modeling receipt based on single statutory policies such as benefit maximums (Aizer et al. 2022) and uniform take-up rates (Schmidt et al. 2025), we estimate benefits based on actual spending and rates of receipt. Generosity, as a measure of benefits received, is not calibrated to any threshold of adequacy.
Social Safety Net Policy (SSNP) Measure Descriptions and Data Sources
The inclusiveness of receipt (or the proportion of the potentially needy or eligible population receiving benefit) is calculated by dividing the number of program recipients in a state by the number of potentially needy individuals or families in the state. The estimates of potentially needy individuals or families are calculated using three-year moving averages from the Annual Social and Economic Supplement (ASEC) of the Current Population Survey (Ruggles et al. 2025). For means-tested programs, the estimate of the potentially needy is the number of individuals or families who (a) fall into categorically eligible groups and (b) have market pretax, pre-transfer incomes below the federal poverty threshold, or below some percentage of the threshold depending on the income eligibility criteria of the program.4 The inclusion measure combines reported caseloads with a population denominator capturing the reach of a program. The population denominators, in turn, are based on broad need-based categories rather than narrow programmatic criteria, on who should be covered rather than on who is technically eligible.
The SSNP measures, comparable across multiple programs, offer a holistic and capacious view of social policies as they are actually provided and received. They capture persistence and change in the governance of social policy—including deference to labor markets, devolution, and delegation. By focusing on inclusion and generosity, they provide a more accurate measure than expenditure data of the ways in which social protections are structured and administered in state and federal policy and experienced by their recipients. They facilitate comparison across programs and over time, effectively harmonizing data for multiple programs administered through different entities at the state and local levels. And they facilitate consideration of the relationship between programs, of the ways in which multiple policies are configured or reconfigured, and the ways in which such configurations (or packages of support) are experienced by recipients.
These measures are designed to capture the structure and exercise of state and local discretion across these programs, but also to turn our attention from discrete and formal policy choices to the range of decisions and practices, and to the diverse mechanisms of policymaking and implementation that determine the resulting scope and generosity of social provision. They reflect statutory policy choices (such as maximum benefits), administrative rules (such as those regarding eligibility), and operational procedures and practices that collectively determine the generosity and inclusion of social provision for low-income families with children. Capturing policy choice, design, and administration is especially important in social programs that devolve authority to state or local governments. States create plans that meet the strict or broad guidelines of the authorizing federal policy through an administrative process. These state plans reflect a plethora of decisions made at the state level by legislative policymakers and executive agency administrators. They shape the rules and processes that dictate program access and benefits, with administration often allowing some discretion in decision-making and implementation practices for frontline workers. With these measures, we aim to capture a fuller picture of ways in which decisions are made—by legislators, by administrators, and by front-line caseworkers—regarding the responsibility to render assistance, the goals of social policy, the deservingness of its recipients, and the terms and boundaries of social citizenship. These measures serve as an explanatory foundation for our understanding of deep and persistent inequalities—across jurisdictions, across programs, and across populations—in social provision, and for documenting the connection between subnational discretion and equitable social citizenship.
ANALYTIC METHODS
To describe the benefits (generosity) and boundaries (inclusion) of social citizenship, and specifically in social safety net provision, we use yearly state-level measures and calculate central tendencies (mean and median) from 1994–2022.5 In the results, we primarily report the 50-state median given the skewed distribution of many of the measures. We examine the patterns and trajectories of multiple programs using these comparable measures to capture the level and change over time (with particular attention to upward or downward trends) in the substantive commitment to social citizenship.
To examine the inequity in social citizenship across states, we describe the extent or magnitude of cross-state variation using four measures: standard deviation (SD), coefficient of variation (COV), Gini coefficient (Gini) and the ratio of the 90th to 10th percentile (90-10 ratio). This approach to examining cross-state variation focuses on the distribution of states in a given year. However, to capture the individual state patterns and changes over time, we complement these variation measures with scatterplot graphs, which display the state-specific starting (1994 or earliest year available) and ending (2022 or most recent year available) values. We also describe how cross-state variation has changed over time, and whether those changes represent states converging (becoming more similar) or diverging (becoming more different). We assess these changes by looking at the absolute size of the change and by using significance tests (Levene test for variance, bootstrapping for the COV and Gini).6 Combining the median and variation trends, we characterize programs as upward or downward, converging or diverging from 1994–2022. Finally, for selected programs, we document state-specific values and regional patterns of generosity and inclusion, and how these have shifted from 1994 to 2022.
FINDINGS
We begin with a snapshot view of safety net provision in 1994, prior to the substantial changes brought about by PRWORA (see figure 1 and table 3), and at about the moment when Kathryn Edin and Laura Lein (1997) were collecting their data for the book Making Ends Meet. Looking first at programs that provide income support, we see that the annualized AFDC cash assistance benefit in 1994 was about $7,200 (in 2022 dollars) for low-income families with children in the median state and an average annualized Food Stamp benefit (a near-cash benefit) for households with children of about $4,000, a combination which represented less than 50 percent of the poverty threshold for a family of three (United States Census Bureau 2025). Other income supports betrayed similar inadequacies: the most generous (SSI for disabled children) still fell below half the poverty line ($9,500 annualized average benefit); UI offered an average benefit of less than $5,000 per spell of receipt (average benefit × duration); and for a family of three at the poverty line, state income taxes (including credits) yielded a small net loss or tax liability of about $45. Child support collected through state enforcement agencies resulted in an average transfer from noncustodial parents to custodial parents of just over $5,000. In terms of education and health supports for children, annual per-child spending on preschool and early education and childcare were the most robust (approximately $7,800 and $7,500 in the median state, respectively), while child health insurance spending ($3,300 per child) was substantially less.
Median Trends in Generosity and Inclusiveness, 1994–2022
Source: Authors’ calculation from the SSNP data.
Note: Dotted and solid lines represent income and in-kind support programs, respectively.
Generosity and Inclusion Cross-State Variation, 1994 and 2022
Turning to inclusion, cash and food assistance had the broadest reach in 1994 but even then were received by just under 60 percent and 65 percent of low-income families with children, respectively (see figure 1).7 Just over 40 percent of low-income single-parent families received payments from a child support obligation or fell under the state income tax threshold for incurring tax liabilities, and just over a third of unemployed workers received unemployment benefits. Fewer than two in ten low-income children under age thirteen received subsidized childcare, and the relatively generous SSI program reached fewer than three in one hundred low-income children (defined as up to 200 percent of the federal poverty level [FPL]). In terms of child-focused services, approximately 50 percent of low-income children (defined as up to 300 percent of the FPL) received health insurance through the Medicaid program in 1994 but fewer than 20 percent of three- to four-year-olds were enrolled in Head Start or state pre-K programs.
In sum, social citizenship, as captured by levels and rates of provision in 1994, was a patchwork of inadequate and fragmented support with limited reach into the population of low-income families with children. Narrow and categorical boundaries of assistance underscored uneven inclusion across programs and states; meager benefit levels underscored the uneven and inadequate generosity of assistance for those receiving assistance.
Between 1994 and 2022, a combination of policy and program changes and policy inaction (that is, policy drift) yielded a universe of social policies that increasingly undermined the autonomy of its recipients (both by erecting conditions to direct assistance and by leaning more on in-kind supports), narrowed its attention to low-income children, and offered support (most notably during recessions) with temporary or time-limited benefits. These changes are illustrated by looking at the multi-program trends in the generosity of benefits and the inclusiveness of receipt.
The passage and implementation of PRWORA in 1996 ushered in substantial changes not only to the cash assistance program for low-income families with children but also to food assistance, child support, childcare, and child health insurance (Blank 1997; Ziliak 2016). PRWORA rendered cash assistance conditional and temporary, leading to a sharp decline in the rate of receipt (from 43 percent of low-income families with children in 1997 to 24 percent by 2007). The increase in poverty brought on by the Great Recession underscored the limits of the new program; instead of a corresponding increase in TANF caseloads, caseloads continued to fall, resulting in an almost 20 percent reduction in the rate of inclusion during a period of heightened need. This lack of responsiveness was more troubling when paired with a parallel decline in the average annualized benefit which fell from about $7,250 in 1994 to $5,800 in 2007, to just $4,700 by 2013, and recovered only slightly ($5,100 in 2022) during the COVID recession.
While the decline in cash assistance occurred in all states, there was substantial variation in the levels and trends of support (see figure 2 and table 3). Under AFDC, federal standards were minimal, and states enjoyed broad discretion with respect to the definition of need and corresponding benefit levels. This is reflected in the wide cross-state variation in AFDC generosity in 1994: at the 10th percentile, the annualized average monthly benefit was about $4,000 (in 2022 dollars), but the benefit at the 90th percentile was almost three times that. With the transition to TANF, states were granted nearly unfettered discretion in the expenditure of TANF block grant funds (Parolin 2021), and the extent of cross-state variation in the generosity of benefits increased from a Gini of 0.21 in 1994 to a Gini of 0.25 by 2022, representing a 20 percent increase in cash benefit inequity overall. All states reduced spending per family on cash assistance over the next two decades, but the most dramatic absolute reductions were in the most generous states. This can be seen in figure 3 by comparing the location of the most generous states in 1994 to those in 2022, a decline in generosity of more than $3,700 at the 90th percentile.
Cross-State Variation Trends in Generosity and Inclusiveness, 1994–2022
Source: Authors’ calculation from the SSNP data.
Note: Tax = State income tax; Cash = Cash assistance; CCare = Childcare; Pre-K = Preschool and Early Education; CHSupp = Child support; UI = Unemployment insurance; Health = Child health insurance; Food = Food assistance; SSI = Supplemental Security Income. The left and right ends of each bar represent the 10th and 90th percentiles, respectively.
Turning to inclusion in cash assistance, in 1994 only 38 percent of low-income families with children received cash assistance at the 10th percentile, versus 78 percent at the 90th percentile (see figure 2 and table 3), reflecting states’ broad rulemaking (eligibility) and administrative discretion. With the transition to TANF, states were required to specify terms of receipt related to work requirements, family caps, time limits, and sanctions (Soss et al. 2001; Nadal-Fernandez et al. 2025). As a result, all states experienced rapid declines in inclusion, especially throughout the 1990s and 2000s. By 2022, inclusion had fallen at the 10th percentile by an incredible 89 percent (representing a collapse in the rate of inclusion from almost four in ten families in 1994 to just four in one hundred by 2022). The decline in inclusion at the 90th percentile was also substantial, ending the period in 2022 at 0.43—nearly the same rate as that of the 10th percentile in 1994 (0.38).
The overall pattern is one of downward divergence on both measures: generosity and inclusion both declined at the median, generosity collapsed at the top end of the distribution, inclusion collapsed at the bottom, and cross-state variation increased for both. Comparing the levels of generosity and inclusion of individual states in 1994 to that of 2022 illustrates the downward divergence of both generosity and inclusion, as well as some telling regional patterns (figure 3). Southern states (indicated by the diamond symbols in figure 3), for example, crowded below the median on both measures in both 1994 and 2022—a legacy of the deeply racialized history of cash assistance. By the same token, Northeastern and Western states were the most inclusive and generous in both years.8 The larger patterns here are telling. Broad state discretion under AFDC yielded wide variation with regard to both inclusion and generosity. The transition to TANF increased discretion in financing and administration, while at the same time constraining states in rulemaking (by requiring new conditions and terms).
Cash Assistance, Inclusion and Generosity, 1994 and 2022
Source: Authors’ calculation from the SSNP data.
Note: The same shapes are used to represent US regions across both years. Shapes with dark outlines correspond to 1994, and shapes without outlines correspond to 2022. Animated versions of figures 3 through 7 are available at https://www.rsfjournal.org/content/12/1/34/tab-supplemental.
While traditional cash assistance declined, the federal Earned Income Tax Credit (EITC) (Halpern-Meekin et al. 2015) and tax credits for children (McCabe and Popp Berman 2016) increased. However, these shifts were not just from a transfer to tax mechanism. They reflected a narrowing of the boundaries of provision, effectively restructuring the most generous forms of assistance as rewards for work rather than responses to need (Blank 2018; Edin and Shaefer 2015), and they provided the largest benefits to the near poor, leaving the deeply poor worse off (Parolin et al. 2023).
The one form of income support that remained stable and relatively generous over this time was cash assistance for disabled children (SSI), a program marked by clear (if narrow) boundaries and a high floor of federal benefits. The average SSI benefit remained relatively stable during this period (hovering just above an annualized benefit of $9,000 at the median). Median inclusion remained very low, with between 3 and 4 percent of low-income children receiving assistance from 1994–2022 (figure 1). Cross-state variation in generosity and inclusion was minimal (figure 2), reflecting the limited discretion afforded states. Some states added small supplements to the federal benefit (Erkulwater 2015; Duggan et al. 2016), often to sustain commitments made in pre-1972 programs. Limited discretion in administration also dampened state-to-state variation in SSI inclusion, because the program relied heavily on referrals by intermediary organizations providing health and education services to children, intrastate variation (that is, rural vs. urban) was more profound (Schmidt and Sevak 2017; Wittenburg et al. 2015).
The new regime of work-conditioned assistance rested, at least in theory, on the provision of ancillary supports (especially childcare) that would make labor market participation feasible. However, both the amount provided to subsidize the cost of childcare, and the availability of this benefit have been limited. At the median in 1998, the average benefit was $7,600 which increased only slightly (to about $9,000) by 2020 (figure 1); fewer than two in ten low-income children received subsidized childcare in 1998, a rate that barely budged over the next 24 years. The absence of such support compounded the erosion of autonomy and the narrowing of options for support.
Reflecting a history of different state welfare-to-work programs (Greenberg et al. 2009) and the consolidation of childcare funding streams into the CCDF (Meyers et al. 2002), the provision of childcare varied widely. In terms of generosity, in 1998 (ranging from $5,500 at the 10th percentile to $11,400 at the 90th), a gap that closed marginally by 2020 as overall generosity rose, a slight upward convergence (figure 4). Variation in inclusion was similar but widened between 1998 and 2020, as a small increase at the 90th percentile (from 0.30 to 0.36) was accompanied by a slight decrease (from 0.11 to 0.10) at the 10th (a slight upward divergence). Tellingly, inclusion and generosity in childcare echoed the same regional pattern: those states at the bottom of the distribution in cash assistance (most from the South) also failed to offer meaningful work supports in the bargain.
Childcare, Inclusion and Generosity, 1998 and 2020
Source: Authors’ calculation from the SSNP data.
Note: The same shapes are used to represent US regions across both years. Shapes with dark outlines correspond to 1998, and shapes without outlines correspond to 2020.
The paternal (and punitive) logic of post-1996 social provision was especially evident in the increasingly muscular and aggressive system of child support enforcement and collections. Increased enforcement and collection efforts yielded a substantial increase in the proportion of families with a collection (from 49 percent in 1997 to 82 percent in 2007), but a corresponding decline in the average amount collected per case per year, from approximately $5,400 to $3,700 (figure 1). Post-TANF enforcement mandates narrowed state variation on both generosity and inclusion. Variation in generosity, increasingly shaped by differences in income and judicial support guidelines (Venohr 2013), showed a downward convergence; variation in inclusion, reflecting the dramatic increase in collections, showed an upward convergence (figure 2).
After 1994, social policy increasingly narrowed its categorical boundaries, largely by reserving its attention to children (or families with children), especially in the provision of in-kind benefits and services. In the case of health insurance, CHIP in 1997 and expansions of the eligibility criteria for CHIP and Medicaid, resulted in dramatic increases in inclusion—from 48 percent of low-income children enrolled in 1994 at the median, to 82 percent by 2007, to over 100 percent by 2020.9 While states enjoyed considerable discretion in setting eligibility thresholds, CHIP’s generous financing formula incentivized and enabled them to raise those thresholds, yielding dramatic increases in inclusion across the distribution (figure 2): in 1994 inclusion ranged from 0.34 at the 10th percentile to 0.58 at the 90th ; by 2021 it ranged from 0.90 to 1.47. This dramatic pattern of upward convergence (figure 5) pushed the least inclusive states in 2021 beyond the most inclusive from 1994 and left only six states (three in the Midwest, three in the Mountain West) with inclusion rates under one. The rollout of the CHIP program, alongside other changes in health-care financing and access (including the Affordable Care Act), narrowed state-to-state differences in generosity (the 90-10 ratio fell from 2.03 to 1.78).
Child Health Insurance, Inclusion and Generosity, 1994 and 2021
Source: Authors’ calculation from the SSNP data.
Note: The same shapes are used to represent US regions across both years. Shapes with dark outlines correspond to 1994, and shapes without outlines correspond to 2021.
Food assistance (across several programs including school-based meal programs, the Special Supplemental Nutrition Program for Women, Infants, and Children [WIC], and SNAP) also underscored the shift to in-kind supports targeted at children. In the case of SNAP, inclusion initially declined in the wake of welfare reform but rebounded when the 2002 Farm Bill instituted several changes designed to increase enrollment among eligible populations (Ziliak 2015). Between 2001 and 2007 alone, inclusion at the median grew from 49 percent to 67 percent (figure 1). Lack of responsiveness in cash assistance was contrasted with the continued growth in SNAP inclusion, buttressed by temporary expansions during the Great Recession (Bitler and Hoynes 2016): the number of participating households with children outpaced the increase in family poverty, resulting in a rate of inclusion that grew to 82 percent by 2010 and 105 percent by 2020. As with health insurance, the dramatic increase in inclusion was reflected across the distribution (figure 6).
Food Assistance, Inclusion and Generosity, 1994 and 2020
Source: Authors’ calculation from the SSNP data.
Note: The same shapes are used to represent US regions across both years. Shapes with dark outlines correspond to 1994, and shapes without outlines correspond to 2020.
During this period, cross-state variation increased in both generosity and inclusion (table 3). While SNAP is paid entirely with federal dollars, states can exercise administrative discretion—by investing in outreach or streamlining applications, for example—in such a way as to boost participation (Currie 2003; Gunderson 2015; Ziliak 2015; Ziliak et al. 2015) or to slow program growth and discourage or deny applicants, sometimes called “soft diversion” (Edin and Shaefer 2015). Uneven administrative capacities or intentional administrative burdens (Barnes et al. 2023; Fox et al. 2023) leave some states short of the USDA’s expectation that eligible households receive benefits within thirty days of application. This, paired with the use of broad-based eligibility in many states, generated growing variation in inclusion (figure 6) even in a putatively federal program. Much of this variation occurred above the median: inclusion nearly doubled at the 90th percentile (from 0.76 to 1.4) between 1994 and 2020—a pattern over the full 1994–2020 era of upward divergence.
Other child-focused programs betrayed mixed patterns of growth and variation. In early childhood education (ECE), generosity at the median grew from $7,800 in 1994 to $10,500 in 2022, a change that encompasses a marked increase from 1994 to 2006, a corresponding decline through 2014, and another increase in the last decade. Inclusion grew modestly (from 0.14 to .24) with growth concentrated at the median and above, reflecting disparate growth in state pre-K programs (by 2022, inclusion ranged from 8 percent of three- to four-year-olds participating in state pre-K or Head Start programs at the 10th percentile to 40 percent at the 90th). Variation in generosity reflected a modest upward divergence. As with generosity (table 1), higher levels of state discretion (ECE policies other than Head Start are set almost entirely by states or school districts) yielded widening variation in program inclusion.
In this new world of conditional and narrowly targeted social policy, the provision of temporary supports was increasingly important. Unemployment insurance is strictly work-conditioned, both by its eligibility terms (a record of earnings in covered employment) and by benefit levels and job search requirements designed to hasten reemployment. Rulemaking in UI is premised on substantial state discretion, which leaves the level and duration of benefits, and the details of monetary and nonmonetary eligibility to state plans (O’Leary et al. 2020; Blaustein 1993). Trends in generosity and inclusion from 1994–2022 (figure 1) are not linear in either direction but instead reflect shifts in rates of unemployment and temporary federal extensions or supplements (Woodbury 2015; Whittaker and Isaacs 2013; Congdon and Vroman 2022). During the Great Recession, federal spending increased the median UI benefit from $5,600 to over $6,600 and pushed inclusion at the median from 44 to 57 percent; by 2020, UI generosity and inclusion were back at or near their 2007 levels (table 3).
Variation in UI generosity (the average weekly benefit times the average weekly duration of receipt) reflects both underlying differences in wages from state to state, and discretion in program rules—especially constraints on duration imposed by a number of states in the wake of the Great Recession. That variation widened over time as net increases at the median and above were accompanied by a small decrease at the 10th percentile (upward divergence). Changes in variation with respect to inclusion were similar, with the largest losses at the 10th percentile (downward divergence). Such discretion—and its outcomes—is regionally patterned (figure 7), generated by both lower base wages in southern states and more restrictive terms of receipt. Such discretion is important during economic expansions but is muted during recessions when federal programs (statutory and temporary) increasingly step in to relieve state programs: federal spending met 20 percent of total UI costs during the 2001 recession, 56 percent during the Great Recession, and over 78 percent during the COVID pandemic (O’Leary 2013; Spadafora 2023).
Unemployment Insurance, Inclusion and Generosity, 1994 and 2022
Source: Authors’ calculation from the SSNP data.
Note: The same shapes are used to represent US regions across both years. Shapes with dark outlines correspond to 1994, and shapes without outlines correspond to 2022.
We identify three main shifts in social provision since the early 1990s: first, that all forms of assistance are increasingly conditioned (especially by labor force participation); second, that the boundaries of inclusion are increasingly narrow (especially targeted at children); and third, that benefits are inadequate (in part because they have been confined to temporary programs sustained only during exceptional moments of economic crisis). We observe evidence of these shifts in the decline in cash assistance generosity and inclusion following PRWORA in 1996, growth in programs such as early education and child health insurance, and the fleeting expansions of UI and food assistance during the Great Recession and the COVID pandemic. Further, we show that state discretion in rulemaking, financing, and administration contributes to wide variation in social provision, and that the level and form of state discretion generate often stark inequities in social provision across both states and programs.
States use discretion, in some cases, to make similar choices, resulting in a reduction in cross-state variation (convergence); in others, they make dissimilar choices, resulting in an increase in variation (divergence). Mapping this onto the trends in the median levels of generosity and inclusion, we find that, across programs, some states were more generous or inclusive over time (upward trending) and others were less so (downward trending). This yields, as summarized in figure 8, program-specific patterns of upward divergence, upward convergence, downward divergence, and downward convergence. Most remarkably, we see no single dominant pattern. We see—across programs and across both inclusion and generosity—both upward and downward convergence, and upward and downward divergence. This demonstrates that no single dynamic drives states to change their level of commitment to social provision (race to the bottom, rise to the rooftops) or to become more similar (laboratories of democracy) or more distinct. Regional patterns reflect some historical legacies but also exhibit considerable variation across programs, suggesting a complex pattern of state choices shaped by levels and mechanisms of discretion, by target populations, by state demographics and economies, and the like.
Changes over Time (1994–2022) in Generosity and Inclusion, Median, and Variation
Source: Authors’ calculation from the SSNP data.
Note: Values for state tax generosity are not included due to negative values. CA=Cash assistance, CC=Childcare, CHI=Child health insurance, CS=Child support, ECE=Preschool and early childhood education, SNAP=Food assistance, SSI=Supplemental Security Income. TAX=State income tax, UI=Unemployment insurance.
In this new configuration of social policy, wide disparities in program generosity and inclusion are not variations on national policy; these variations are the policy—fifty worlds of welfare in the place of any pretense of national standards or commitments. National averages or state medians mean little in a context in which state policy choices mark sometimes divergent, sometimes convergent, and sometimes idiosyncratic responses to insecurity or need. Across the recent history of such policies, generosity and inclusion plummeted in some programs, remained stable in others, and improved in a few. In some, generosity and inclusion moved in tandem; in others, the reach of the program and the adequacy of its benefits diverged or converged. As a result, low-income families relied on different combinations of support or suffered the disappearance of any meaningful support at all.
DISCUSSION
We understand the substantive shifts described earlier not simply as a reconfiguration of the targets and tools of social assistance, but as an approach to poverty governance that compromises an already shaky architecture of social citizenship. In the US case, and increasingly after 1996, the aspirational norms of robust social citizenship—universal boundaries of eligibility, generous or adequate benefits, and unconditional terms of assistance—have faltered. While benefits for some working parents have increased, the jurisdictional and categorical tangle of social provision invites and widens inequality in social protection across jurisdictions and populations, and calls into question the very compatibility of federated politics and social citizenship (Weaver 2019; Banting 2006).
Across this era, the reach of social policies and the experience of recipients were stunted by increasingly burdensome and conditional terms of receipt—especially the expectation of labor market participation as a condition of assistance or as a source of eligibility. Such conditions corroded the autonomy and security of both those who pursued them and those who were excluded by them. They marked a stark retreat from the idea that social policy might reduce or eliminate the necessity of commodifying one’s labor in order to maintain an adequate standard of living (Dukelow 2021; Kessler-Harris 2001). Prior to PRWORA, the choice to work for wages reflected a complex and often difficult calculation of material needs and family responsibilities (Edin and Lein 1997); after 1996, such choices were displaced by mandates and sanctions. Rooted in the welfare-to-work logic of PRWORA, such conditions have been extended to other programs (such as SNAP and Medicaid) on the assumption that they might discipline recipients or dissuade applicants (Wething 2025; Bauer et al. 2018). More broadly, such paternal terms of receipt enmesh recipients in coercive and punitive relationships with state authority as a baseline condition for assistance (Haney 2018; Fong and McCarthy 2026, this issue). Our inclusion and generosity measures, on this score, trace the boundaries and benefits of social policy but cannot fully capture the qualitative terms of receipt, including the often-punitive lived experience of social policies (Cherlin et al. 2002; Amerikaner et al. 2025). Such terms evoked a narrow and contractual form of citizenship in which social or civic standing retreated, as Margaret Somers (2008, 4) has argued, from that “based on shared fate among equals to that of conditional privilege.” The terms of social assistance—which demand demonstrations of worth, or administrative rituals of humiliation—becomes more important than the assistance itself.
Such terms, alongside a finer-grained categorization of eligibility across programs, have contributed to a dramatic narrowing of boundaries of social provision (Bruch et al. 2023). In this respect, the relative generosity and inclusiveness of child-focused programs also means that others are shut out as the targeting of assistance narrows from low-income families to working families, from parents to children. “Each categorization,” as Alice Kessler-Harris (2001, 65) reminds us, “open[ing] or clos[ing] a door to the status, social rights, and economic security that measured progress toward economic citizenship for someone.” Child-focused provision undermines universalism not only by narrowing the ambit of assistance, but by often cleaving the interests of children from those of their parents—curtailing the efforts of more holistic two-generation or whole-family strategies to break out of categorical silos (Sommer et al. 2018). Despite the child-focus, some groups of children, and in particular children with immigrant backgrounds, face both legal or formal exclusions from some programs (Bernstein et al. 2022; Acevedo-Garcia et al. 2021), and recent research also documents avoidance or hesitation to take up programs due to immigration concerns (Gonzalez et al. 2025). Even the security of children is subject to widely varying jurisdictional choices or circumstances with regard to childcare accessibility and quality, the expansion of public pre-K, and the administration of eligibility for programs like SNAP and SSI.
As the terms of receipt harden and the categorical reach of social programs narrows, the simple adequacy of social provision erodes even further. This reflects both the principle of less eligibility that accompanies programs calibrated to labor market expectations and labor market participation (Bonnet 2019; Piven and Cloward 1993) and the outright exclusion of those who cannot clear the conditional terms or administrative hurdles to receipt (Herd and Moynihan 2025). State and local policy discretion exposes recipients to the disparate demands of state and local labor markets. The resulting inequalities are evident in both choices made by individual states, and in their regional patterns. Policy devolution multiplies opportunities for inequities in policy design, administration, and outcomes (Kelly and Lobao 2021; Soss et al. 2011; Michener 2019; Skandalis et al. 2022) and has yielded a renewed racial patterning in both state policy and local administration (Soss et al. 2008, 2011). States with larger non-White caseloads or populations, or exhibiting deep racial inequality, are reliably less equitable, inclusive, and generous in their social policies (Hero 2003; Lieberman and Lapinski 2001; Pierson 2019; Hardy et al. 2019).
Onerous terms of receipt, narrowing categorical boundaries, and the fleeting and inadequate scope of provision—all compounded by jurisdictional inequities—shape not just social provision and social citizenship, but the broader parameters and experience of civil and political citizenship. Unequal protection, of course, is not lost on its recipients, whose economic security and social citizenship rest on both the generosity and reliability of income and social supports (Gonalons-Pons et al. 2026, this issue). More broadly, the structure and substance of social policies shape the experience of those targeted (or neglected) by state authority, and they shape their standing (or self-perception) as citizens or valued members of the polity commanding autonomy, respect, recognition, and protection (Mettler and Soss 2004; Soss and Weaver 2017; Fraser 1995). While policies structured on relatively universal and unconditional terms might generate civic inclusion and incorporation (Bidadanure 2019), those relying on conditional, paternal, and punitive terms have the opposite effect (Somers 2008; Bruch et al. 2010; Michener 2018).
While the drift of social policy since the 1990s has undermined the promise and performance of social citizenship, the outlook is not entirely bleak. The targeted universalism and autonomy promised by local guaranteed basic income pilots such as the Compton Pledge (Constantino et al. 2026, this issue) or large-scale experiments such as Baby’s First Year (Flanagan and Halpern-Meekin 2026) provide a welcome contrast to the increasingly conditional and categorical logic of our core social programs. In a similar vein, state and local innovations in a wide range of policies—including minimum wage, paid leave, and fair scheduling—offer fragmented promises of more robust social citizenship. Such innovations present challenges of scope and scale for any jurisdiction, and they are magnified where states have preempted the ability of municipalities and counties to even entertain such options (Bogle 2024; Briffault 2018). Finally, our assessment of social provision since the 1990s underscores the prospect of more generous and inclusive social policies. The growing generosity of programs like SNAP or ECE, and the marked expansion of inclusion in programs like SNAP and child health insurance in some states, suggest that our social policies are not governed by iron laws of decentralized austerity or declension. Such programmatic and jurisdictional bright spots illuminate our ability and capacity to do better.
FOOTNOTES
↵1. State discretion might also be shaped by policies or conditions that lie outside the policies in question. State and local policy choices on childcare, for example, might be shaped by the high costs of provision in some jurisdictions or a dearth of providers in others. The long history of workfare is shaped not only by policy decisions privileging labor market participation, but also by the uneven ability—across time and space—of labor markets to provide commensurate opportunities. Levels of child support, in turn, are determined less by state policies than by state and local judicial conventions regarding levels of noncustodial support (Venohr 2013; Pirog and Ziol-Guest 2006).
↵2. The SSNP dataset was created by Marcia Meyers, Sarah Bruch, and Janet Gornick from publicly accessible state and federal administrative records, and original population estimates calculated using the Annual Social and Economic Supplement (ASEC) of the Current Population Survey. The SSNP does not include programs for low-income families that do not include any state discretion, such as the federal EITC or CTC. The SSNP is not exhaustive of all programs for low-income families with decentralized designs; for example, it does not include WIC, school-based assistance, or housing assistance (public housing and subsidized voucher-based programs administered by state and local housing authorities). Earlier work (Bruch et al. 2018) based on the SSNP included a tenth program, targeted work supports. Due to concerns about the accuracy of state reporting on participation and expenditures on work supports, we exclude this program here.
↵3. In cases where there is a missing value for an observation (a state) or year, values are imputed using neighbor averages (that is, the average of year before and after the missing value). As with most administratively reported data, there is quite a bit of variability in the data obtained from many of the sources used in the construction of these policy indicators. To help reduce this type of measurement variability, the indicator values are top- and bottom-coded at two standard deviations from the mean for that year and are double-smoothed by first using three-year moving averages in the construction of the numerators and denominators and by smoothing the final indicator using three-year moving averages. For in-kind or service benefits, we use item-specific price indices: for food assistance, the US city average CPI for food at home; for health insurance, medical care services; for childcare and early childhood education, day care and preschool.
↵4. The potentially needy population denominators differ from estimates of the potentially eligible population, which incorporate additional program- and state-specific eligibility criteria (see the Urban Institute’s TRIM3, for example). We have chosen to calculate the potentially needy population defined by broad categorical criteria of programs to capture the depth of program receipt in the economically needy population. This approach allows for comparability over time within programs; our measure of the potentially needy, over time, does not reflect changes in program eligibility rules.
↵5. To test for significant changes over time, we use a t-test of means to compare the 1994 and 2022 values. The 50-state median differs from a national average, in that each state (regardless of population) is given equal weight.
↵6. While there is no standard metric for determining significant differences in the COV or Gini, we use a bootstrap method that generates a sample of COVs or Ginis for each yearly comparison and estimates the probability that the observed difference is random. In the bootstrapping process, we resample pairs of observations by state rather than resampling based on year values. This leads to much lower variation in the bootstrap estimates because state values are highly correlated over years. To determine when a change in COV or Gini is statistically significant, we test the differences across two years using a cut-off of p < .05 (see Kenworthy 1999 for a similar approach).
↵7. The designation of low-income in the findings follows rough programmatic thresholds: pretax and transfer (market) income below 100 percent of the federal poverty level (FPL) for cash assistance, childcare, and taxes; 130 percent for food assistance; 200 percent for SSI; and 300 percent for child health insurance (see table 2 for more specific measurement details). These differences are important for our measure of inclusion, as low (or high) rates of inclusion carry different implications depending on the FPL thresholds.
↵8. In some cases, the change was much starker on one measure. In Louisiana and Georgia, for example, inclusion fell dramatically (from 0.40 to 0.03 in Louisiana, from 0.70 to 0.03 in Georgia) between 1994 and 2022, but because both states legislated benefit increases in recent years, generosity actually increased (from $3,700 to $5,100 in Louisiana, from $5,700 to $8,500 in Georgia). At the other extreme, Oregon saw little change in inclusion (0.46 in 1994, 0.49 in 2022) but slashed its annualized benefit by almost three-quarters (from $8,900 to $2,500).
↵9. An estimated inclusion measure over 100 percent results from states having eligibility criteria and enrolling children in Medicaid and/or CHIP whose family income is above 300 percent of the FPL (see table 2 for more measurement details).
- © 2026 Russell Sage Foundation. Bruch, Sarah K., Arun Chaudhary, Colin Gordon, and KaLeigh K. White. 2026. “Fifty Worlds of Welfare: State Discretion and Social Citizenship Since 1994.” RSF: The Russell Sage Foundation Journal of the Social Sciences 12(1): 34–66. https://doi.org/10.7758/RSF.2026.12.1.02. Direct correspondence to: Sarah K. Bruch, at skbruch{at}udel.edu, 111 Academy Street, Newark, DE, 19716, United States. Arun Chaudhary, at achaudh{at}udel.edu, 111 Academy Street, Newark, DE, 19716, United States. Colin Gordon, at colin-gordon{at}uiowa.edu, 280 Schaeffer Hall, Iowa City, IA, 52242, United States. KaLeigh K. White, at kaleighkwhite{at}gmail.com, 1180 Observatory Drive, 3412 William H. Sewell Social Sciences Building, Madison, WI 53706-1320, United States.
Open Access Policy: RSF: The Russell Sage Foundation Journal of the Social Sciences is an open access journal. This article is published under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
REFERENCES
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.