The neglected ones: Time at home during COVID-19 and child maltreatment

Abstract

The early months of the COVID-19 pandemic led to extreme social isolation, precarious employment and job loss, working from home while tending to children, and limited access to public services. The confluence of these factors likely affects child health and well-being. We combine early release child maltreatment reports in Indiana with unique and newly available mobile phone movement data to better understand the relationship between staying at home intensively during the COVID-19 pandemic and child maltreatment. Our findings indicate that the prolonged stays at home promoted by the early public health response to COVID-19 resulted in reductions in child maltreatment reports overall and substantiated reports of maltreatment. However, relative to areas that stayed home less, children in areas that stayed home more were more likely to be both reported for and a confirmed victim of maltreatment, particularly neglect. These areas have historically been socioeconomically advantaged and experienced lower rates of maltreatment. We only observe increases in confirmed child maltreatment in metropolitan counties, suggesting that the effects of staying home on child maltreatment may reflect both the differential risk of leaving home and access to services in metropolitan–rather than non-metropolitan–counties. Staying at home has been challenging for many families. Families likely need assistance as the pandemic persists, evolves, and when it ends.

1. Introduction

In the spring of 2020, the United States experienced the first wave of the Sars-Cov-2 (or the novel coronavirus known as COVID-19). As a result, nearly 95% of Americans were under instructions to “stay-at-home” to curtail the spread of COVID-19. By May 2020, most states had lifted or relaxed these stay-at-home orders (National Academy for State Health Policy, 2020). Understanding families' struggles during the first wave of the pandemic will be paramount to child serving agencies and policymakers as they respond to the needs of families as we emerge from this unprecedented chapter in our nation's history and prepare for future crises.

The early months of the pandemic, which is the primary focus of this paper, required many Americans to stay home exclusively. This presented families with a host of new challenges, and as a result, American family life changed dramatically and abruptly. Working parents faced stress surrounding any combination of the following: unemployment, meeting the in-person expectations of an essential job that did not allow for staying at home, working from home, a shift in employment responsibilities, and/or other acute employment pressures.

Parents had to navigate this new employment landscape with one additional complication: as a result of school and daycare closures due to public health precautions, their children were home, too. Beyond the early employment changes, COVID-19 imposed other substantial challenges to parenting including being solely responsible for providing safe and consistent care to children without the typical support of daycare or schools, lack of daily routines and normalcy, limited hours for essential services (e.g., grocery stores), increased parental stress, and social isolation (Gassman-Pines, Ananat, & Fitz-Henley II, 2020). Indeed, emerging research indicates that the pandemic was associated with stark increases in parents’ reports that they had lost their temper with their child and felt overwhelmed by the responsibility of being a parent (Kalil, Mayer, & Shah, 2020). Each of these factors are important for navigating parenthood and caregiving, and the compounding of these experiences are particularly concerning to child health and welfare professionals (Abramson, 2020, American Academy of Pediatrics, 2020).

In this paper, we describe how staying at home due during COVID-19′s first wave influenced rates of child maltreatment. Child maltreatment, which includes sexual abuse, physical abuse, and neglect, is a pervasive problem in the United States. In 2018, 678,000 children were victims of substantiated child maltreatment (U.S. Department of Health & Human Services [DHHS], 2020). At some point in their childhoods, 37% of all children are the subject of a Child Protective Service (CPS) investigation (Kim et al., 2017). Neglect, which is characterized by inadequate supervision and failure to provide a child with basic necessities, including safety, is the most common form of child maltreatment. It accounts for approximately 75% of child maltreatment reports and is present in nearly 80% of child maltreatment related deaths (DHHS, 2020).

Ex ante, the effects on children of spending more time at home with their parents or primary caregivers are unclear. If this increased time translates into quality “family time,” then children might be better off. However, the popular press is replete with stories of parents experiencing heightened fatigue and providing distracted or disengaged supervision as they juggle childcare, homeschooling, and their formal employment responsibilities in the wake of COVID-19 stay-at-home orders (see e.g., Edwards & Snyder, 2020). Moreover, many of the typical resources (local public services, places of worship, daycare, etc.) available to families were either stressed due to high demand or also temporarily closed due to stay-at-home orders. Stated differently, COVID-19 created stressors at both the family and community levels. Contractions of resources – such as those brought about by COVID-19 – could put children at an increased risk of maltreatment.

Given the emergent nature of COVID-19 research, data for answering questions surrounding COVID-19 and outcomes like child maltreatment are limited – despite their immediate importance. Common sources of child maltreatment information (e.g., formal child maltreatment reports to Child Protective Services (CPS) nationwide, medical records, parent- and child-reported measures) are not yet available or are slowly becoming available to the research community. For example, the National Child Abuse and Neglect Data System (NCANDS): Child File usually releases data to the research community with a roughly two-year lag.

We use early release child maltreatment report data made publicly available by one state agency: Indiana’s Department of Child Services (DCS). We combine these data with proprietary mobile phone movement data on Indiana residents. We take great care to confirm the validity of the data, as they are unique and novel, and we are confident that they are well suited to shed light on COVID-19′s influence on child maltreatment.

We draw on these two unique sources of data to describe (1) overall trends in child maltreatment during the pandemic and (2) the relationship between the intensity of staying at home during the first wave of COVID-19 and child maltreatment.

Our interpretation of the findings is framed by the assumption that child maltreatment trends are a good representation, but not an exact measure, of true child maltreatment both before and during the COVID-19 pandemic. If changes in surveillance during COVID-19 is the primary source of change among child maltreatment reports, then lower child maltreatment report rates induced by isolation during the pandemic may imply that true maltreatment did not change but simply became undetected. We address this question below.

We show that across Indiana (and consistent with other jurisdictions’ experiences across the United States as reported by media outlets), substantiated reports of maltreatment declined in April and May 2020, likely due to reduced reports from educators as a result of school closures (Baron et al., 2020, Bullinger et al., 2021). However, counties in Indiana with the largest increases in time spent at home between early March 2020 and early April 2020 experienced significantly more maltreatment reports and substantiated cases of maltreatment relative to counties whose residents stayed home less. The results are driven by substantiated child neglect. Areas with relatively more substantiated maltreatment during the first wave of the pandemic are generally socioeconomically advantaged and historically have had lower rates of maltreatment. Finally, this finding is unique to metropolitan counties.

Our findings represent a deviation from typical trends in child maltreatment research. According to our calculations, the counties for which we observe the highest relative increase of substantiated maltreatment had the lowest, pre-pandemic substantiation rates and the fewest traditional risk factors for reported child maltreatment (these counties have the lowest poverty rate, highest median income, and a larger share of the population identifying as Non-Hispanic white). Yet within the reduced counts, patterns emerge that show a heterogeneous experience of children during the pandemic. Specifically, within the reports that did occur, children in counties who were at home the most experienced increased substantiated reports compared to children from counties with the lowest time at home. Policymakers should consider the unique challenges to providing safe and consistent care for all children during a pandemic while implementing appropriate public health responses. Though this paper focuses on child maltreatment in the early pandemic, we note that the COVID-19 pandemic is not over, could evolve (Cohan, 2020), and is not guaranteed to be an isolated event (Erlanger, 2021). Therefore, as governments issue, re-issue, or prolong the closures of public services, like schools, they may need to issue complementary policies or services to off-set or mitigate the challenges families face stemming from new time and economic constraints. Finally, understanding the hardships families endured during COVID-19 will be critical in formulating a post-pandemic response that helps families heal.

2. Theories related to child maltreatment

2.1. Primary causes of child maltreatment

Extant research has focused on two primary causes of child abuse and neglect (parental psychopathology and economic hardship); however, research also indicates that there may be different pathways between the causes and the most common maltreatment subtypes (i.e., physical abuse, supervisory neglect, and physical neglect). An extensive literature has examined the ways in which parental psychopathology, including stress and depression, are associated with increased risk for child maltreatment, particularly child abuse (Kempe et al., 1962, Pinderhughes et al., 2000, Taylor et al., 2009). A second line of research has sought to understand the ways in which economic hardship increases the risk for child maltreatment, particularly supervisory and physical neglect (Berger, 2004, Bullinger et al., 2020). Psychopathology and economic hardship likely work both independently and in tandem to increase the risk for child maltreatment. Economic hardship, for example, may exacerbate or instigate mental health problems, increasing the risk for child maltreatment. Economic hardship and mental health problems diminish two resources that are paramount for positive parenting involvement: time investments in children (Sandberg & Hofferth, 2001) and economic resources, namely income (Yeung et al., 2002).

2.2. How stress and mental health affect child maltreatment

According to prior research, stress, depression, and related mental health problems increase the risk for child maltreatment by reducing parents’ ability to cope with the rigors of parenting (Easterbrooks et al., 2013). Depression, for example, may influence parents’ perceptions of children’s behavior leading to harsher parenting and potentially increased child maltreatment (Bugental & Happaney, 2004). Research drawing on the Family Stress Model (Conger et al., 2000) indicates that parental stress is linked to increased conflict between parents, which in turn is associated with increased harsh and detached parenting. Moreover, prolonged intense parenting – especially when few respite or support resources are available – may lead to parental burnout, which can increase the risk of neglect and abuse (Griffith, 2020).

2.3. How economic hardship affects child maltreatment

Economic hardship has increasingly been identified as a primary causal factor in child maltreatment, particularly child neglect (Bullinger et al., 2020, Feely et al., 2020). This work often argues that economic hardship may diminish parents’ ability to provide safe and consistent care for children by increasing their exposure to unstable and unsafe environments, reducing parents’ ability to provide materially for children, and increasing stress, depression, and other mental health problems that inhibit coping skills (Feely et al., 2020, Pelton, 2015). Emerging work investigating the causal effects of income on child maltreatment largely finds that income support programs reduce the risk for child maltreatment (Berger et al., 2017, Cancian et al., 2013, Raissian and Bullinger, 2017), while negative economic shocks increase the risk (Lindo et al., 2018, Schenck-Fontaine et al., 2017, Schenck-Fontaine and Gassman-Pines, 2020, Schneider et al., 2017).

A robust sociological and economic literature has investigated the important role that parental resources, including time and income, play in providing parents with the means to parent effectively (Becker, 1981, Brooks-Gunn and Duncan, 1997). This literature demonstrates that income influences child wellbeing through its effect on parents, namely, less parenting stress, more parental warmth, and less use of harsh discipline (Linver et al., 2002, Mistry et al., 2002).

Though employment is traditionally considered a way for families to obtain economic resources, which promotes child well-being and would be expected to reduce maltreatment, the relationship between employment and maltreatment is complex. Employment introduces competing and potentially off-setting demands for caregivers’ time and introduces the need for consistent childcare, which is often unaffordable for formal and licensed arrangements or puts parents at the mercy of unreliable informal childcare arrangements. Market work may decrease the risk for child maltreatment if it effectively provides parents with the ability to purchase additional resources that enable them to provide safe and consistent care of their children (e.g., childcare). Alternatively, employment may increase the risk for child maltreatment if it is unstable or in other ways reduces parents’ ability to ensure their children’s safety (Paxson & Waldfogel, 2002) (e.g., if parents are unable to purchase or afford quality childcare). However, there is a dearth of research on the mechanisms between unemployment, including the nuances of the effects of the employment of different household members, and maltreatment, which has multiple subtypes that could be differentially affected by household employment patterns. Although the pathways between these factors have not been specified or demonstrated, studies have established that community-level changes in employment often result in changes in net changes in maltreatment rates.

Theory and prior work suggest that when employment, i.e., demands on parents' time, is combined with potential economic and psychological distress, the risk for child abuse or neglect increases if there are no appropriate child supervision options. Without access to traditional supports the Family Adjustment and Adaptation Response Model (FAAR) model indicates that child neglect may increase as parents are forced into an imbalance between resources and demands. In this regard, the abrupt, erratic, unusual, and simultaneous changes in employment and childcare as a result of COVID-19 are likely to change the demands and available resources of most families.

2.4. Macro shocks and child maltreatment

Research on the effect of macro shocks such as natural disasters on child maltreatment is limited. This work generally hypothesizes that natural disasters increase stress and experiences of adverse mental health that may result in increased child maltreatment. Prior work examining outcomes such as traumatic brain injury in young children in the wake of hurricanes has found mixed results. Keenan et al. (2004), for example, find increased traumatic brain injuries among young children in counties severely affected by hurricane Floyd in North Carolina. Similarly, Curtis et al. (2000) examined three natural disasters, hurricane Hugo, the Loma Prieta earthquake, and Hurricane Andrew. They found significant increases in child abuse reports in the aftermath of hurricane Hugo and the Loma Prieta earthquake, but not after hurricane Andrew.

Other work has examined the influence of macroeconomic shocks on child maltreatment. Some research indicates that rising unemployment is associated with increased child maltreatment (Brown and De Cao, 2020, Frioux et al., 2014) while other studies imply that unemployment may actually be protective against child maltreatment if it provides children with greater supervision and increased access to caregivers (Paxson and Waldfogel, 2002, Raissian, 2015). Lindo et al. (2018), for example, find that rising male unemployment is associated with increased child maltreatment while rising female unemployment is associated with decreased child maltreatment. Similar work has examined the effects of mass job layoffs on child maltreatment. Schenck-Fontaine et al. (2017) find that job losses increase the severity of child maltreatment cases, but their findings are restricted to economically disadvantaged areas. In sum, although the literature on the effect of disasters and macro shocks on child maltreatment is mixed, there is growing evidence that these events are linked to increased child maltreatment through economic and stress-based pathways.

3. Covid-19 policy responses and child maltreatment

The public health threat from COVID-19 has had a clear effect on parents’ economic resources and implications for time allocations and investments in their children. Indiana’s Governor, Eric Holcomb, was among the first American governors to declare a public health emergency on March 6, 2020. This declaration coincided with the diagnosis of the first known COVID-19 positive case in the state. In addition, 46 counties in Indiana (out of 92) declared independent public health emergencies (authors’ calculations)1 . Other public policy responses to reduce the spread of COVID-19 included stay-at-home orders, school closures, and face mask mandates, among others. For example, both public and non-public K-12 schools were closed on March 19, 2020, at first temporarily, and then for the remainder of the 2019–2020 Academic Year.2 Daycares were not ordered to close, but many did close (at least temporarily) due to staffing, child exits from care, and the additional expense of new public health measures (Rickert, 2020).

These policies may be related to child maltreatment through strains on household financial resources, parental time, and/or mental health. First, the pandemic has introduced unprecedented increases in unemployment and financial hardship throughout the U.S. (Unemployment Insurance Weekly Claims, 2020). In compliance with Indiana’s stay-at-home order, effective March 25, 2020, non-essential businesses closed, which rendered many parents furloughed or unemployed. This sudden economic shock may have increased child maltreatment through the inability to purchase goods and services that serve to reduce the risk for maltreatment (e.g., safe housing, food security, childcare). Even among parents who remained employed, families may face more economic hardships from reduced work hours, pay cuts, or unpaid leave from work.

A second group of parents were able to work from home and maintain a stable income if they were able to adequately fulfill their employment responsibilities. This second group sizable group. The Bureau of Labor Statistics (Dey et al., 2020) estimates that about 45% of workers were in jobs that allowed for teleworking, and approximately 42–50% of parents were in jobs that allowed for teleworking. Teleworking placed parents under substantial pressure to split their time between market work and childcare. Because of the need to work and care for children simultaneously, the pandemic may have increased child neglect - and in this scenario, supervisory neglect in particular - as employed parents were forced to manage the conflicting requirements of work and childcare without the support of childcare centers and schools.

Finally, parents who are essential workers and were required to leave their homes for work, were confronted by a lack of traditional out of home childcare and schools, and generally faced challenges in navigating childcare while they continued to work. The pandemic also heightened the risk and therefore stress of their employment, often without appropriate compensation.

For parents and caregivers in each of these scenarios, the pandemic has likely imposed additional mental health strain, increases the risk of physical abuse. The widespread uncertainty associated with the spread of COVID-19 itself, social isolation from social distancing, and absence of regular resources (including childcare) have all contributed to difficulties in coping with everyday life. Early research indicates that both parental and child psychological wellbeing deteriorated in the early months of the pandemic (Gassman-Pines et al., 2020). Prescriptions for antidepressants are reported to have increased by 12 percent over previous years (Edney, 2020) and are also in short supply. Due to both increases in demand and disruptions in the supply chain, the Food and Drug Administration has added commonly prescribed tablets to a list of drugs experiencing a shortage.

Importantly, theory indicates that – as is the case with COVID-19 – these processes need not occur in isolation, but are likely mutually re-enforcing. Stay-at-home orders may have simultaneously increased economic hardship, mental health problems, and time constraints which together may further increase the risk for child maltreatment. The Family Stress Model, for example, indicates that economic hardship increases parental stress, resulting in parental conflict and harsh parenting. Similarly, the FAAR predicts that economic hardship may increase stress. The FAAR model describes a process through which families draw on their resources and capabilities to balance and adjust to demands and crises. In the context of COVID-19, new employment demands emerged, traditional resources disappeared, and few or no substitute childcare resources materialized. Such a dynamic could increase familial conflict and stress and parental burnout and withdrawal, resulting in child maltreatment. Indeed, early evidence indicates increases in domestic violence across the U.S. as a result of stay-at-home orders (Bullinger et al., 2020, Leslie and Wilson, 2020, Sanga and McCrary, 2020).

In sum, theory and prior empirical research indicate that stay-at-home orders may increase child maltreatment through three potentially interactive pathways. First, if parents become unemployed as a result of the crisis then increased economic hardship might result in child maltreatment. Second, the challenges associated with staying home intensively might increase psychological distress leading to child maltreatment, regardless of unemployment status. Third, among parents who remained employed, the competing demands of work and childcare may have resulted in increased child neglect as parents were unable to provide safe and consistent childcare in the face of simultaneous work and lack of out-of-home childcare and school.

Theory and extant research indicate that parents who closely adhered to state emergency orders to shelter in place may have been at increased risk for child maltreatment through the mechanisms described above. If greater time spent at home during the pandemic is linked to increased maltreatment, then comparing areas with greater or lesser compliance with the stay at home order may provide important insights.

Although the present study cannot adjudicate between mechanisms, each of these forces may contribute to changing trends in child maltreatment during the pandemic and to the influence of intensive staying at home on different forms of child maltreatment.

There may also be important differences by geography. Although the literature is mixed, some work indicates that child maltreatment is higher in rural areas among White families and in urban areas among non-whites (Maguire-Jack, Jespersen, Korbin, & Spilsbury, 2020). Raissian (2015) also shows that unemployment has a differential effect on metropolitan versus non-metropolitan counties. Although the pandemic altered traditional patterns of access to services, these too vary by geography (Allard, 2017). Widespread closures, potentially greater fear of infection, and more stark changes in daily routine in urban areas might indicate differential effects by urbanicity. In this sense, although the stay at home order was universal, the variation in adherence to the order provides a unique opportunity to examine the overall and geographical effects on child maltreatment.

4. Study data and methods

4.1. Child maltreatment reports

To examine the relationship between time at home during the pandemic and child maltreatment, we use two primary sources of unique and newly available data. First, to measure child maltreatment, we use early release, administrative records of child maltreatment reports to Indiana Child Protective Services (CPS) from January 2013 through May 2020. These records document the number of assessed reports of child abuse, including sexual abuse, physical abuse, and neglect, for each month in all counties, by disposition (substantiated or unsubstantiated). The sample consists of all 92 Indiana counties from January 2013 through May 2020 (89 month-years) measured monthly (N = 8,188).

The primary outcome measure is the monthly incidence of substantiated reports of maltreatment made to the Indiana Department of Child Services (DCS) (i.e., confirmed maltreatment following a CPS investigation). We constructed a rate per 10,000 children. We focus on substantiations – as opposed to total child maltreatment reports – because reports have dropped precipitously across the country following stay-at- home orders. In Indiana after a maltreatment report is made staff follow a detailed protocol for conducting an assessment. Per DCS's policy, the assessment process lasts no more than 45 days from the time of the initial report and includes a home assessment, Interviews with the child (depending on age), the parent, the alleged perpetrator (if different than parent), the reporting source(s), and any witnesses to the incident. The evidence is reviewed and assessed for credibility and the investigating case manager – and when necessary in consultation with their supervisor – makes a finding of “substantiated” or “unsubstantiated”. A finding of substantiated is made “when facts obtained during the assessment provide a preponderance of evidence sufficient to lead a reasonable person to believe that Child Abuse and/or Neglect (CA/N) has occurred or when the alleged perpetrator admits to having abused and/or neglected the alleged child victim” (Indiana DCS, 2020).

Although overall reports are good indicators of child risk and exposure to maltreatment in the aggregate (Drake, 1996, Hussey et al., 2005), educational personnel account for about 20% of reports both nationally and in Indiana (authors’ calculations using NCANDS: Child File 2017 data). Substantiation rates (1) offer a less volatile measure of child maltreatment, and (2) require evidence of harm. Therefore, substantiation rates allow us to understand if children experiencing confirmed maltreatment changed following stay-at-home orders. We also report results for child maltreatment report rates, and our findings are robust to outcome variables.

4.2. Time spent at home

Data on social distancing and physical movement come from SafeGraph, Inc. a data company that aggregates anonymized location data from smartphone devices. In April 2020, SafeGraph, Inc. provided their “Social Distancing Metrics” to researchers through free, non-commercial agreements. SafeGraph reports that it tracks 35 million unique devices per month that have turned on their location services with exact known location in the United States. This database assigns each device to a home location (a 153-meter by 153-meter area that receives the most frequent GPS pings) using its nighttime (6 pm to 7 am) location. Among other measures, the data track the percent of each day that a device dwells at its home location. The data reported to researchers reflect the median percent of the day that devices spend at home at the Census block group level. For confidentiality reasons, SafeGraph excludes Census block groups with <5 devices. For a balanced panel, we drop block groups that are not in the panel every day during our time period (0.04% of block groups). We then aggregate the data to the county-month level to merge with the county-month maltreatment data. These data have been used in recent analyses examining the effect of state government restrictions due to COVID-19 on mobility (Dave et al., 2020, Farboodi et al., 2020, Friedson et al., 2020, Gupta et al., 2020).

4.3. Analytical approach

To assess whether prolonged stays at home during the COVID-19 pandemic affected child maltreatment, we first calculate the change in the time spent at home between the beginning of March and the beginning of April for each county.3 We do this because we seek to investigate changes in time spent at home as a result of the pandemic (i.e., relative to some baseline) rather than simply the amount of time spent at home following the emergency declaration. Then we divide counties into four quartiles based on the change. Fig. 1 shows the differences in both the percent of the day spent at home and the percent of devices that spent the entire day at home across quartiles. On average, counties in Indiana increased their time at home by 26.8 percentage points, or roughly 6.4 h per day between weeks 10 and 14 of 2020. During that same time, the percent of devices staying completely at home increased by 19 percentage points, a 90 percent increase. Counties that had the most dramatic increase in their time spent at home during this time period increased by 7.3 h per day, on average, compared to 5.3 h per day on average among the lowest quartile. We use this variation in the increase in time spent at home to construct four different groups of counties.4

Fig. 1.

Trends in time at home by quartile of changes in time spent at home.

We compare trends in child maltreatment report rates before and after Indiana’s Governor declared an emergency due to COVID-19 on March 6, 2020 across changes in the time spent at home among these four groups. Specifically, we estimate the following equation:

$$Y_{\mathit{cmy}}=\alpha +{\beta }_1{{Q2}_c\ast Post}_{\mathit{my}}+{\beta }_2{{Q3}_c\ast Post}_{\mathit{my}}+{\beta }_3{{Q4}_c\ast Post}_{\mathit{my}}+{\log \left(adultpop\right)}_{\mathit{cy}}+{\delta }_c+{\gamma }_{\mathit{my}}+{\epsilon }_{\mathit{cmy}}$$ (1)

where Y is the child maltreatment substantiation rate for county c in month m during year y. Q2 = 1 if county c was in the second quartile of increased time at home and zero otherwise. Q3 = 1 if county c was in the third quartile of increased time at home, and Q4 = 1 if the county was in the fourth quartile. The first quartile (i.e., the group of counties that increased their time at home the least) is the reference group. Post represents the quarantine effects and equals 1 if m = April or May and y = 2020, and zero otherwise.5 The coefficients of interest are ${\beta }_{1-3}$, the interactions between each quartile binary variable and the Post binary variable. This approach is akin to a difference-in-differences model with three treatment groups, or a treatment intensity model, where each of the top three quartiles is a treatment group. The reference (control) group is the first quartile: those counties that had the smallest change in their time at home.

Additional components of the main model include county fixed effects, ${\delta }_c$, which account for characteristics of a county that may be correlated with both substantiated child maltreatment rates and whether residents of a county stayed home, such as a county’s culture or attitude toward aggressive parenting behaviors. Month-year, ${\gamma }_{\mathit{my}}$, fixed effects account for seasonality and general trends in reports over time, such as drops that might have occurred statewide due to school closures, allowing for month-to-month comparisons. We also adjust for the log of the adult population, which is a time-varying proxy for metropolitan status. Due to the recency of data, we are limited in our ability to adjust for potential confounders. However, studying changes over time in the same geographical area helps to account for within-county, statewide, and seasonal changes in parenting behaviors. All regressions are weighted by child population, and standard errors are clustered at the county-level to account for serial correlation.

4.4. Challenges with using child maltreatment reports

The month in which the data are reported reflect cases that were closed in that month. Indiana’s DCS policy requires that cases be closed within 45 days from the date of the initial report to the assessment (i.e., a decision on whether the report is substantiated or not). To determine the proportion of cases that were initially reported in the same month in which they closed, we compared the Indiana early release data to the National Child Abuse and Neglect Data System: Child File (NCANDS)–the national data repository–in Indiana for March 2017 (the latest year available for NCANDS at the time of the analysis). It took approximately 33 days to close a substantiated case of neglect in both metropolitan counties and non-metropolitan counties (authors’ calculations). Therefore, it is possible that some reports in April 2020 could have been made initially in March 2020 and that some reports initially made in April 2020 were not closed until May 2020. We have taken three steps to mitigate this concern.

First, we use April and May 2020 data as the post-COVID time periods. Due to the 45-day requirement by DCS, we are confident that nearly all reports closed in the post-period were initially reported in the post-period, as well.

Second, the closure date in these early release reports and the initial report date in the final data the state of Indiana submits to NCANDS are highly correlated; historical data from January to May 2017 indicate a correlation of 0.82 for all substantiations and 0.86 for neglect substantiations. Using these correlations, we surmise that approximately 82 percent of substantiations closed in a given month were also initially reported in that same month.

Finally, since we compare counties to themselves in earlier time periods, methodologically these differences should not affect our results. As long as the distance between initial report and case closure within counties are not substantially different in April and May 2020 compared to earlier months, the results from this study reflect differential trends in April and May 2020 relative to preceding trends.

5. Results

5.1. Descriptive statistics

Before presenting our main results, we first present descriptive statistics. Fig. 2 shows trends in the total report rate and the total substantiation rate for the most recent 12 months of our study period, split across the increase in time spent at home quartiles (authors' calculations, Indiana Department of Child Services). These trends demonstrate several noteworthy points. First, both reports and substantiations fell during April and May 2020 for all counties in Indiana. This trend is consistent with early evidence from Florida (Baron et al., 2020), news reports that calls to child maltreatment hotlines are down in neighboring Illinois (Jackson, 2020), and declines in 911 calls reporting child abuse in Chicago, Illinois (Bullinger et al., 2020). Second, counties that increased time at home the least (quartile 1) consistently have the highest rate of reports and substantiations of all groups (authors' calculations, Indiana Department of Child Services). Similarly, the group of counties that increased their time at home the most (quartile 4) consistently has the lowest rate of reports and substantiations. Quartiles 2 and 3 are generally in between these two groups. Finally, between June 2019 and March 2020, trends in both report and substantiation rates across the four quartiles are parallel, lending support for our analytical approach.

Fig. 2.

Trends in maltreatment report rates by quartile of changes in time spent at home.

Table 1 presents descriptive statistics (based on 2017 data) for each quartile. Approximately 53% of children in Indiana reside in the top two quartiles. Counties that increased their time at home the most (quartile 4) tend to be relatively less populous, have higher incomes and lower poverty rates, and have a larger share of the population identifying as non-Hispanic white (authors' calculations, American Community Survey). In addition, children in these counties are historically less likely to be confirmed victims of maltreatment (authors' calculations, Indiana Department of Child Services).

Table 1.

Descriptive statistics by increases in time spent in the home from March 1–15 to April 1–15.

	Quartile 1	Quartile 2	Quartile 3	Quartile 4
Average Increase in Hours Spent at Home Per Day	5.3	6.3	6.7	7.3
Sexual Abuse Substantiation Rate Per 10,000 Children	1.5	1.7	1.7	1.0
Physical Abuse Substantiation Rate Per 10,000 Children	1.1	1.3	0.9	0.8
Neglect Substantiation Rate Per 10,000 Children	17.9	16.0	13.4	9.8
Total Maltreatment Substantiation Rate Per 10,000 Children	20.6	19.0	16.1	11.6
Total Maltreatment Report Rate Per 10,000 Children	150.9	121.0	122.5	80.3
Metropolitan County (0/1)	0.80	0.71	0.84	0.74
Population	530,339	229,163	185,609	146,936
Poverty Rate (%)	0.17	0.14	0.13	0.08
Median Income ($)	48,159	52,530	55,015	73,064
Percent of the Population that is:
Non-Hispanic White (%)	0.72	0.77	0.83	0.89
Non-Hispanic Black (%)	0.17	0.12	0.08	0.03
Hispanic (%)	0.08	0.10	0.07	0.05
Asian (%)	0.03	0.01	0.02	0.03
Aged 0–17 (%)	0.23	0.24	0.24	0.25
Aged 18–25 (%)	0.14	0.10	0.10	0.09
Aged 26–45 (%)	0.26	0.24	0.25	0.25
Aged 46–64 (%)	0.23	0.26	0.25	0.26
Aged 65+ (%)	0.14	0.17	0.16	0.15
Number of Counties (% of total child population)	23 (29.1)	23 (18.2)	23 (30.4)	23 (22.3)

Sources: 2020 SafeGraph data, 2017 DCS data, 2017 Census Bureau’s Small Area Income and Poverty Estimates Program, and 2017 National Center for Health Statistics Bridged-Race Population Estimates.

5.2. Main results

Moving to the main results, Table 2 presents the fully adjusted models, based on Eq. (1). Table 2 shows how trends in maltreatment reports changed in April and May 2020 relative to the trends before COVID-19 (January 2013-March 2020) differentially across counties by the increase in time spent at home. Overall, the figure shows universal decreases in rates of maltreatment report and substantiation per 10,000 children. In an effort to identify the effects of staying at home during the pandemic we compare geographies where households stayed at home intensively to those where they stayed at home less intensively. When we look at changes in substantiated reports for sexual abuse and physical abuse in April and May 2020, we do not see any differences among the quartile groups. Substantiated reports of neglect are higher in quartile 4 during April and May 2020 relative to the same period for quartile 1 counties. Specifically, substantiated neglect reports in quartile 4 counties increased an average of 2.45 reports per 10,000 children more than the counties in quartile 1. When we compare this to the pre-COVID-19 mean of 12.34 reports per 10,000 children, this estimate reflects a relative increase of about 20 percent. Similarly, the overall increase in the substantiation rate is higher in these counties by 2.67 reports per 10,000 children, or about an 18 percent increase. Importantly, these results imply that relative to the large reduction in substantiated reports experienced in quartile 1, counties in quartile 4 experienced a much smaller reduction.

Table 2.

Social distancing & substantiated child maltreatment in Indiana.

	Substantiation rate				Report rate
	Sexual abuse	Physical abuse	Neglect	Total	Total
Post*Quartile 2 (6.3 Hours More/Day)	0.16	−0.28	1.78	1.66	8.67
	(0.31)	(0.18)	(1.48)	(1.61)	(5.57)
Relative % Change	10.7%	−25.7%	14.4%	11.1%	8.3%
Post*Quartile 3 (6.7 Hours More/Day)	−0.24	−0.20	1.24	0.80	6.03*
	(0.26)	(0.17)	(1.34)	(1.36)	(3.59)
Relative % Change	−16.1%	−18.3%	10.0%	5.4%	5.8%
Post*Quartile 4 (7.3 Hours More/Day)	0.35	−0.13	2.45*	2.67*	25.15***
	(0.27)	(0.16)	(1.36)	(1.37)	(3.93)
Relative % Change	23.5%	−11.9%	19.9%	17.9%	24.1%
Mean Y Pre-Covid	1.49	1.09	12.34	14.93	104.37
R2	0.19	0.19	0.55	0.56	0.78
N	8188	8188	8188	8188	8188

Notes: Data from Indiana DCS January 2013-May 2020. Post = 1 for April and May 2020. Outcome: Rates Per 10,000 Children. Models include log(adult population), county FE, and month-year FE, and are weighted by child population. Coefficient estimates represent changes relative to the comparison group: Quartile 1, counties that stayed home 5.3 h more/day. Robust standard errors are clustered at the county-level. *p < 0.10, **p < 0.05, ***p < 0.01.

Although we focus on substantiated reports due to the reporting issues addressed earlier, in the last column of Table 2, we show that the overall report rate also increased in quartile 4 by 24 percent, relative to quartile 1. We also observe relative increases in the total report rate for counties in quartile 3 of about 6 percent. Further analysis of total reports broken down by maltreatment type (available upon request) shows that increases in neglect reports are driving the effect in the total report rate. Our analyses also show relative increases in sexual and physical report rates among quartile 4 counties.

The nature of the COVID-19 pandemic is such that population density plays a large role in infectious disease transmission. Further, the macroeconomic context has different implications for child maltreatment across metropolitan and non-metropolitan areas (Raissian, 2015). Therefore, we next determine whether there are differences in the role of staying at home on child maltreatment across metropolitan status. Table 3 reports the main results disaggregated by metropolitan status.6 The relative increases in neglect and overall substantiations in quartile 4 counties are driven by metropolitan counties. Further, when split by metropolitan status, all three quartiles have significantly more maltreatment reports in April and May 2020 relative to quartile 1. Fig. 3 is a visual depiction of Table 3. As in Table 2, the increases in substantiations are driven by neglect, which may suggest child suffering due to job loss, precarious employment, or less supervision while parents work (Moyer, 2020).

Table 3.

Social distancing & substantiated child maltreatment in Indiana, by metropolitan status.

	Substantiation rate				Report Rate	Relative % Change in Total Substantiation Rate
	Sexual abuse	Physical abuse	Neglect	Total	Total	Total
Panel 1: Metro (n = 44)
Post*Quartile 2 (6.3 Hours More/Day)	0.28	−0.04	3.04**	3.28**	14.50**	23.6%
	(0.33)	(0.12)	(1.47)	(1.58)	(5.59)
Post*Quartile 3 (6.8 Hours More/Day)	−0.34	−0.11	2.04	1.59	8.96**	11.5%
	(0.28)	(0.13)	(1.34)	(1.35)	(3.58)
Post*Quartile 4 (7.3 Hours More/Day)	0.44	0.13	3.70***	4.27***	30.39***	30.8%
	(0.30)	(0.11)	(1.14)	(1.04)	(3.13)
Mean Y Pre-Covid	1.36	1.00	11.53	13.88	99.45
R2	0.27	0.28	0.67	0.67	0.82
N	3916	3916	3916	3916	3916
Panel 2: Non-metro (n = 48)
Post*Quartile 2 (6.2 Hours More/Day)	−0.03	−1.01**	−1.48	−2.53	−3.77	−13.6%
	(0.49)	(0.48)	(3.22)	(3.44)	(11.64)
Post*Quartile 3 (6.6 Hours More/Day)	0.20	−0.50	−2.33	−2.63	−7.03	−14.1%
	(0.53)	(0.59)	(2.73)	(3.00)	(9.16)
Post*Quartile 4 (7.5 Hours More/Day)	0.20	−0.97**	−2.19	−2.96	8.31	−15.9%
	(0.46)	(0.46)	(3.01)	(3.19)	(10.28)
Mean Y Pre-Covid	1.97	1.42	15.25	18.64	121.92
R2	0.08	0.11	0.35	0.34	0.63
N	4272	4272	4272	4272	4272

Notes: Data from Indiana DCS January 2013-May 2020. Post = 1 for April and May 2020. Outcome: Rates Per 10,000 Children. Models include log(adult population), county FE, and month-year FE, and are weighted by child population. Coefficient estimates represent changes relative to the comparison group: Quartile 1. Robust standard errors are clustered at the county-level. *p < 0.10, **p < 0.05, ***p < 0.01.

Fig. 3.

Effect of relatively larger increase in time at home during COVID-19 on substantiated child maltreatment reports (April and May 2020 relative to quartile 1 of changes in time spent at home).

5.3. Robustness checks

Our results indicate that children in counties that increased their time at home the most between March and April 2020 were more likely to be confirmed victims of maltreatment than children in counties that were less likely to increase their time at home. We next test the sensitivity of our results to changes in functional form and study time period. We report these results in Appendix A.

First, the public health response to the COVID-19 pandemic began dramatically changing in mid-March. For example, Indiana Governor Holcomb ordered K-12 schools to close beginning on March 19, 2020. Since we have data aggregated to the monthly level, we are limited in our ability to look more carefully at March 2020, when COVID-19 precautions were going into effect (e.g., at the weekly or daily level). In the main results, March 2020 is treated as a non-COVID-19 month; in other words, the Post binary variable equals zero. However, column 2 of Appendix A shows that when we drop March 2020 from the analysis–thereby comparing April and May 2020 to January 2013 through February 2020–the results are nearly identical for both substantiation rates (Panel A) and overall report rates (Panel B).

Second, the time period in our main model consists of January 2013 to May 2020. For our primary analysis, we opt for more statistical power since the recency of the data make us unable to adjust for potential confounders. When we limit the sample to May 2019 through May 2020 (12 months of data), results are less precise, but substantively similar for both substantiations and total reports.

Third, given that quartile 1 has substantiation and report rates that are approximately 88% higher than those in quartile 4, we probe whether or not our results are due to scaling in the outcome variables. That is, the substantiation rate in quartile 1 can fall much more than the substantiation rate in quartile 4 simply because quartile 1 rates have a much higher starting value. Therefore, we log the substantiation rates, which allows us to perform the analysis with percent changes rather than level changes. As can be seen in Appendix A, although we lose precision in the substantiation result, the point estimate is the same (approximately 18 percent for quartile 4). We prefer our main model because we are able to include counties with zero substantiation rates.

Finally, instead of the quartile-based analysis, we also explore nonlinearities using the amount of time spent at home and the quadratic form. The results from this analysis are consistent with our main results. Specifically, the substantiation rate decreases at an increasing rate by time spent at home, relative to quartile 1 (results available upon request).

6. Limitations

This study has limitations, which should be taken into account when interpreting the results and also addressed further when conducting future research. COVID-19 has affected families throughout the country (and world). While there has been differential spread in the disease across geographic areas, there are concerns about data quality related to infection and community transmission rates. Furthermore, the indirect consequences of the disease are more likely to affect children than the direct consequences of the illness. Therefore, we use the change in hours spent at home to capture how a community is responding to COVID-19. To be sure, a community’s ability and desire to stay home are likely endogenous, and all counties in Indiana increased their time at home in the initial phase of COVID-19 lockdown. Our estimates are a measure of that intensity, and should be interpreted as such.

We use early release maltreatment data that are reported monthly at the county level. The data do not include detailed information on the victim, reporter, perpetrator, or date the report was made. Further, the measures of child maltreatment reports are cases where the assessment closed in a given month. Therefore, they reflect the month that a decision was made by a CPS worker, and for some cases may not be the same as the month the report was made. To confirm that our results are not sensitive to the timing of assessment decisions, we additionally test multiple time periods including ending the sample period in (1) April 2020, (2) May 2020, and (3) June 2020, and our results are robust to each of these time periods (results available upon request). Further, counties that increased their time at home between early March and early April are generally the same counties that continued to say home throughout the rest of April 2020 (authors’ calculations). Because of the methodological choice to determine changes at home at the start of the COVID-19 public health response and the average time between initial report and assessment decision of 33 days, we opt to include April and May 2020 in the main results. However, our general conclusions about how families fared after COVID-19 related policies were implemented are robust to this decision. Notably, although the time to case closure from prior years indicates that cases were generally closed within 45 days, it is possible that the pandemic resulted in case closure times beyond the norm.

We interpret our findings to mean that substantiated child maltreatment in quartile 4 increased relative to quartile 1 during the COVID-19 response. An alternative interpretation is that substantiated child maltreatment in quartile 1 declined more than that in quartile 4 (see Fig. 2). As the descriptives in Table 1 show, the quartile 4 counties are much more advantaged on typical markers of socio-economic status. The concept of “surveillance bias” has been well described in the literature (Ards, Myers, Malkis, & Hagerty, 2003), and a potential limitation of this study is that the data do not reveal who made the child maltreatment report to DCS. To understand how this omitted variable might affect our study, we used the NCANDS: 2017 Child File data to compute the percent of referrals that were made by each reporter type both in Indiana and in each of our quartile groups. The distribution of reporters is strikingly similar across our quartiles. For example, education personnel are the largest reporting group in Indiana and made 18.85% of all child maltreatment reports in 2017. Education personnel made 19.5% of the reports in quartile 4 and 19.67% of the reports in quartile 1 counties. This provides confidence that the children residing in our constructed quartiles are subject to similar kinds of surveillance and reporting tendencies.

All states (including Indiana) are expected to make their child maltreatment data available to the National Child Abuse and Neglect Data System: Child File. Once these data are available, researchers can probe more nuanced questions. As noted before, historically, the early release reports used in this study and the final data Indiana submits to the National Child Abuse and Neglect Data System: Child File–the national data repository–based on date of initial report are strongly correlated. While we do not doubt the internal validity of our study, extensions around subgroups and external validity to other states are warranted by the research community.

The data we use are at the county-level, and there is likely within-county variation in both maltreatment rates and staying home. Moreover, when using aggregate data, one should worry about an ecological fallacy, though using aggregate data to study child maltreatment is not uncommon due to data access limitations.

The smartphone mobility data are a new data source for social policy research. Though they are a convenience sample based on a panel of smartphone devices, the sample of 35 million users is well-balanced across demographics and geographies (Farboodi et al., 2020), and represents a proxy for how counties responded to COVID-19, on average. However, we cannot distinguish between essential and non-essential mobility.

Finally, we are limited in our ability to adjust for potential confounders due to the recency of data. Studying changes over time in the same geographical area helps to account for within-county, statewide, and seasonal changes in parenting behaviors. However, given the quick nature of the change in time spent home, it is unlikely that many covariates varied at the county level between the April to May 2020 timeframe, and so this is unlikely to be a major – if any – source of bias.

7. Discussion & conclusion

Our findings indicate that, relative to areas that stayed home less, children in counties that stayed home more in response to COVID-19 policies were more likely to be both reported for and a confirmed victim of maltreatment, particularly for neglect. Importantly, these results took place within a context of total absolute declines in maltreatment reports across quartiles and maltreatment types. Further, the sizable relationships are likely conservative estimates due to statewide declines in overall reports to CPS, if the declines in CPS reports are primarily the result of undetected child maltreatment rather than true declines in maltreatment. These findings are driven by metropolitan counties that typically have lower rates of maltreatment, higher incomes, and more non-Hispanic white residents. This latter finding is consistent with recent evidence on the effect of stay-at-home orders on domestic violence in that increases in domestic violence are driven by households without a history of domestic violence and among married households with children (Bullinger et al., 2020, Leslie and Wilson, 2020, Sanga and McCrary, 2020).

At first glance, this result may be somewhat surprising based on the assumptions of the etiology of child maltreatment. We also cannot disaggregate neglect into its component parts, for example, medical, educational, supervisory, or material neglect. We do, however, think it is possible - especially given that our results are driven by neglect changes in quartile 4, the most economically advantaged group - that supervisory neglect may be responsible for at least part of our results. Such a theory is consistent with the experiences of working parents and is plausible when interpreted in the context of their COVID-19 experiences. Consistent accounts emerged via several outlets, which demonstrated children of working parents were at an increased risk for supervisory neglect in the early stages of the pandemic.

One New York Times article reports that although children’s overall emergency department visits fell during the early pandemic (much like overall reports to CPS), physicians report seeing an increase in accidental injuries from trampolines, scooters, bicycles, and inflatable pools because “parents [are] often unable to provide constant supervision because of work and other obligations” (O’Connor, 2020). Yet another piece details how parents who engaged in “intense hovering” pre-pandemic are now unable to watch their children while they work. Though this may allow children to explore and develop independence, it has also been associated with spikes in injuries (Caron, 2020). Data from social media accounts also suggest that children’s exposure to violence increased in the early stages of the stay-at-home restrictions (Babvey et al., 2020). Finally, perhaps most relevant to this study, more time at home during April and May 2020 was associated with more referrals of supervisory neglect in Georgia (Bullinger et al., 2021).

Related but distinct from employment pressures is the increase in parental stress due to the sudden concurrent and unprecedented obligations of providing childcare or supervision while working from home. Parallel accounts of children resisting new routines combined with added expectations from virtual or homeschooling increased parental burden and likely exacerbated parental fatigue. As early as April 2020 (Cooney, 2020, Manjoo, 2020), there were accounts of parents in many different situations feeling stressed, overwhelmed, and guilty for being alternately consumed by parenting responsibilities and trying to ignore them altogether. The prolonged parental burden leading to a range of responses from fatigue to (at the more extreme) parental burnout could explain the relatively higher rates of substantiated neglect cases (Griffith, 2020), especially among families that have experienced the largest increase in time at home. Simply put, parents may be worn down by the constant and elevated demands of childcare and need a respite from childcare responsibilities. With no alternate childcare provider, children are at risk for maltreatment. It is difficult to fully account for the differences in child neglect among groups who stayed home intensively and those that did not. Families who stayed at home the most were from comparatively advantaged areas. Although the difference in total time spent at home between families in quartile 1 and quartile 4 averaged approximately 2 h, it is possible that the transition to intensive staying at home was more pronounced for families in quartile 4 who may also have been more burdened with the need to balance work from home with childcare duties.

Early research from the medical literature also supports these claims. For example, the National Poison Data System has reported sharp increases in daily calls to poison centers regarding ingestions and inhalation in 2020 relative to the same period in 2019 (Chang et al., 2020). Calls increased the most regarding children aged 5 and under. Bram et al. (2020) show that while childhood injuries related to sports have decreased between March and April 2020 compared to earlier years, fractures and injuries occurring in the home, such as bicycle injuries, have increased. Finally, Chaiyachati et al., (2020) demonstrate an overall decrease in pediatric emergency department visits, but show an increase in high acuity trauma in 2020 relative to 2018 and 2019. Chaiyachati and colleagues also note that the demographics of patients seeking care is different. They note a smaller proportion of African American patients and observe a decrease in patients using public insurance (with a shift to privately insured patients). This demographic shift is consistent with our findings that children not traditionally at risk for child maltreatment are at heightened risk during the COVID-19 pandemic. This shift reveals that all families – regardless of traditional risk factors – need services and policies to promote child well-being. This may be especially relevant as school closures and daycare shortages persist or re-emerge.

Though our main result – that substantiated neglect in counties with residents experiencing the largest change in time at home is higher relative to counties experiencing less of a change in time at home – is consistent with the challenges of providing safe and consistent care to children while they work from home, a few points deserve further examination. Like emergency department visits, substantiation rates (and overall report rates) are down for all counties. These results suggest that families are struggling universally, but going without the traditional resources available to them. There are adverse consequences for children when there is a shock to family demands and there are insufficient resources, in the form of money or time, to provide safe and consistent care for children. In this sense, parents who were previously able to provide safe and consistent care with the aid of child care, schools, and other social supports may no longer be able to, or parents who previously had difficulty providing safe and consistent care may be further disadvantaged in the context of the pandemic. Of course, there remains a concern about which children are not being seen by CPS. If there are children not coming before CPS, this does not negate the findings of this paper, but it speaks to a need to continue to reach out to and serve all children in various communities.

Two points require further investigation. First, these data do not allow us to comment on the sources of reports of increased child neglect among families who stayed home intensively. Given that these families spent the most time at home they may not have come to the attention of CPS via traditional surveillance-related pathways. Similarly, the geographic variation in sources of child neglect reports require additional investigation. We find evidence of increased neglect in traditionally lower-risk areas. This may be a reflection of increased surveillance in these areas during the stay-at-home order, or it may be that the transition to staying at home was a greater transition for families in these areas.

Relatively more maltreated – specifically neglected – children may be an unintended and negative consequence of the public health response to the pandemic. Policymakers should increase outreach to families who may be at risk – including those who have not historically been considered at risk – and provide emergency social and economic supports (Bullinger et al., 2020). Healthcare professionals working in areas with strict social distancing may consider prolonged durations of time spent at home an additional risk factor for child maltreatment, and should enhance bilateral contact with children, youth, and families throughout the crisis to identify families who are at increased risk. Finally, the results of this study are important when considering the risks and benefits of resuming children’s activities such as schools, daycares, and camps.

The COVID-19 crisis represents a unique challenge for child maltreatment prevention. Targeted intervention for families with prior CPS contact as well as those traditionally considered at-risk may help to reduce child maltreatment. It may also be that aspects of the CARES Act and related legislation providing extended and enhanced unemployment insurance benefits, government cash transfers, and expanded access to paid family leave, may (even if temporarily) serve as a near-universal prevention strategy by increasing family income and relieving stress (Brown & De Cao, 2020).

Unfortunately, due to the ongoing COVID-19 pandemic and emerging variants, access to childcare and formal schooling remains uncertain or unstable for many families. This is particularly problematic for working parents who are forced to balance the competing demands of work and childcare, which may result in increased risk for child neglect. Social distancing is a necessary public health approach to curtail the spread of COVID-19 in communities, but it may still create new problems that affect broad swaths of the population. The pandemic has brought about new and persistent challenges for families. With evidence about the American family experience, policymakers and child serving agencies can craft meaningful policies and interventions to bring them much needed assistance.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Social distancing & substantiated child maltreatment in indiana, robustness checks

	Total substantiation rate
	Main model	Dropping March 2020	Limit sample to 2019–2020	Log (total rate)
Panel A: Total Substantiation Rate
Post*Quartile 2 (6.3 Hours More/Day)	1.66	1.59	2.51	0.06
	(1.61)	(1.60)	(1.63)	(0.13)
Post*Quartile 3 (6.7 Hours More/Day)	0.80	0.76	1.01	−0.05
	(1.36)	(1.37)	(1.16)	(0.10)
Post*Quartile 4 (7.3 Hours More/Day)	2.67*	2.64*	1.28	0.18
	(1.37)	(1.39)	(0.95)	(0.14)
Mean Y Pre-Covid	14.93	14.91	13.85	2.48
R2	0.56	0.56	0.67	0.68
N	8188	8096	1104	8019
Panel B: Total Report Rate
Post*Quartile 2 (6.3 Hours More/Day)	8.67	8.43	12.95*	0.06
	(5.57)	(5.45)	(7.41)	(0.05)
Post*Quartile 3 (6.7 Hours More/Day)	6.03*	5.85	10.20**	0.01
	(3.59)	(3.60)	(4.29)	(0.04)
Post*Quartile 4 (7.3 Hours More/Day)	25.15***	25.04***	23.56***	0.14***
	(3.93)	(3.95)	(4.04)	(0.04)
Mean Y Pre-Covid	104.37	104.05	110.46	4.54
R2	0.78	0.77	0.82	0.84
N	8188	8096	1104	8188
Notes: Data from Indiana DCS January 2013-May 2020. Post = 1 for April and May 2020. Outcome: Rates Per 10,000 Children. Models include log(adult population), county FE, and month-year FE, and are weighted by child population. Coefficient estimates represent changes relative to the comparison group: Quartile 1, counties that stayed home 5.3 h more/day. Robust standard errors are clustered at the county-level. *p < 0.10, **p < 0.05, ***p < 0.01