Responding Holistically: Does Total U.S. State Safety Net Support Generosity Prevent Child Maltreatment?

Abstract

Background:

Prior studies show that policies strengthening families’ economic well-being may help reduce child maltreatment and child protective services (CPS) involvement. However, limited research has examined multiple programs holistically or accounted for substantial cross-state differences in both policy generosity and inclusivity.

Objective:

This study examines the association between total state safety net support and CPS investigation rates at the (state-year) population level, by considering nine major safety net policies that vary in generosity and inclusion across states and over time.

Participants and setting:

We draw administrative data from the 2004-2019 National Child Abuse and Neglect Data System child file. We link this data to unique state-year data that captures both the inclusivity and generosity of each safety net program.

Methods:

We construct a “total expected state safety net support” measure that captures the average expected benefit package for a potentially needy family in each state and year. We estimate two-way state and year fixed-effects models.

Results:

We find that a $1000 annual increase in total expected state safety net support is associated with a 4 percent decline in state-year CPS investigations among children of all ages and those under age 5, with similar associations for neglect and physical abuse investigations. Estimates were comparable across racial-ethnic groups, although estimated magnitudes were slightly larger for Black non-Hispanic children under age five.

Conclusions:

Our findings underscore the importance of measuring state safety net support holistically when examining its impact on CPS involvement, including accounting for both the generosity and inclusivity of multiple programs. Implications for policy are discussed.

1. Introduction

In 2023, 3.7 million children were the subject of a child protective services (CPS) investigation for alleged child abuse or neglect, nearly 15% of whom were substantiated as victims of child maltreatment (US Department of Health & Human Services, Administration for Children and Families, 2025). Cumulative lifetime rates are even higher; before turning 18, more than 1 in 3 U.S. children are subject to a CPS investigation (Kim et al., 2017). Identifying policy strategies for reducing risk of child maltreatment remains a critical social and economic concern. Research demonstrates that maltreatment increases risk for mental health challenges, substance abuse, criminal and delinquent behavior, and health and economic well-being (Currie & Tekin, 2012; Currie & Widom, 2010; Herrenkohl et al., 2013; Mersky & Topitzes, 2010). The annual cost of maltreatment to society is non-trivial, ranging from $592 billion for substantiated CPS investigations, to $2.96 trillion for all investigated CPS cases (Klika et al., 2020). Further, child maltreatment is a key factor that reinforces social inequality in the U.S., as lower income and racial-ethnic minority children are at greater risk of CPS involvement (Edwards et al., 2021; Imran et al., 2019).

We examine whether greater total safety net support in U.S. states is associated with lower CPS investigation rates, overall and by race and ethnicity, for all children and for children under age five. Prior research suggests that poverty and material hardship contribute to an elevated risk of both child maltreatment and CPS involvement (Austin et al., 2023; Beimers & Coulton, 2011; Berger et al., 2017; Brown et al., 2019; Bullinger et al., 2023; Cancian et al., 2013; Chanda et al., 2025; Harden & Berger, 2025; Hong et al., 2026; Kovski et al., 2022; McGinty et al., 2022; Pac et al., 2023; Puls et al., 2021, 2024; Rittenhouse, 2025; Rochford et al., 2024; Wildeman & Fallesen, 2017; Xu et al., 2021). Thus, it stands to reason that policies that support families’ economic wellbeing and financial security may help prevent CPS involvement. While prior research provides powerful insights, it has largely focused on a single program, such as the Earned Income Tax Credit (EITC), Supplemental Nutrition Assistance Program (SNAP), child support enforcement (IV-D), Temporary Assistance for Needy Families (TANF), Medicaid, and early care and education (ECE) programs (Berger et al., 2017; Brown et al., 2019; Cancian et al., 2013; Ginther & Johnson-Motoyama, 2022; Hong et al., 2026; Johnson-Motoyama et al., 2022; Kovski et al., 2022), or on only a limited subset of U.S. safety net programs (Klevens et al., 2015; Puls et al., 2021, 2024).

Our study extends prior research by taking a more holistic view of the U.S. state social safety net landscape. Specifically, we construct a state-level measure of average expected safety net support that captures both program generosity (program expenditures or benefit amount per recipient) and inclusion (likelihood of receiving a benefit for a potentially needy family) to compute the average amount of social welfare program support that a potentially needy family can expect to receive in a given state and year. Our measure includes nine safety net policies and programs, many, though not all, of which are means-tested, and for which U.S. states have substantial discretion in setting eligibility rules, benefit levels, financing, or administration. These include cash and near-cash programs—TANF, Supplemental Security Income (SSI), Unemployment Insurance (UI), child support, state income taxes for low-income families, and SNAP—as well as in-kind programs, such as childcare subsidies, early education (state pre-kindergarten (pre-k) and Head Start), and public health insurance (Medicaid and State Children’s Health Insurance Program (SCHIP)).

Focusing on total expected safety net support has two major advantages. First, accounting for the full range of support allows us to estimate the combined association of multiple public investments with CPS involvement. In contrast to many Western European countries which provide more universal, broad-based support in the form of social insurance, the U.S. safety net is a patchwork system that delivers limited, targeted support to low-income families through various cash and in-kind benefit programs (Bruch et al., 2018; Esping-Andersen, 1990). As a result, low-income families in the U.S. rely on a range of programs – and often receive them simultaneously – to make ends meet (Halpern-Meekin et al., 2015; Macartney & Ghertner, 2023). Thus, it is crucial to consider the full package of potential support to families in addition to individual programs. Second, our approach recognizes that the U.S. social safety net is a complex, decentralized, and fragmented system that varies substantially both between states and within-states over time in both inclusivity (i.e., program access, or the share of needy households receiving benefits) and generosity (i.e., benefit amount). Under U.S. federalism, individual states exercise substantial discretion in eligibility rules, financing, administration, and benefit levels (Bruch et al., 2018). Resulting inequalities in inclusion and generosity highlight the importance of considering both inclusivity and generosity. Program inclusivity is especially relevant to consider when assessing potential preventative impacts for populations that face greater barriers to the social safety net, such as racial-ethnic minorities (Bitler et al., 2023; Parolin et al., 2023).

To the best of our knowledge, very little research on CPS involvement has accounted for differences in both program generosity and inclusion across the range of programs that constitute the U.S. safety net. Previous studies have operationalized generosity as state-level annual spending on TANF, Medicaid, SCHIP, EITC, Head Start, Child Care and Development Fund (CCDF), and pre-k per person living below the federal poverty line (Puls et al., 2021, 2024). This work finds an inverse relation between per-capita spending on these programs and CPS investigation rates that holds only for White children and not for Black and Hispanic children. Although this approach provides vital insight on the importance of considering multiple programs, it is limited by the omission of several important programs for low-income families, such as child support, SNAP, SSI, and UI. Further, while measures of spending per person below the poverty line are relevant indicators of program generosity, they do not fully capture the benefit amounts potentially needy populations are likely to receive – for each individual program – as a function of both benefit generosity and program inclusion.

Whether, and in what direction, U.S. state safety net support is associated with CPS involvement is theoretically ambiguous. There are likely three mechanisms linking safety net support to CPS involvement. First, given plausibly causal evidence on the impact of income on child maltreatment and CPS involvement (Bullinger et al., 2023; Cancian et al., 2013; Kovski et al., 2022; Rittenhouse, 2025; Wildeman & Fallesen, 2017), greater state safety net support may reduce CPS investigations by increasing disposable income by reducing out-of-pocket expenses or increasing total income directly, via in-kind or cash assistance, respectively. Additionally, safety net support may raise income through increased parental employment and work hours (through, for example, more generous and inclusive state ECE and EITC programs) (Humphries et al., 2025; Jackson et al., 2025; Marifian, 2021; Timilsina, 2025). This “income pathway” may result in reduced parental stress, anxiety, and depression, which may, in turn, promote positive parenting behaviors and increase parents’ abilities to meet children’s needs (Duncan et al., 2014), thereby reducing CPS involvement risk.

Second, some safety net supports may mechanically lower the risk for CPS involvement. For example, ECE programs may limit time spent unsupervised or children’s exposure to potentially abusive or neglectful caregivers (Hong et al., 2026; Pac, 2021). Medicaid and SCHIP – which enhance access to health care – may improve parents’ mental health (McMorrow et al., 2017; Winkelman & Chang, 2018), and in turn, help parents provide safer and more consistent care for their children (Brown et al., 2019). In addition, children’s health may improve with greater access to health care and, as a result, parents may find caregiving easier and, thus, be less likely to engage in maltreatment. Moreover, health care providers may make referrals to other public benefit programs that may help reduce maltreatment and CPS involvement rates (National Academies of Sciences, Engineering, and Medicine, 2024).

Finally, while the income and mechanical mechanisms suggest an inverse relation between safety net support and CPS involvement, greater support in (some) safety net programs may heighten families’ exposure to mandated reporters—including health care providers, teachers, social workers, and childcare providers—which could lead to greater CPS reporting for alleged maltreatment (National Academies of Sciences, Engineering, and Medicine, 2024). Though quantitative research largely suggests that surveillance is not a major driver of disparities in CPS involvement (Berger & Harden, 2025), qualitative interviews of CPS-involved families suggest that it cannot be ruled out (Fong, 2019, 2020; Merritt, 2021).

2. Methods

2.1. Data and sample

Data on CPS investigations are drawn from the 2004–2019 National Child Abuse and Neglect Data System (NCANDS) child file. For children who were investigated multiple times, we use their latest investigation in a given year, treating it as their single investigation, such that our investigation rate is scaled on a per-child base. We aggregate individual CPS investigation records to the state-year level for all children aged 0–17 and separately for children aged 0–4. For each age group, we additionally aggregate individual records to the state-year-racial group level for White Non-Hispanic (NH), Black NH, and Hispanic (of any race) children. We leave out “other” NH populations, as they comprise a range of different racial-ethnic groups, such as Asian, American Indian or Alaska native, Native Hawaiian or other Pacific Islander children, who have differing risks for child maltreatment and CPS involvement. Further, these heterogenous groups have small populations in some state-years, which creates measurement challenges to constructing state-year CPS investigation rates for each group. Race and ethnicity categories were reported by parents or children.

The NCANDS child file reports data for most states with a few exceptions: North Dakota (missing 2004-2008) and Oregon (missing 2004-2010). Hence, in our full sample analyses and subgroup analyses by race and ethnicity, those states (in their relevant years), as well as the District of Columbia (in all years), are excluded. We exclude the District of Columbia because our key policy variables are available only for the 50 U.S. states. In addition, reported race and ethnicity of the child is missing in Pennsylvania in 2004-2013. We therefore exclude Pennsylvania in all years from our subgroup analyses by race and ethnicity. We do not consider post-COVID years due to fluctuations in CPS reporting during the COVID-19 pandemic. Our final analytic sample includes 788 state-year observations and 2256 state-year-race observations (i.e., 752 state-year observations in each group).

We link NCANDS state-year data to the State Safety Net Policy Dataset (SSNP) (Bruch, Meyers, et al., 2023; see also (Bruch et al., 2018, 2022; Bruch, Naald, et al., 2023; Rhodes et al., 2026) for other applications of the SSNP data). The SSNP provides yearly state-level information on safety net programs for all 50 states, beginning in 1994, is constructed primarily from publicly available state and federal administrative records, and covers programs with some degree of state discretion¹ that serve as primary sources of support for most low-income or unemployed working-age adults and their dependents. SSNP has important advantages over other policy data sources. First, its measures uniquely capture both generosity and inclusion. Second, its generosity and inclusion indicators are comparable across multiple programs and over time. Generosity is calculated by dividing program expenditure (total expenditure or expenditure on benefits, if available) by the number of recipients in a state. Inclusion is measured as the proportion of potentially needy population in a state that is enrolled in a given program, which reflects eligibility restrictions and administrative burdens that make it more or less difficult to access benefits.² For means-tested programs, the estimate of the potentially needy is the number of individuals or families with (pre-tax and transfer) income below the federal poverty line or below another threshold determined by the program’s income eligibility criteria. For non-means-tested programs, the definition of the potentially needy population depends on the program – for example, we use the number of unemployed individuals and the number of single parent families with children as the denominators for UI and child support, respectively. Additional information on the data sources and specific definitions is provided in Table 1.

Table 1.

Information on SSNP measure descriptions and data sources

Program	Dimension	Measure Construction
Temporary Assistance for Needy Families	Generosity	State and federal dollars spent on cash benefits in TANF program divided by the monthly average number of recipient families.	ACF TANF Financial Data
	Inclusion	The monthly average number of families receiving TANF, divided by the number of pre-tax and transfer poor families with children (at 100% Federal Poverty line (FPL)).	OFA Caseload Data, Population denominators from CPS ASEC (three-year moving averages)
Child Support	Generosity	Child support distributions per child support case in which a child support collection was made on an obligation.	OCSE Annual Report to Congress
	Inclusion	Number of child support cases for which a collection was made on an obligation divided by the number of single parent families with children.	OCSE Annual Report to Congress, Population denominators from CPS ASEC (three-year moving averages)
Supplemental Nutrition Assistance Program	Generosity	Expenditures on benefits divided by the number of participating households.	USDA Food and Nutrition Service Food Stamp Program Data
	Inclusion	Number of households with children participating divided by the number of pre-tax and transfer poor families with children (130% FPL).	USDA Food and Nutrition Service Characteristics of Food Stamp Households Annual Reports, Population denominators from CPS ASEC (three-year moving averages)
Unemployment Insurance	Generosity	Average weekly benefit received multiplied by the average number of weeks of receipt.	Department of Labor Employment and Training Administration Unemployment Insurance Data Summaries
	Inclusion	Number of recipients in all programs divided by the total number of unemployed.	Department of Labor Employment and Training Administration Unemployment Insurance Data Summaries, Population denominators from CPS ASEC (three-year moving averages)
Supplemental Security Income	Generosity	Average yearly child disability benefit received (includes federally administered state supplementation payments).	Social Security Administration SSI Annual Statistical Reports
	Inclusion	Number of children < 18 receiving SSI divided by the number of pre-tax and transfer poor children < 18 (200% FPL)	Social Security Administration SSI Annual Statistical Reports, Population denominators from CPS ASEC (three-year moving averages)
State Income Tax for Low-Income Families	Generosity	State income tax that a single-parent family of three pays when their income is at the poverty line (positive values indicate a tax credit, negative values indicate a liability).	NBER TAXSIM tax calculation tool
	Inclusion	Proportion of poor single parent families of 3 (100% FPL) under state income tax threshold for single parent family of 3.	NBER TAXSIM tax calculation tool, Population denominators from CPS ASEC (three-year moving averages)
Child Health Insurance	Generosity	Federal and state expenditures on Medicaid child beneficiaries and SCHIP enrollees divided by the number of Medicaid child beneficiaries and SCHIP enrolled children.	MACPAC MACStats Reports
	Inclusion	Medicaid beneficiaries and SCHIP enrolled children divided by the under 18 pre-tax and transfer poor population (300% FPL).	CMS CHIP Annual Enrollment Report, Population denominators from CPS ASEC (three-year moving averages)
Preschool and Early Education	Generosity	Federal and state expenditures on Head Start and state pre-K divided by the number of children enrolled in Head Start and state pre-K.	National Institute for Early Education Research State of Preschool 2002-2021
	Inclusion	Children enrolled in state pre-K and Head Start divided by the number of children 3-4 years old.	National Institute for Early Education Research, Population denominators from CPS ASEC (three-year moving averages)
Childcare Subsidies	Generosity	Total spending (CCDF and TANF) on childcare per child served by TANF and CCDF.	Dept. of HHS ACF's CCDF Data, Dept. of HHS ACF TANF Financial Data
	Inclusion	Number of children served by TANF and CCDF divided by the number of pre-tax and transfer poor children under 13 (100% FPL)	Dept. of HHS ACF's CCDF data, Dept. of HHS ACF TANF Financial Data, Population denominators from CPS ASEC (three-year moving averages)

Source: Bruch, Meyers, et al., 2023.

2.2. CPS involvement

We measure CPS involvement using the state-year or state-year-race specific CPS investigation rate, overall and separately for neglect and physical abuse allegations. The rate is computed as the number of investigations per 1,000 children (either under 18 or under 5), and we draw on population denominators from the 2004-2019 Surveillance, Epidemiology, and End Results (SEER) county-level bridged race population database. Given that investigation rates are as important an indicator of risk as substantiated cases (Drake et al., 2003), and that substantiation is partially influenced by agency-related factors such as local accessibility to prevention services (Font & Maguire-Jack, 2015), we follow previous studies and use investigation rates as our primary outcome. However, we also present findings using substantiated rates in supplemental analyses. Figure A.1 illustrates state-level trends in CPS investigation rates for all children (panel A) and children under age 5 (panel B).

2.3. State safety net support

Following prior research (Rhodes et al., 2026), we construct a measure of total expected state safety net support using SSNP data. It is calculated by first multiplying the generosity value by the inclusion value for each program, thereby discounting generosity by inclusivity. Multiplying the two gives us the average program generosity one would expect for a potentially needy family in a given state and year. There is one exception in our formula: when the state income tax generosity variable is negative, indicating a tax liability, we multiply it by (1-inclusion). This adjustment is made to be consistent with higher inclusion rates leading to lesser discounts in generosity. We then sum these multiplied values across all nine programs and adjust for inflation (in 2021 dollars), using the Consumer Price Index. This produces a measure that reflects the overall supportiveness of the safety net package in each state and year.

Figure 1 illustrates that there is substantial variation in total expected safety net support across states and over years (see panel A for the overall measure and panel B for individual policies), with 12% of the total variation reflecting variation within states over time, and 88% reflecting variation between states. This within-state temporal variation suggests there is enough change in state safety net support to estimate models with two-way state and year fixed effects.

Figure 1.

Trends in safety net policy variables

Note. Authors’ analysis of 2004-2019 SSNP data. TANF = Temporary Assistance for Needy Families; SSI = Supplemental Security Income; SNAP = Supplemental Nutrition Assistance Program; Pre-K = Pre-Kindergarten; SCHIP = State Children’s Health Insurance Program.

2.4. Covariates

To account for potential confounders of associations between total state safety net support and CPS involvement, we include the following time-varying state-level demographic characteristics and economic conditions that have been shown to be, or are potentially related to, both CPS involvement and state safety net support: the proportion of Black NH and Hispanic children, whether the governor is a Democrat, the unemployment rate, the teen birth rate (Drake et al., 2023; Puls et al., 2021), and the proportion of single-parent households (Drake et al., 2023; Paxson & Waldfogel, 2002). Table A.1 shows summary statistics for our CPS involvement measures, total safety net support measures, and state-level covariates, as well as their data sources.

2.5. Statistical methods

We estimate ordinary least squares (OLS) two-way fixed-effects (TWFE) regressions that leverage within-state variation in total state safety net support over time, controlling for the state-year level covariates described above. In the regressions, we include state fixed effects and year fixed effects, which adjust for observed and unobserved time-invariant state-level characteristics (e.g., permanent state differences in CPS policies, cost-of-living differences) and national year-specific factors (e.g., macro-economic shocks), respectively. These models are analyzed for all children (aged 0-17) and for children aged 0-4, as the latter have greater access to some policies, such as ECE programs, and face the highest risk for child maltreatment (US Department of Health & Human Services, Administration for Children and Families, 2025). We weight these models by the state population to prevent small states from disproportionately influencing the estimates. Given well-documented racial ethnic disparities in CPS involvement (Edwards et al., 2021) and access to the social safety net (Bitler et al., 2023; Parolin et al., 2023), we also conduct subgroup analyses by race and ethnicity and compare estimates across subgroups using Z-tests.

Finally, we conduct several sensitivity and supplementary analyses. As sensitivity tests, we consider two alternative specifications of state safety net support: (i) considering only benefit generosity, defined as the maximum benefit available in each program; and (ii) excluding one policy at a time to confirm our findings are not driven by any single policy. In the former, we construct a variable that is the sum of state maximum TANF benefits and state plus federal EITC benefits for a family of three with two children. Because we are unaware of any dataset that provides consistent measures of maximum benefits across all programs considered in this study, we focused on those two policies. This test helps ensure that variation in total safety net support is not driven by changes in the composition of program beneficiaries. As supplementary analyses, we estimate models in which we separate total safety net support into cash or near-cash and in-kind benefits to assess whether these program types differ in their associations with maltreatment. Additionally, we estimate models using substantiated child maltreatment rates, rather than investigation rates, as the outcome. Our analyses were conducted using Stata SE, version 18.5.

3. Results

Table 2 presents our main findings from TWFE models that estimate associations between total safety net support and CPS investigation rates. We present both the absolute and percent changes in CPS investigation rates; the percent change is calculated relative to the investigation rate mean in the study sample. As indicated in panel A, $1000 increase in total state safety net support in a given year is associated with 1.86 fewer CPS investigations per 1000 children of all ages (a 3.7 percent reduction from the sample mean), with similar estimated effect magnitudes for neglect (4.9 percent reduction) and physical abuse investigations (5.6 percent reduction). We find a similar magnitude of association for young children (see panel B): $1000 greater safety net support is associated with 2.5 fewer CPS investigations per 1000 children aged 0-4 (a 4.0 percent reduction). While physical abuse allegation magnitudes are larger (1.1 fewer investigations, i.e., 9.4 percent) than neglect allegations (1.9 fewer investigations, i.e., 4.5 percent), Z-tests indicate that the two estimates are not statistically distinguishable from one another. Table A.3 illustrates the full regression outputs.

Table 2.

Associations between total safety net support and CPS investigation rates, overall and for neglect and physical abuse

	Total investigation rates per 1000	Neglect investigation rates per 1000	Abuse investigation rates per 1000
Panel A. All children aged 0-17.
Total safety net supportiveness	−1.86* (0.92)	−1.53* (0.73)	−0.62** (0.22)
Percent effect	3.7	4.9	5.6
Outcome mean	50.9	31.0	11.1
Panel B. Children aged 0-4.
Total safety net supportiveness	−2.49* (1.05)	−1.91* (0.94)	−1.06*** (0.30)
Percent effect	4.0	4.5	9.4
Outcome mean	62.1	42.0	11.3
Observations	788	788	788

Note. Authors' analyses of 2004-2019 NCANDS data linked to policy data from the SSNP. Total safety net support is reported in $1000s; percent effect is the percent reduction relative to the CPS investigation mean. Standard errors are reported in parentheses. All models control for state fixed effects, year fixed effects, proportion of Black and Hispanic children, whether a governor is a Democrat, unemployment rate, teenage birth rate, and percentage of single-parent households. The full table of coefficients is presented in appendix Table A.3. Estimates are weighted by the state population of children (either 0-17 or 0-4) in each year.

^***p<0.001

^**p<0.01

^*p<0.05

⁺p<0.1.

We next turn to subgroup analyses by race and ethnicity. Table 3 shows that, in terms of effect size, total safety net support appears to play a stronger role in reducing CPS investigations among racial and ethnic minority children under age 5, particularly Black NH children, relative to White NH children. For instance, $1,000 greater annual total expected safety net support is significantly associated with 4.0 fewer CPS investigations (a 3.6 percent decrease), 2.9 fewer neglect investigations (a 3.8 percent decrease), and 1.8 fewer abuse investigations (a 9.2 percent decrease) per 1000 Black NH children under age 5. These estimated effect sizes exceed or are comparable to those of White NH children under age 5 (3.0, 3.8, and 8.2 percent, respectively), and the estimates for White NH children are either statistically nonsignificant or attain only marginal significance (p<.10). We found similar trends among all children with respect to statistical significance, but the estimated magnitudes were more similar across racial and ethnic groups compared to those found among young children. However, Z-test results indicate that the negative associations between total state safety net support and CPS investigations do not statistically differ between White NH, Black NH, and Hispanic children in either age group (see Table A.4, A.5 for full regression outputs).

Table 3.

Subgroup analyses by race and ethnicity: Associations between total safety net support and CPS investigation rates, overall and for neglect and physical abuse

Panel A. All children aged 0-17.	Total investigation rates per 1000			Neglect investigation rates per 1000			Abuse investigation rates per 1000
	White	Black	Hispanic	White	Black	Hispanic	White	Black	Hispanic
Total safety net supportiveness	−1.31 (0.79)	−2.54+ (1.47)	−0.92 (0.84)	−1.18+ (0.64)	−2.12* (1.04)	−0.94 (0.64)	−0.49* (0.21)	−0.87* (0.39)	−0.51* (0.22)
Percent effect	3.0	2.9	2.2	4.3	3.9	3.7	5.4	4.3	5.5
Outcome mean	44.4	90.0	42.0	27.9	55.0	25.4	9.2	20.5	9.3
Panel B. Children aged 0-4
Total safety net supportiveness	−1.61+ (0.93)	−4.03* (1.69)	−1.00 (1.02)	−1.42+ (0.79)	−2.91* (1.42)	−0.93 (0.84)	−0.80* (0.30)	−1.84** (0.57)	−0.78** (0.25)
Percent effect	3.0	3.6	2.2	3.8	3.8	3.0	8.2	9.2	9.2
Outcome mean	54.2	111.3	45.9	37.7	75.9	30.7	9.7	20.0	8.5
Observations	752	752	752	752	752	752	752	752	752

Note. Authors' analyses of 2004-2019 NCANDS data linked to policy data from the SSNP. Total safety net support is reported in $1000s; percent effect is the percent reduction relative to the CPS investigation mean. Standard errors are reported in parentheses. All models control for state fixed effects, year fixed effects, proportion of Black and Hispanic children, whether a governor is a Democrat, unemployment rate, teenage birth rate, and percentage of single-parent households. The full table of coefficients is presented in appendix Tables A.4 and A.5. Estimates are weighted by the state population of children (either 0-17 or 0-4) in each year.

^***p<0.001

^**p<0.01

^*p<0.05

⁺p<0.1

Finally, the results in Table 4 demonstrate that our findings are robust to considering a more narrowly defined measure of policy generosity, giving confidence that our main results are not driven by changes in the composition of program beneficiaries. They also demonstrate that excluding each policy in turn yields similar results, indicating that our findings are not driven by any single program but instead reflect the effects of a holistic package of safety net programs. Additionally, supplemental analyses suggest that, for children under age 5, cash or near-cash support, rather than in-kind support, is the primary driver of the association between greater total safety net support and lower CPS investigation rates (see Table A.2). However, in contrast to our results for investigations, we find that total safety net support is not significantly associated with substantiation rates (see Table A.3).

Table 4.

Associations of alternative state safety net measures with CPS investigation rates

	All Children Aged 0-17	Children Aged 0-4
Maximum TANF and EITC Benefits	−2.36** (0.78)	−2.60** (0.93)
Total Safety Net Support Without TANF	−2.22+ (1.22)	−2.91* (1.37)
Total Safety Net Support Without Child Support	−2.01+ (1.13)	−2.66* (1.31)
Total Safety Net Support Without SNAP	−1.92* (0.90)	−2.60* (1.06)
Total Safety Net Support Without SSI	−1.88* (0.93)	−2.51* (1.06)
Total Safety Net Support Without Head Start/State Pre-K	−1.63+ (0.86)	−2.33* (0.97)
Total Safety Net Support Without Childcare Subsidies	−2.49* (1.20)	−3.18* (1.35)
Total Safety Net Support Without Medicaid/SCHIP	−1.80+ (0.90)	−2.43* (1.05)
Total Safety Net Support Without State Income Taxes for Low-Income Families	−1.96* (0.94)	−2.62* (1.09)
Total Safety Net Support Without Unemployment Insurance	−1.78+ (0.93)	−2.35* (1.11)

Note. Authors' analyses of 2004-2019 NCANDS data linked to policy data from the SSNP. Total safety net support is reported in $1000s; percent effect is the percent reduction relative to the CPS investigation mean. Standard errors are reported in parentheses. All models control for state fixed effects, year fixed effects, proportion of Black and Hispanic children, whether a governor is a Democrat, unemployment rate, teenage birth rate, and percentage of single-parent households. Estimates are weighted by the state population of children (either 0-17 or 0-4) in each year. Each estimate is drawn from a separate regression and represents the key coefficient of interest of the relation between total state safety net support and the CPS investigation rates. N=788 in all models.

^***p<0.001

^**p<0.01

^*p<0.05

⁺p<0.1.

4. Discussion

This study contributes to the child maltreatment, social policy, and social work literatures by providing new evidence that overall U.S. state safety net support is associated with lower rates of CPS involvement. Given that child maltreatment and CPS involvement have detrimental consequences for children’s health and well-being in childhood and adulthood (Currie & Tekin, 2012; Currie & Widom, 2010; Herrenkohl et al., 2013; Mersky & Topitzes, 2010), their prevention may serve as an important mechanism by which state policy contexts improve children’s social and economic mobility and long-term health.

Using measures capturing a wide range of state safety net programs that serve as important sources of income for low-income families and accounting for state-year differences in both program generosity and inclusion, we find that $1000 greater safety net support is associated with a 4 percent reduction in CPS investigations, both among all children and children under age 5, with a statistically similar association for both child neglect and physical abuse allegations. Our results further suggest that, among children under age 5, cash or near-cash support is associated with a greater reduction in CPS involvement than in-kind support. To the extent that cash or near-cash benefits more directly alleviate economic hardship than in-kind benefits (Center on Budget and Policy Priorities, 2021; World Bank Group, 2024), the stronger role of cash or near-cash support is consistent with the income pathway serving as a key mechanism linking total safety net support to CPS involvement, although we could not conduct a formal mediation analysis to test this pathway.

Because we lack individual-level data on program receipt status and income, our findings reflect an intent-to-treat (ITT) estimate of the association between total safety net support and CPS involvement at the (state-year) population level, rather than an effect of receiving greater benefits among low-income families who are likely eligible. Although our effect sizes are somewhat smaller than those reported in prior studies, they remain within the reported range, and are sizable for population-level ITT estimates of associations of additional income with CPS involvement. Berger et al. (2017) and Rittenhouse (2025), for instance, found a 2-10 percent decrease in CPS involvement (e.g., the likelihood of being contacted by CPS in the first three years of a child’s life) per $1000 increase in income among low-income single-mother families, and all low-income families, respectively. Bullinger et al. (2023) found a $1000 increase in family income during the first few months of a child’s life to be associated with a 10 percent reduction in CPS involvement. An additional potential explanation for the generally lower magnitudes of our estimates is that our measure of total expected safety net support includes in-kind benefits. They may not provide the same flexibility as direct cash assistance for managing unexpected income shocks that may elevate the risk for maltreatment and CPS involvement. Consistent with this argument, we find that cash or near-cash safety support is more strongly linked to reduced CPS investigations among young children.

Our subgroup analyses indicate that total safety net support has statistically similar associations with CPS involvement across racial and ethnic groups. This finding differs from some prior research on the aggregate influence of multiple safety net programs, which found that the association of state safety net spending with CPS investigations is smaller for racial and ethnic minorities than for White NH populations (Puls et al., 2024). A separate body of research on the safety net finds a higher likelihood of benefit receipt among White NH low-income populations, driven in part by state eligibility and administrative policies, as well as broader policy climates that disadvantage program take-up for racial and ethnic minorities (Bitler et al., 2021; Brantley et al., 2020; Minoff et al., 2021; Ullrich et al., 2019). By incorporating variation in state-year level inclusivity for each program, our analyses may better account for differential probabilities of safety net receipt by race and ethnicity, given differences in states’ racial and ethnic composition. Hence, this may explain our finding of similar associations across racial and ethnic groups. However, our state-year level inclusivity measure enables us only to account for differences in aggregate-level program access across states and over time. It does not allow us to assess differential access to safety net programs by race and ethnicity within a state and year. Future research should use individual-level data to examine whether similar results emerge after accounting for differential program receipt by race and ethnicity and other demographic characteristics (e.g., income). If so, identifying the mechanisms underlying such patterns could inform policy to reduce racial and ethnic disparities in CPS involvement.

This study has additional limitations. First, our findings do not lend themselves to causal interpretation. It is possible that there remain confounding factors for which we are unable to account, such as labor market or economic shocks that are not fully captured by the state unemployment rate, or state-year specific differences in the infrastructure for and accessibility of various prevention services. Second, data limitations prevent us from directly testing the mechanisms that may underlie the relation between greater overall state safety net generosity package and lesser CPS involvement. Future research on causal mechanisms is needed. While it is challenging to empirically distinguish between the income and mechanical pathways discussed earlier, access to individual-level data on state of residence, CPS involvement, family income, total expected safety net support, total safety net support received, parental stress, and parenting behavior would help assess whether one of the pathways plays a dominant role or both are at play vis-a-vis the association between greater total state safety net support and lesser CPS involvement. Third, the aggregated (state-year) nature of our data limits the statistical power to precisely estimate associations between specific individual policies that constitute total safety net support and CPS involvement. This is also a crucial topic for future research, which could inform policymakers about which types of policies warrant greater investment.

5. Conclusion

Notwithstanding its limitations, by examining associations of overall expected safety net support with risk of child maltreatment (as approximated by CPS involvement), this study contributes to a more holistic understanding of the full range of potential costs and benefits of states’ investments in the social safety net, and offers timely implications for state policymaking debates. From a methodological standpoint, our findings highlight the importance of taking a holistic approach to measuring state safety net support, accounting for both the generosity and inclusivity of multiple programs. For policy, our findings suggest that expanding the generosity and inclusion of economic support programs—through state policy expansions that may be more politically feasible than federal expansions—may help reduce risk of CPS involvement. Given the substantial share of public benefit spending undertaken by states (Hembre et al., 2024; Urban Institute, n.d.), as well as the important implications of cross-state inequality in social welfare policies (e.g., Bruch et al., 2018), increased investment in state safety net policies may have meaningful preventative benefits.