Debt Strain and Child Protective Services Involvement

Abstract

Research has identified a likely causal relation of economic precarity with both child maltreatment and child protective service (CPS) involvement. Yet, little is known about the relation between credit use (debt) and CPS involvement despite credit becoming an increasingly normative aspect of attempting to manage economic precarity for low-income families. We link individual-level longitudinal data on credit use to administrative records on CPS involvement, to examine whether credit related economic precarity—or ‘debt strain’—is associated with increased CPS reports, allegations of child neglect and child abuse, and CPS removal of children for low-income mothers in Wisconsin from 2016–2021. Results from standard and mother-specific fixed-effects logistic regressions suggest that debt strain is associated with greater risk of CPS involvement, particularly for child neglect. Heterogeneity analyses reveal that these associations are concentrated among low-income and White mothers. We discuss implications for debt-regulation and CPS policy and programs.

Introduction

Child welfare system involvement is a relatively common experience for American children and is especially common among children of color and those from low-income families (Kim et al. 2017; Wildeman and Emanuel 2014; Wildeman et al. 2014). Recent estimates suggest that more than one-third of all children and more than one-half of Black children in the US become involved with the child welfare system by the time they turn 18 (Font and Maguire-Jack 2020; Kim et al. 2017). Lower economic resources and associated greater economic precarity in the family or community are strongly linked to both child maltreatment and child protective services (CPS) involvement (Berger and Waldfogel 2011; Fong 2019b; Font and Maguire-Jack 2020), and a growing body of research suggests that this relation is likely causal (Pac et al. 2023). Indicators of economic precarity, such as home foreclosures (Berger et al. 2015) and decreases in welfare benefit generosity (Wildeman and Fallesen 2017), have been linked to increased CPS involvement and foster care placement, whereas minimum wage increases (Raissian and Bullinger 2017), larger cash transfers (Bullinger, Packham, and Raissian 2023; Berger et al. 2017) and greater child support receipt (Cancian, Yang, and Slack 2013) have been shown to reduce CPS involvement. Moreover, racial and ethnic disparities in economic resources and precarity are highly salient drivers of the well-documented disparities in child maltreatment and CPS involvement by race and ethnicity (Barth et al. 2020; 2022; Kim and Drake 2018; Putnam-Hornstein et al. 2013; Thomas and Waldfogel 2022).

Access to and use of credit are now (increasingly) common aspects of attempting to manage economic precarity but also have the potential to worsen economic precarity in the United States, where the aggregate household debt balance has trended upward since the COVID-19 pandemic and associated recession, reaching $17.7 trillion in the first quarter of 2024 (Munsch 2024). State policies and credit markets have historically excluded low-income and minority populations from access to prime credit instruments (Dwyer 2018). However deregulation and expansion of subprime credit markets in recent decades and associated increases in subprime debt among low-income populations suggest that the management of debt has become a crucial component of inequality and economic precarity (Morduch and Schneider 2017; Wherry, Seefeldt, and Alvarez 2019; Lin and Neely 2020). Access to credit and the cost thereof are particularly salient for low- and moderate-income families, in terms of both navigating precarious financial circumstances and accessing goods and services, such as education and housing, that promote upward economic mobility. Yet, significant debt strain (a high level of debt relative to income or capacity for repayment) may also, itself, contribute to economic precarity. Despite evidence that economic resources are strongly linked to CPS involvement and that such associations reflect not only absolute income, but also economic precarity, more generally (Thomas and Waldfogel, 2022), research has yet to examine whether family debt strain may be associated with CPS involvement.

This study begins to fill this gap by using novel individual-level proprietary data that include ‘mainstream’ and ‘alternative (subprime) financial services’ credit report records for the entire population of Wisconsin from 2016 to 2021, linked to the Wisconsin Administrative Data Core (WADC). The WADC includes the full population of administrative records spanning CPS involvement, employment and earnings, social welfare program participation and benefit levels, corrections, and public education. Using these extensive data, we employ standard logistic regressions and (conditional) logistic regressions with mother-specific fixed effects to examine associations of debt strain – operationalized alternatively as high credit card utilization, use of high-cost alternative financial services, and delinquency on debt payments, as well as a global indicator of any of these three – with CPS investigations (overall and specifically for child neglect and physical abuse) and foster care placements.

Background

An extremely large proportion of children are impacted by the child welfare system over the course of their childhood. Each year, approximately 6 percent of U.S. children are reported to CPS, 5 percent are investigated by CPS, 1 percent are determined to be a ‘victim’ of child abuse and/or neglect, and 1 percent spend some time in foster care (US Department of Health and Human Services 2019; Administration on Children, Youth and Families, Children’s Bureau 2020). However, prevalence rates over the course of childhood are substantially higher with 37 percent of U.S. children being investigated by CPS (lifetime prevalence rates of being reported to CPS are not available), 13 percent being ‘substantiated’ as a maltreatment victim, and 6 percent spending some time in foster care (Kim et al. 2017; Wildeman and Emanuel 2014; Wildeman et al. 2014). The vast majority of CPS-involved children come from low-income and poor families, and children of color—particularly Black and American Indian children—are considerably overrepresented in CPS caseloads. Indeed, whereas 28 percent of White children are investigated, 11 percent substantiated, and 5 percent placed in foster care by CPS at some point during childhood, rates for Black children are 53 percent, 21 percent, and 11 percent, respectively, and are 23 percent, 15 percent, and 15 percent for American Indian children. They are 28 percent, 11 percent and 0.05 percent for Hispanic children, and 10 percent, 1 percent, and 0.02 percent for Asian and Pacific Islander children (Kim et al. 2017; Wildeman and Emanuel 2014; Wildeman et al. 2014).

Economic Precarity and Child Maltreatment

Research has consistently documented associations of low-income and poverty status, economic scarcity, and material hardship with child maltreatment and CPS involvement (Administration for Children and Families 2021; Putnam-Hornstein et al. 2013; Eckenrode et al. 2014; Pelton 2015). Cross-nationally, low-income status is the most consistent and strongest predictor of CPS involvement across the wealthy countries (Berger and Waldfogel 2011). Moreover, a growing body of research (especially in the US) indicates that the relation between income and CPS involvement is likely causal (Berger et al. 2017; Bullinger, Packham, and Raissian 2023; Cancian, Yang, and Slack 2013; Raissian and Bullinger 2017; Wildeman and Fallesen 2017). Slack and colleagues (2024) detail several pathways through which economic scarcity can impact CPS involvement. Economic scarcity can increase risk of CPS involvement directly, as limited economic resources can constrain parents’ abilities to secure adequate goods and services to meet children’s basic needs which, in turn, can lead to child maltreatment, particularly neglect. The onset or persistence of economic scarcity may also adversely influence caregiver wellbeing in terms of stress, distress, and other domains of mental and physical health, which can result in harsher and less responsive parenting and lesser time investments in children, and, thereby, increased risk of child abuse and neglect (Berger and Waldfogel 2011; Slack et al. 2024). Moreover, whereas caregiver characteristics are likely to be jointly associated with both a family’s economic resources and its risk of child maltreatment, in the context of economic scarcity, the likelihood that such characteristics manifest in child maltreatment may be exacerbated (Slack et al. 2024).

Economic precarity appears to be a particularly important element of low-income status in increasing CPS-involvement risk. Low-income families experience fluctuations in hours and employment, making it difficult to plan for children’s supervision and increasing stress in ways that affect parenting behaviors (Gennetian et al. 2015; Wolf et al. 2014; Bauer, East, and Howard 2025). State social welfare support can fluctuate as well, and administrative burdens and associated benefit-receipt delays, may further constitute a distinct source of economic precarity (Parolin 2023). In the context of economic scarcity, unexpected expenses such as rent increases, medical costs, or car repairs, can result in family crises (Fulford and Low 2024) that may further strain parental care. Economic scarcity thus both entails and is bound up with economic precarity, and the uncertainties therein may be particularly consequential for child protective services involvement.

Structural factors may also contribute to both economic precarity and CPS involvement. Disparities in economic resources by race and ethnicity, which stem from widespread historical and contemporary forms of structural and institutional racism and bias that pervade U.S. society, have been shown to largely ‘explain’ racial and ethnic disparities in CPS involvement and outcomes (Barth et al. 2020; 2022; Kim and Drake 2018; Putnam-Hornstein et al. 2013; Thomas and Waldfogel 2022) (although systemic biases within CPS may also be salient [Dettlaff et al. 2011; Font, Berger, and Slack 2012]). Structural factors external to the family, such as generosity of government benefits, may both impact family economic resources and also mediate the relation between economic precarity and child maltreatment. Moreover, low-income and minoritized families may face heightened surveillance from mandated reporters, which may also increase their risk of CPS involvement (Merritt 2020; Fong 2019a; 2020; 2017).

Debt Strain as an Indicator of Economic Precarity

Structural factors in US society have also increased the likely importance of credit and debt vis-à-vis economic precarity. The expansion of credit markets, especially with respect to high-cost credit, in the United States has made debt a significant element of both managing and potentially exacerbating economic precarity for low-income and poor families (Tach and Greene 2014). Limited access to and generosity of (particularly cash) social welfare transfers have made the market for unsecured and subprime (high cost) credit a significant source of private fallback; indeed, research has shown associations of less generous social welfare provision with greater presence of payday loan lenders and greater use of unsecured credit (Bea, Amorim, and Friedline 2023; Fitzpatrick and Coleman-Jensen 2014). Access to credit may (at least temporarily) relieve family economic precarity (and associated stress) by allowing for consumption smoothing and facilitating purchases of necessary goods and services. Yet, low-income and racialized populations have been systematically excluded from asset-building credit markets (Dwyer 2018; Morduch and Schneider 2017; Wherry, Seefeldt, and Alvarez 2019) and remain overrepresented among subprime borrowers (Charron-Chénier 2018; 2020; Rugh, Albright, and Massey 2015). Notably, subprime borrowing is far more expensive than mainstream borrowing and includes borrowing via ‘predatory’ credit mechanisms. Moreover, because debt is expected to be repaid, borrowing may ultimately increase economic precarity and other economic stress for families, particularly those high-cost (subprime) debt.

Debt can be viewed as contributing to or constituting economic precarity when it is used to support ongoing consumption due to inadequate income and, in particular, when it is characterized by utilization of high-cost borrowing. That is, debt acquired and, especially, accumulated via high-cost unsecured credit, such as credit cards and alternative financial services loans, may imply that a family’s other economic resources have fallen short of its consumption needs. Those shortfalls may arise due to a range of factors including a drop in income or benefits, expense shocks, or changes in family or household composition. Research indicates that families with low incomes and earnings (Maroto 2021; Amorim and Schneider 2022), and those in poverty (Fitzpatrick and Coleman-Jensen 2014), are disproportionately likely to turn to high cost credit, including subprime financial institutions, compared to those with greater income and wealth (Chen, Friedline, and Lemieux 2022; Charron-Chénier 2020). As such, financially burdensome (unsecured) debt, itself, can be understood to be an indicator of economic precarity (McCloud and Dwyer 2011; Sweet 2021; Finnigan and Meagher 2019; Parolin 2023).

Moreover, unpaid and past due debts (delinquent debts) directly signal constrained financial resources, and often come with additional fees or penalties that further aggravate economic precarity. Indeed, past due debts are associated with other dimensions of economic precarity, such as income poverty and limited availability of household financial resources, and subjective financial stress (psychological stress emanating from economic hardship) (Bradshaw and Finch 2003; Gauthier and Jr 2010; Marks 2007; Neckerman et al. 2016). In turn, they are associated with poorer mental health (Drentea and Lavrakas 2000; Drentea and Reynolds 2015; Dunn and Mirzaie 2016; Mendes de Leon and Griggs 2021; Sun and Houle 2020; Sweet et al. 2013; Rhodes, Dwyer, and Houle 2024). Thus delinquency constitutes another and distinct credit-related indicator of economic precarity.

Given the racial wealth gap as well as historic and continued racial discrimination in financial markets, credit-related indicators of economic precarity may capture differences by race in navigating uncertainty that have implications for CPS involvement, particularly in the context in which Black families are already highly overrepresented in CPS systems. Black families may face even more precarity than White families but, at the same time, typically have lesser access to credit, with the exception of the highest cost alternative financial services, which are disproportionately located in minoritized neighborhoods (Bea et al. 2023).

In short, high levels of utilization of high-cost credit, including card use, alternative financial services borrowing, and past due payments are key indicators of ‘debt strain,’ which comprises an important component of economic precarity in the contemporary United States. The inclusion of debt strain as a form of economic precarity serves to capture the financial consequences of a myriad of sources of financial unpredictability in the lives of (particularly lower-income) families, including adequacy and volatility of earnings, social welfare benefits, and family composition. Given well-established links among economic precarity, child maltreatment, and CPS involvement, it is possible that debt strain may be associated with CPS-involvement risk.

The Current Study

To the best of our knowledge, research has yet to examine whether debt strain may be associated with CPS involvement, with the sole exception of housing-related debt (falling behind on a mortgage or rent) (Berger et al. 2015; Berg and Brännström 2018). Berger and colleagues (2015) find experiencing a home foreclosure filing to be associated with an increased likelihood of CPS involvement. Moreover, they find that risk of CPS involvement rises in the months prior to the foreclosure filing itself (when a family is likely to be increasingly falling behind on its mortgage payments) and remains elevated in the year following the filing. Others have studied the relation between evictions (which are more common than foreclosures among low-income populations) and CPS involvement. For example, Berg & Brännström (2018) find that children from evicted families are twice as likely than other children to be placed in out-of-home care, while Bullinger & Fong (2021) find a positive association between the number of eviction notices and CPS cases at the neighborhood level. However, the relation between debt strain and CPS involvement, outside of the context of housing insecurity, has not to the best of our knowledge been explored.

Given that debt is becoming an increasingly common component of household finances (Munsch, 2024), debt strain may constitute an additional explanatory mechanism of the association of economic precarity, in general, with CPS involvement. Thus, it is important to examine whether there is an explicit association of debt strain with CPS involvement, as such a relation may imply the need for debt-specific policy reform that is separate from economic support-related (transfer) policy levers directed at alleviating economic precarity more generally. While state social welfare benefit generosity has been linked to lesser unsecured borrowing (Berger, Brown, et al. 2024; Berger, Bruch, et al. 2024), debt-specific policies, such as regulations on AFS lending and ‘usury laws,’ which regulate interest and fees, may also be important for reducing debt strain. Thus, examining whether there is a relation between debt strain and CPS involvement has important implications for understanding the full range of costs and benefits of such policies.

Our study extends prior literature to examine whether several indicators of debt strain—high credit card utilization, accumulation of subprime AFS debts, and/or delinquent overdue payments on any credit—are associated with increased CPS involvement, and whether there is heterogeneity in these associations by race and income. We expect a positive relation between debt strain and CPS involvement, particularly for reports alleging child neglect, given that neglect has been more explicitly linked to economic precarity than have other forms of child maltreatment, such as child abuse (Slack et al. 2024; Berger and Waldfogel 2011). Since neglect allegations make up the majority of CPS reports, we expect a positive association of debt strain with child neglect to drive positive associations of debt strain with CPS reports and out-of-home placement of children (child removal), in general. We expect this association to be weaker for allegations of child abuse, which is less directly linked to economic scarcity (Slack et al. 2024; Berger and Waldfogel 2011).

Because CPS involvement, itself, could lead families to take on debt, in order to manage the costs and disruptions of investigations and subsequent sanctions, we conduct several sensitivity analyses with increasingly stringent restrictions in order to isolate the role of debt as a precursor to subsequent CPS involvement. We also perform heterogeneity (subgroup) analyses by prior CPS involvement status, both because debt strain may have a stronger relation with first-time versus subsequent CPS involvement, and also both to better address the possibility that CPS involvement may lead to debt acquisition and to minimize selection bias that may occur if mothers who have multiple encounters with CPS also have characteristics that are associated with high levels of debt strain. Finally, given the overrepresentation of non-White (particularly Black and Native American) and low-income families in CPS caseloads, as well as disproportionately high levels of debt strain among low-income populations, we expect the relation between debt strain and CPS involvement to be strongest among non-White and low-income populations. Given racial and income gaps in credit access and spatial segregation, these differences may be particularly pronounced for AFS debts.

Data and Methods

Data and Sample

Our data are drawn from two sources: (1) the Wisconsin Administrative Data Core (WADC) (Brown et al. 2020), which includes all individuals participating in a wide range of public social welfare programs in Wisconsin, and (2) the Ohio State University-University of Wisconsin Consumer Credit Panel (OSU-UW CCP), which includes credit history data for the population of individuals in Wisconsin with a credit report. The WADC includes data on CPS allegations, investigations, and out-of-home placements, including the timing of each event, as well as on employment and earnings, social welfare program participation and benefit levels, and demographic characteristics such as race and education. An individual enters the WADC when they interact with one of these systems (e.g., are reported to CPS; have earnings reported to the Unemployment Insurance [UI] program; enroll in a social welfare benefit program such as the Supplemental Nutrition Assistance Program [SNAP], Temporary Assistance for Needy Families Program [TANF], or Medicaid [MA]; enter public school; enter state prison), or are a member of a social welfare benefit program ‘case’ for which another individual is the ‘case head.’ The WADC follows individuals backwards and forwards in time from the point at which they enter. An individual’s presence in the WADC implies that they or the individuals they live with (typically family members) have participated at some point in at least one of the programs included in WADC.

The OSU-UW CCP comprises mainstream proprietary data from one of the major US credit bureaus, including quarterly records on credit accounts of all individuals in Wisconsin who have a credit report (approximately 87% of the adult population). These data are available for the fourth quarters of 2015, 2016, and 2017, and all quarters from 2018–2021. The OSU-UW CCP contains information on the types of accounts individuals hold, the outstanding balance on each account, the status of payments on those accounts, new account openings, and inquiries about opening new accounts, among other elements. The mainstream credit report data are further linked with subprime credit history data a subsidiary of the credit bureau that aggregates data on alternative financial services (AFS) reported by creditors offering (subprime) small dollar, high cost, and shorter-term credit including payday loans, single-period micro loans, and high-interest rate short-term installment loans. Some, but not all, of these loans are also included in the mainstream credit database.

We link the 1,296,840 mothers in the WADC to the OSU-UW credit panel and subsequently drop individuals who are deceased (67,165), who are residing out-of-state throughout the study period (151,999), and who were less than 13 or more than 60 years old (100,827) at any point during the study period. We restrict our period of analysis from 2016 to 2021, using the fourth quarter of 2015 only to construct lagged values of some predictors of interest. Our analyses leverage a balanced panel of 17,583,282 quarterly observations of 976,849 mothers spanning the 18 quarters for which we have data.

Measures

CPS involvement.

Our data include information on CPS involvement at the child level, including details on investigations and child removals, such as relevant dates (referral, investigation start and end dates, case opening and closing date, out-of-home placement start and end dates), and the type(s) of maltreatment alleged. We aggregate this information to the mother level and create an indicator for whether any CPS events (investigations overall and by alleged maltreatment type, and/or child removals) occurred for any child linked to a mother in each quarter of our observation period.

Debt strain.

We use information in the OSU-UW CCP on both concurrent and pre-existing debt to construct three measures of debt strain all calculated at the mother-quarter level. High credit card utilization is defined as having outstanding debt totaling 70% or more of one’s total credit limit on all credit cards combined. A credit card utilization rate of more than 30 percent of available credit can negatively impact one’s credit score, and utilization of 70 percent or above signals economic precarity (Office of the Comptroller of the Currency 2024). We construct mother-quarter indicators of high credit card utilization, which we lag by one quarter in our empirical analyses to both ensure proper causal ordering between debt strain and CPS involvement and rule out reverse causality. We estimate models for high credit card utilization using the full sample as well as a sample restricted to those with a credit card balance. Alternative Financial Services (AFS) use is a mother-quarter indicator that the mother has at least one AFS account, such as a payday or auto title loan, which charge subprime and, in some cases, predatory, levels of interest. We lag the AFS use indicator by one quarter in our analyses. Delinquency is a mother-quarter indicator that, over the prior 6-month period, a mother had any account or trade (credit action) for which payment is (or was) 60–180 days overdue. We view this measure as a proxy that a mother’s expected debt payments have exceed her economic resources. Delinquent payments serve reduce one’s credit score, are at risk of being sent to collections (which may more severely harm one’s credit score), and may have legal consequences. Since our measure of delinquency is lagged by definition, we use its value in the current quarter in our analyses. Finally, we include in our analyses a mother-quarter indicator of any debt strain, defined as the mother having utilized greater than 70 percent of their credit limit in the prior quarter, having taken an AFS loan in the prior quarter, and/or having any delinquent payments in the prior six-month period.

Covariates.

Because any association between CPS involvement and debt strain may be correlated with mother and family characteristics, we adjust in our regression analyses for a range of covariates, including maternal race, education, and age at first birth, quarterly earnings reported to the UI system, lagged credit score, and indicators that the mother is employed (again drawn from the UI system), is receiving TANF, and is receiving UI benefits. We also include county of residence and quarter-year fixed effects to account for county- and time-specific unobserved factors such as CPS practices, access to AFS institutions, and macroeconomic trends.

Analytic Approach

We estimate associations of debt strain with CPS involvement using both standard and conditional (mother-specific fixed-effects) logistic regressions. The standard logistic regressions consist of between-mother analyses comparing the likelihood of CPS involvement in a given quarter for mothers who did and did not experience debt strain in the prior quarter (prior six months with respect to delinquency), net of the full set of covariates described above. Standard errors from these models are clustered at the mother level. Whereas these models leverage the full sample of mothers and adjust for a range of observed characteristics, they are subject to potential bias due to unobserved factors.

The conditional logistic regressions consist of within-mother analyses comparing the likelihood of CPS involvement for the same mother in periods with and without (lagged) debt strain. This approach has the advantage of adjusting for unobserved time-invariant factors that may differentiate mothers by both debt strain and CPS-involvement status and, if not accounted for, may bias estimates of associations between the two. Given that there is likely selection into both debt strain and CPS involvement based on factors beyond those we observe, these analyses are particularly informative, and their estimates can be interpreted as less subject to bias than those from the standard logit models.

Notably, however, the conditional logit estimates, like those from the standard logits, are subject to bias due to unobserved time-varying factors and cannot, therefore, be assumed to be causal in nature. In addition, the conditional logit estimates are informed only by within-mother variation over time in debt strain and CPS involvement; information from mothers who exhibit no variation over time in these factors does not contribute to their estimates. In essence, this means that the standard and conditional logits are estimated on different samples, with the standard logits being estimated for all mothers and the conditional logits being estimated only for those who exhibit variation in debt strain and CPS involvement. Were this not the case, the difference in estimates between the two models would be interpreted as being explained by differences in time-invariant maternal characteristics that are associated with (contribute to selection into) both debt strain and CPS involvement. To better assess the extent to which differences in our estimates are driven by differences in the samples versus the specification of each model, we also estimate standard logistic regressions using only the sample of mothers who contribute to the conditional logit estimates. For ease of interpretation, we present our estimates as odds ratios (exponentiated logit coefficients) with associated confidence intervals.¹

Results

Descriptive Statistics

Table 1 presents descriptive statistics for the characteristics of our sample, overall and by debt strain status. The mothers in our sample, on average, were 23 years old when they gave birth to their first child, have more than two children in their household, are predominantly White (62 percent), and have a high school degree/GED or greater education (56 percent). About 28 percent have no UI-reported earnings in any quarter, and five percent have no UI-reported earnings in a given quarter. Among those with reported earnings, average quarterly earnings are about $8,800. Approximately three percent of mothers receive UI benefits and less than one percent receive TANF. Fourteen percent of mothers receive child support and about 1.2 percent make child support payments. Mothers with debt strain are, on average, from more vulnerable populations, as we would expect for an indicator of economic precarity; specifically they were younger (22 versus 24) at the time of their first birth, have more children (2.6 versus 2.3), are disproportionately non-White (18 percent versus 11 percent), and are disproportionately likely to have less than high school education (20 percent versus 16 percent) than those without debt strain. Debt-strained mothers are also far less likely to have no UI-reported earnings at all (19 percent versus 33 percent) but more likely to have no UI-reported earnings in a given quarter (7 percent versus 5 percent), conditional on ever having UI-reported earnings. Mothers with debt strain earn, on average, $2,000 less per quarter than mothers without debt strain; however, a similar proportion of mothers with and without have below median earnings in the sample. Finally, mothers with debt strain are significantly more likely than those without debt strain to receive TANF and UI benefits and, in particular, to be involved in the child support system, with 20 percent of mothers with debt strain receiving and two percent paying child support, versus 12 percent and 1 percent of mothers without debt strain.

Table 1.

Sample characteristics, overall and by debt strain status

	Full Sample	Any debt strain	No debt strain	Difference: Any v. no debt strain
Observations (persons)	976,849	569,821	407,028
Observations (person-quarters)	18,560,131	5,422,169	13,137,962
Mother’s age at first birth	23.06 (5.38)	22.13 (5)	23.45 (5.48)	−1.32***
Number of children	2.34 (1.44)	2.57 (1.56)	2.25 (1.38)	0.32***
Race
Non-Hispanic White	61.79%	61.40%	61.92%	−0.56***
Non-Hispanic Black	13.23%	18.20%	11.16%	7.04***
Missing	16.09%	16.95%	15.7%	−1.22***
Education
Less than high school	17.23%	19.57%	16.25%	3.32***
High school degree/GED or greater	55.78%	67.13%	51.06%	16.07***
Missing	26.9%	13.30%	32.69%	19.38***
Earnings
% did not match with UI data	28.25%	18.43%	32.34%	−13.91***
% with no earnings	5.29%	7.11%	4.49%	2.62***
Mean when positive	8,868.18 (7,218.59)	7,293.42 (5,408.97)	9,752.59 (7,990.29)	−2459.17***
% below sample median# ($7,630)	24.31%	34.19%	34.19%	0
% above sample median ($7,630)	24.31%	23.13%	24.83%	−1.70***
Program participation
% receiving UI benefits	3.57%	4.88%	2.99%	1.89***
% receiving TANF	0.72%	1.51%	0.38%	1.13***
% receiving child support	13.76%	19.48%	11.24%	8.24***
% paying child support	1.21%	2.08%	0.83%	1.25***

Note: Means (and standard deviations) or percentages presented for mother-quarter-level observations.

^#Median is defined on quarterly wage when positive.

⁺p < .10

^*p < .05

^**p < .01

^***p < .001.

Table 2 shows average CPS involvement rates for sample mothers during the observation period, overall and by debt strain status and type. Mothers who have had debt strain are about three times more likely to also experience CPS involvement for one or more of their children (9.9 percent versus 1.8 percent), with mothers with delinquent credit accounts being at greatest risk (11.9 percent), followed by those with AFS loans (11.2 percent), and those with high credit card utilization (10.4 percent). Those with debt strain are about five times more likely to have investigations that involve a neglect allegation, seven times more likely to have investigations that involve an abuse allegation, and five times more likely to have a child removed, with relatively similar risk patterns across the debt strain types.

Table 2.

CPS involvement rates, overall and by debt strain status

		Debt strain
	Full Sample	Any debt strain	Delinquency	High CC Utilization	AFS loans	No debt strain	Difference: Any vs no debt strain
Observations	976,849	552,415	400,514	505,970	87,814	424,434
Percent of sample	100	56.55	41.00	51.80	8.99	43.45
Any CPS involvement	6.41%	9.93%	11.92%	10.36%	11.23%	1.83%	8.10***
Any investigation	4.34%	6.91%	8.55%	7.24%	7.44%	0.99%	5.92***
Neglect investigation	3.38%	5.23%	6.18%	5.46%	6.16%	0.98%	4.25***
Abuse investigation	0.88%	1.41%	1.77%	1.48%	1.29%	0.19%	1.22***
Any child removal	6.48%	10.03%	12.03%	10.46%	11.36%	1.86%	8.17***

Note: Column percentages presented for mother-level observations.

⁺p < .10

^*p < .05

^**p < .01

^***p < .001

Standard and Conditional Regression Analyses

Table 3 presents results from standard logistic regressions for the full sample (Panel A), standard logistic regressions for the conditional (maternal fixed-effect) logistic regression sample (Panel B), and conditional logistic regressions (Panel C). The standard logistic regression results (Panel A) indicate that mothers with any debt strain are 23 percent more likely to experience a CPS investigation in the subsequent quarter than those with no debt strain, and that those with delinquency and high credit card utilization are, respectively, 26 and 62 percent (53 percent among those with a credit card balance) more likely to experience a CPS investigation than those without each form of debt strain. In contrast, those with AFS debt are 5 percent less likely than those without AFS debt to experience a CPS investigation. We see a similar pattern of estimates for investigations that included child neglect and those that included child abuse allegations, although the magnitudes of the estimates are larger with respect to neglect than abuse. The pattern is also similar for child removals, with the notable exception that AFS loans are not significantly associated with child removals.

Table 3.

Standard and conditional logistic regression results of associations between debt strain and CPS involvement

	Any Investigation	Neglect Investigation	Abuse Investigation	Child Removal
Panel A: Standard logistic regression, full sample
Any debt strain	1.227*** [1.206,1.248]	1.308*** [1.281,1.335]	1.166*** [1.138,1.194]	1.394*** [1.326,1.465]
Delinquent	1.263*** [1.241,1.285]	1.356*** [1.327,1.384]	1.191*** [1.162,1.221]	1.494*** [1.422,1.570]
High CC utilization (full sample)	1.618*** [1.578,1.658]	1.991*** [1.927,2.057]	1.421*** [1.374,1.469]	2.155*** [1.981,2.345]
High CC utilization (those with CC balance)a	1.527*** [1.488,1.567]	1.852*** [1.790,1.916]	1.350*** [1.303,1.398]	1.995*** [1.830,2.175]
Has AFS accounts	0.951*** [0.916,0.988]	0.905*** [0.863,0.950]	0.935** [0.888,0.986]	1.02 [0.912,1.140]
Observations	17,583,282	17,583,282	17,583,282	17,583,282
Panel B: Standard logistic regression, conditional (maternal fixed-effect) sample
Any debt strain	1.047*** [1.030,1.063]	1.040*** [1.020,1.060]	1.023* [1.001,1.046]	0.971 [0.920,1.025]
Delinquent	1.079*** [1.062,1.095]	1.071*** [1.051,1.090]	1.054*** [1.031,1.077]	0.965 [0.921,1.011]
High CC utilization (full sample)	1.092*** [1.069,1.117]	1.074*** [1.044,1.105]	1.042*** [1.012,1.074]	0.847*** [0.772,0.929]
High CC utilization (those with CC balance)a	1.325*** [1.293,1.358]	1.343*** [1.300,1.387]	1.255*** [1.213,1.298]	1.197*** [1.100,1.303]
Has AFS accounts	0.987 [0.955,1.019]	1.002 [0.962,1.044]	0.982 [0.941,1.026]	1.244*** [1.104,1.402]
Observations	1,126,530	763,092	594,864	154,854
Panel C: Conditional (maternal fixed-effect) logistic regression
Any debt strain	1.020* [1.001,1.039]	1.020*** [0.997,1.044]	1.012 [0.985,1.039]	1.022 [0.967,1.081]
Delinquent	1.048*** [1.029,1.068]	1.056*** [1.033,1.080]	1.048*** [1.016,1.081]	1.016 [0.962,1.074]
High CC utilization (full sample)	1.011 [0.981,1.042]	1.020 [0.980,1.062]	0.981 [0.942,1.023]	0.955 [0.861,1.058]
High CC utilization (those with CC balance)a	1.039*** [1.005,1.075]	1.059*** [1.012,1.107]	0.992 [0.948,1.039]	0.999 [0.888,1.125]
Has AFS accounts	1.064*** [1.018,1.112]	1.059*** [1.001,1.119]	1.048 [0.979,1.122]	1.379*** [0.977,1.117]
Observations	1,126,530	763,092	594,864	154,854

Note: The estimates presented in each cell are the odds ratio and 95% confidence interval from a logistic regression of the relevant CPS involvement measure (column label) on the relevant debt measure (row label). All models control for unemployment, wage, TANF receipts, net child support receipts, a lagged variable for credit score, county and quarter fixed effects. Standard logit models also include time-invariant controls for race, education, and mother’s age at first birth and cluster errors at the individual level.

⁺p < .10

^*p < .05

^**p < .01

^***p < .001

^aExcludes individuals with no credit card balance (N=11,163,539). Conditional (fixed-effects) sample sizes are 470,944 for any report, 280,906 for neglect report, 260,403 for abuse report, and 44,795 for child removal.

Panel B of Table 3 shows these results for the subsample of mothers contributing to the conditional logit estimates, those who exhibit variation over time in debt strain and CPS involvement. Thus, by definition, all mothers in this sample experienced CPS involvement at some point during the observation period. The estimates represent the difference in the likelihood of CPS involvement in a given quarter between mothers who did and did not experience debt strain in the prior quarter, net of the observed covariates, and among only those mothers who experienced debt strain and CPS involvement at some point during the observation period. These results are consistent with those from the full sample in that any debt strain, delinquency, and high credit card utilization are associated with greater CPS involvement. Notably, however, the magnitudes of the coefficients are substantially smaller than those for the full sample, with the conditional logistic regression estimates indicating that mothers with any debt strain are five percent, those with delinquency are eight percent, and those with high credit card utilization are nine percent (32 percent for those with a credit card balance) more likely than those without debt strain to experience CPS investigations in the subsequent quarter. The magnitudes of these estimates are, respectively, 23 percent, 26 percent, and 62 percent (53 percent) among the full sample. In addition, AFS loans are not significantly associated with CPS investigations in this subsample, whereas they are associated with a decreased likelihood of CPS investigations in the full sample. Also in this subsample, AFS loans are associated with a 24 percent greater likelihood of child removal, whereas they are not associated with child removals in the full sample. Finally, all forms of debt strain, except AFS loans, are associated with an increased likelihood of child removal in the full sample, but only high credit card utilization and AFS loans are significantly associated with child removal in this subsample.

The substantial difference in the estimates produced by standard logistic regressions using the full sample of mothers and the subsample of mothers who experienced both debt strain and CPS involvement at some point during the observation period suggests that unobserved differences between mothers who ever and never experienced debt strain and CPS involvement account for a large portion of the magnitudes of the estimated associations of debt strain with CPS involvement found in the full sample. The change in the estimates for AFS loans suggests that those differences may be particularly linked to credit access and unobserved variation in the likelihood of using the highest cost loans.

Results from conditional logit regressions are presented in Panel C of Table 3. The estimates represent the within-mother difference in the likelihood of CPS involvement in a given quarter when a mother did and did not experience debt strain in the prior quarter, again net of the observed covariates and among only mothers with variation in debt strain and CPS involvement during the observation period. These results indicate that, when mothers experience debt strain in a given quarter, they are more likely to be investigated by CPS in the subsequent quarter than they would otherwise be. Notably, this finding holds for all types of debt strain, with the exception of high credit card utilization, among all mothers in this sample (those both with and without credit card balances). It holds for high credit card utilization among the subsample of mothers with a credit card balance but not for those without a credit card balance.

The estimates from the conditional (maternal fixed-effect) logistic regression specification are generally smaller in magnitude than those produced by the standard logistic regressions using the same sample, suggesting that experiencing debt strain is associated with a 2–6 percent greater likelihood of CPS involvement whereas, in the standard logit models, it is associated with a 5–32 percent greater likelihood of CPS involvement, depending on the debt measure. Moreover, in this specification, we find taking AFS loans to be associated with a 6 percent greater likelihood of a CPS investigation, whereas we find negative association in the standard logistic regression specification. Again, these associations are particularly salient for investigations that include a child neglect allegation. Indeed, in this specification, only the association of delinquency attains statistical significance with respect to investigations that include physical abuse. Finally, we find AFS loan taking to be associated with a 38 percent greater probability of child removal, whereas none of the other forms of debt are significantly associated with child removal in the conditional logistic regression specification.

Taken together, our estimates indicate that debt strain is associated with an increased likelihood of CPS investigation, particularly for neglect allegations and, to a lesser extent, with child removal, albeit with variation by types of debt strain. That the estimates from standard logistic regressions using the full sample are considerably larger than those from standard logistic regressions using the subsample of mothers who are relevant to the conditional (maternal fixed-effect) logistic regression specification (those with variation over time in debt strain and CPS involvement) suggests that unobserved differences between mothers who experience debt strain and CPS involvement and those who do not account for a substantial portion of the estimated between-mother association of the two. Differences in the estimates produced by the standard and fixed-effects (conditional) logistic regressions using the identical (fixed-effects) sample imply that differences in unobserved time-invariant maternal characteristics between mothers who do and do not experience debt strain and CPS involvement account for an additional portion of the estimated between-mother association of debt strain with CPS involvement. Nonetheless, we find the association between debt strain and CPS involvement to hold for most forms of debt strain in each specification. Moreover, the within-mother estimates produced by the fixed-effects models—which are smaller in magnitude than those produced by the other specifications with respect to all of the debt measures except AFS loans (for which the conditional logistic regression estimates are larger)—likely more closely approximate the causal effect of debt strain on CPS involvement than those produced by the standard logistic regression models.

Heterogeneity Analyses

We conducted heterogeneity analyses to examine whether associations between debt strain and CPS involvement may vary with respect to (1) first-time versus subsequent incidents of CPS involvement, given that debt strain and CPS involvement may have reciprocal and dynamic relations over time; (2) race (for White versus Black mothers), given that Black families are disproportionately involved with CPS and face even more credit constraint; and (3) mothers’ earnings from wages (below versus above sample median quarterly earnings of $7,630 for mothers with any UI-reported earnings), given that lower-income families are disproportionately involved with CPS. These results are presented in Table 4. For parsimony, we present only the estimates for ‘any’ debt strain and not those for each type thereof.

Table 4.

Logistic regression results of associations between debt strain and CPS, by subgroups

	Unadjusted Mean	Standard logistic regression	Standard logistic regression, conditional (maternal fixed-effect) sample	Conditional (maternal fixed-effect) logistic regression
Panel A: First-time versus subsequent CPS involvement
First-time CPS involvement
Any debt strain	0.003	1.183*** [1.160,1.206]	1.013 [0.995,1.032]	1.033*** [1.009,1.057]
Observations	17,379,670	17,379,670	917,747	917,747
Subsequent CPS involvement
Any debt strain	0.187	1.008a [0.974,1.043]	1.008 [0.974,1.042]	0.999 [0.966,1.032]
Observations	203,612	203,612	196,063	196,063
Panel B: Race
Non-Hispanic White
Any debt strain	0.010	1.231*** [1.203,1.260]	1.051*** [1.029,1.073]	1.026* [1.001,1.052]
Observations	10,865,502	10,865,502	645,930	645,930
Non-Hispanic Black
Any debt strain	0.013	1.090***b [1.050,1.132]	1.016 [0.980,1.054]	1.015 [0.975,1.057]
Observations	2,324,793	2,324,793	231,511	231,511
Panel C: Income
Low Income
Any debt strain	0.014	1.217*** [1.186,1.248]	1.042*** [1.018,1.067]	1.027+ [0.998,1.057]
Observations	4,314,518	4,314,518	350,397	350,397
High Income
Any debt strain	0.005	1.031c [0.988,1.077]	0.981c [0.943,1.021]	1.015 [0.964,1.068]
Observations	4,330,560	4,330,560	131,548	131,548

Note: The estimates presented in each cell are the odds ratio and 95% confidence interval from a logistic regression of the relevant CPS involvement measure (column label) on the relevant debt measure (row label). All models control for unemployment, wage, TANF receipts, net child support receipts, a lagged variable for credit score, county and quarter fixed effects. Standard logit models also include time-invariant controls for race, education, and mother’s age at first birth and cluster errors at the individual level.

⁺p < .10

^*p < .05

^**p < .01

^***p < .001

^aSignificantly differs from first-time CPS involvement at p < .05.

^bSignificantly differs from non-Hispanic white at p < .05.

^cSignificantly differs from low income at p < .05.

The results in Panel A of Table 4 suggest that the association of debt strain with CPS involvement may be particularly salient for first-time CPS involvement rather than subsequent CPS involvement, though the estimates from the two models are significantly different only in the full-sample standard logistic regression specification. The results in Panel B indicate that this association may differ for White and Black families though, again, the estimates from the two models are significantly different only in the full-sample standard logistic regression specification. For White families, debt strain is associated with a greater likelihood of CPS involvement across all models, even the most stringent conditional logistic regression, indicating that selection may be less substantial for White families than Black families. For Black families, the effect size is significant in the standard logit model (and smaller than that for White populations) but is nonsignificant both in the standard model when estimated using the conditional model sample and in the conditional model, indicating that selection may be especially important for the Black population. The results in Panel C reveal that debt strain may be particularly salient for lower-income families and less so for higher-income families, consistent with our expectations that debt strain is particularly related to economic precarity for disadvantaged populations (the estimates from the two models are significantly different in both standard logistic regression specifications, but not in the conditional logistic regression specification).

Sensitivity analysis

Because we have data for only the fourth quarter of the years 2015–2017 and not for each quarter of those years, our lagged measures of high credit card utilization and AFS use in the fourth quarters of 2016 and 2017 reflect debt in the fourth quarter of the prior year, rather than in the prior quarter. We conducted subgroup analyses for pre- and post-2018 to test whether the one-year lag resulted in different estimates than the one-quarter lags analyzed in post-2018 quarters. Results (available upon request) are similar to those from our main results in both periods, but the estimates on the debt strain variables during the earlier period, characterized by a one-year lag in debt strain, are smaller in magnitude than the coefficients in the later period, characterized by a one-quarter lag in debt strain, suggesting that the longer lag during the earlier period may serve to underestimate the true association of debt strain with CPS involvement in our main analyses.

Finally, about 15 percent of the mothers in our sample appear to be ‘credit invisible,’ in that they do not have a credit report. In our main analyses, code these mothers as not having debt strain given credit invisible individuals cannot encumbered the types debt strain we consider. That said, it could be that credit invisibility is differentially associated with CPS involvement than is the absence of debt strain among the credit visible. As such, we conducted supplementary analyses in which we modeled credit invisibility as distinct from lack of debt strain. Results (available upon request) indicate that distinguishing these two groups has little effect on the estimates of relations of debt strain with CPS involvement produced by our primary models. Indeed, our estimates are quite consistent in both specifications. However, we also find that the relationship between credit invisibility and CPS outcomes are sensitive to changes in model specifications, with our preferred fixed effects specification yielding a negative association between CPS involvement and credit invisibles for each debt strain variable.

Discussion

We used linked longitudinal credit record data and administrative data on CPS reports to investigate associations between debt strain and CPS involvement for mothers in Wisconsin. Our findings indicate that, as expected, debt strain, including high credit card utilization, accumulation of AFS loans, and delinquent payments in a given quarter, is associated with an increased likelihood that a mother’s children will be investigated by CPS in the subsequent quarter, particularly due to child neglect allegations. It is also associated with an increased likelihood that a mother’s children will be placed in foster care in the subsequent quarter. These associations are particularly large in magnitude in between-mother analyses (standard logistic regression specifications), but remain statistically and substantively significant, albeit considerably smaller in magnitude in most cases, in within-mother analyses of associations of changes in debt strain with subsequent CPS involvement (conditional mother-specific fixed-effect logistic regression specifications).

Notably, the estimates for child abuse investigations are considerably smaller in magnitude and less likely to attain statistical significance than those for child neglect. Indeed, only the association of delinquencies, and no other form of debt strain, attains statistical significance with respect to child abuse in the conditional logistic regression specification. This is in line with our expectation that debt strain would be more strongly associated with child neglect than with child abuse, given that neglect is more closely linked to economic precarity (Slack et al. 2024; Berger and Waldfogel 2011). We also find stronger associations of debt strain with first-time CPS involvement than with subsequent CPS involvement, possibly suggesting that mothers who have multiple encounters with the CPS are a select group whose involvement with CPS is determined by factors other than economic precarity, and/or that CPS involvement and debt strain have bidirectional and reciprocal relations.

Contrary to our expectation, we find debt strain to potentially have a stronger association with CPS investigations for White than Black mothers. However, this finding corroborates previous research that has found economic resources to be stronger predictors of CPS involvement for White versus Black families (Thomas, Waldfogel, and Williams 2023; Thomas and Waldfogel 2022), implying that factors beyond economic precarity (such as systemic racism and bias) may play a role in the disproportionate involvement of Black families with CPS. Relatedly, this finding may further reflect that White mothers have greater access to credit (are more able to borrow) or are more credit visible (more likely to borrow using credit mechanisms that are tracked by credit bureaus) compared to Black mothers (Wherry, Seefeldt, and Alvarez 2019). That we find a positive association for Black mothers in the standard logistic regression but not the conditional (fixed-effects) specification supports that differential access to credit may play a role. That is, this finding does not necessarily indicate that debt strain differentially ‘affects’ White and Black mothers’ likelihoods of coming to the attention of CPS, conditional on equal ability to borrow. Future research should further investigate this pattern and seek to identify the mechanisms underlying differential associations of debt strain with CPS involvement by race, if confirmed. In contrast, consistent with our expectation, we find that debt strain is more closely linked to increased risk of CPS involvement for lower-income mothers than their higher-income counterparts.

Turning to child removal, results from standard logistic regression specifications indicate that debt strain, in general, and most types thereof, is associated with an increased likelihood of child removal. Yet, results from conditional (mother-specific fixed-effect) regressions, indicate that only AFS loan taking is associated with an increased likelihood of child removal. This overall pattern suggests that unobserved factors that differentiate mothers who do and do not experience debt strain account for a large portion of the estimated association of debt strain with child removal using the ‘naïve’ (standard between-mother) regression approach. At the same time, the finding that AFS loan taking has a large association with child removal, in particular, is somewhat puzzling. One possible explanation is that mothers who resort to AFS loans are facing particularly severe economic distress, which may also result in a higher probability that they and their children’s circumstances will lead to child removal (e.g., due to inadequate access to goods and services to meet children’s basic needs and/or parental mental health, substance use, stress, and associated parenting behaviors).

Our study has several limitations that are important to consider in contextualizing and interpreting our results. First, CPS involvement in any given quarter is relatively rare, as are some types of debt strain such as AFS loans. This limits our flexibility in model specification. Second, our analyses cannot fully rule out the possibility of reverse causation, despite that we model lagged measures of debt strain relative to the timing of CPS involvement and also conduct subgroup analyses for first-time and subsequent CPS involvement. Existing research on economic precarity and CPS involvement has mainly focused on the effect of economic precarity on subsequent CPS involvement, as we do here. Yet, it is also possible that CPS involvement may lead to greater economic precarity and debt strain, for example by causing parents to miss work to attend mandated services, requiring parents to commit resources to specific goods or services to meet CPS mandates, inducing reductions in the social welfare benefits available to a household when children are removed from the home, and charging parents child support to recover government expenditures on foster care for their child(ren) (Cancian, Yang, and Slack 2013). Future work should examine these potential bidirectional relations. Third, we lack credit data for the first three quarters of 2016 and 2017. As such, our lagged debt strain measures for the fourth quarters of 2016 and 2017 reflect debt strain in the fourth quarter of the previous year rather than in the previous quarter. Results from sensitivity analyses examining, separately, each of these periods suggest that the pronounced lag during the 2016–2017 period is likely to underestimate the association of debt strain with CPS involvement, such that our estimates likely provide a lower bound of the true association of debt strain with CPS involvement in this period. Fourth, despite our use of mother-specific fixed-effects regressions, our estimates cannot be assumed to represent causal effects of debt strain on CPS involvement as they are subject to bias by time-varying unobserved factors, such as parental substance use, physical health and disability status, mental health, and domestic violence. Finally, our sample is drawn only from Wisconsin residents. While we have no a priori expectation that associations of debt strain with CPS involvement would differ in other state contexts, we cannot be sure that our findings generalize to populations in other states. Future research would be well served to examine whether and how state level variation in CPS and/or debt regulatory policy may affect the relation between debt strain and CPS involvement.

With these caveats in mind, this study provides new evidence on the relation between economic precarity and CPS involvement. Specifically, we show both that mothers who experience debt strain are more likely to become involved with CPS than those who do not, and that, when mothers experience the onset of debt strain, the likelihood that they will become involved with CPS increases. Future research should further interrogate relations of indebtedness, debt strain, and CPS involvement, as well as parenting behaviors themselves, with particular attention to variation therein by race and ethnicity and income. Further understanding these relations may have implications for debt-related policies such as usury laws regulating interest rates and other policies specific to subprime lending (e.g., state bans on payday or auto title loans). Research on a ‘credit-welfare state trade-off’ suggests that more generous state social welfare transfers may reduce the use of private fallbacks such as high-cost debt during times of need and, thereby, result in lesser debt strain (Berger, Brown, et al. 2024; Berger, Bruch, et al. 2024). Indeed, helping families avoid debt strain may be a key mechanism through which income support policy reduce CPS involvement (as shown e.g., in Berger and Waldfogel 2011; Fong 2019b; Font and Maguire-Jack 2020; Pac et al. 2023). Thus, our findings suggest that child welfare practice may better serve families by considering their ‘full’ economic contexts, including their expenditures and debts, as well as their earnings and benefits received.