Child maltreatment is a major national concern. Over 1 in 22 U.S. children are reported to and investigated by child protective services (CPS) for child maltreatment concerns each year (DHHS, 2021). An estimated 1 in 3 U.S. children will be investigated by CPS by age 18 (Kim et al., 2017). Child maltreatment has been linked to myriad biological, psychological, and social problems, including premature death (Anda et al., 2006; Cabrera et al., 2020; Lansford et al., 2002; Vaithianathan et al., 2018). A recent estimate suggests that the societal lifetime cost of all investigated child maltreatment reports (CMRs) in a single year in the United States is $2 trillion (Peterson et al., 2018). Understanding community risk and protective factors is important to inform prevention efforts (Jones Harden et al., 2020). Prior research has heavily focused on community risk factors (Coulton et al., 2007; Freisthler et al., 2006; Maguire-Jack, 2014a). Yet, research on community protective factors is sparse. Research on community-level provisions of evidence-based home visiting (EHV) in particular has been absent, even though EHV has been the most prominent approach for child maltreatment prevention. Most studies of the impacts of EHV have been individual-level randomized controlled trials. From a public health and program evaluation standpoint, however, examining impacts on a whole community offers the opportunity to better assess the value of these programs with regard to their effectiveness in real-world community settings and the possibility of spillover effects through a variety of mechanisms. To address this knowledge gap, this study examines the county-level relationship between EHV provisions and CMR rates, using U.S. national data from 2016-2018.

Table 1 reports the descriptive statistics. Each year, on average 61.3 per 1,000 children in a county had a CMR. CMR rates were similar across sex but differed by age and race/ethnicity. Younger children had higher rates. Black children had higher rates compared with White and Latino children. Neglect was the most frequent type, followed by physical abuse and sexual abuse. A relatively small number of counties showed changes in any EHV provisions from Year to Year+1 (“2016 to 2017” or “2017 to 2018”). Among pairs of Year and Year+1 observations, most had provision in both years or no provision in both years (83.4%, and 11.7%, respectively). The percentage that had no provision in Year but provision in Year+1 was 3.3%, and 1.6% had provision in Year but no provision in Year+1. Similarly, changes in MIECHV-funded EHV provisions were not frequent. Most showed consistent no provision (50.9%) or consistent provision (46.8%) in both Year and Year+1. Only 1.4% showed a change from no provision to provision, and 0.8% showed a change from provision to no provision in Year and Year+1.

Table 2 reports the model results for overall CMR rates. The within-effect (i.e., longitudinal change) of EHV provision had a significant negative association. That is, when the EHV provision increased from 0 (absence) to 1 (presence) longitudinally, report rates significantly decreased by 2.13 per 1,000 children with no control (the unadjusted model) and 2.21 per 1,000 children with controls (the adjusted model). The within-effect of EHV provision did not significantly differ by urbanicity (Table S2 in the Supplement). We further assessed state and county random slopes of the within-effect of EHV provision and found that both random slopes were not significant (results not shown). This indicated that the within-effect of EHV provision did not significantly differ by state or county.

The between-effect (i.e., inter-county difference) of EHV provision was not significant in both unadjusted and adjusted models (Table 2). That is, there was no significant difference in report rates between counties with and without EHV provisions. This association did not significantly differ by urbanicity (Table S2) or state (results not shown).

The within-effect of MIECHV-funded EHV provision had no significant association in both unadjusted and adjusted models (Table 2). This association did not significantly vary by urbanicity (Table S2), county, or state (results not shown).

The between-effect of MIECHV-funded EHV provision was significant with no control (the unadjusted model in Table 2). That is, with no control, report rates were significantly higher among counties with MIECHV-funded EHV provision than among counties with no such provision. With controls, this association was not significant (the adjusted model in Table 2) and did not significantly differ by urbanicity (Table S2) or state (results not shown).

Table 3 presents the results of subgroup-specific analysis with all controls. The within-effect of EHV provision had a significant negative association with CMR rates among children aged 0-5 and among children aged 6-11, but not among children aged 12-17. After beginning EHV programs, report rates decreased by 2.88 per 1,000 children aged 0-5 and 2.59 per 1,000 children aged 6-11. The within-effect of EHV provision was also significantly associated with decreasing sex-specific report rates. After introducing EHV provisions, report rates lowered by 2.13 per 1,000 male children and 2.24 per 1,000 female children. The within-effect of EHV provision had no significant association with other subgroup-specific report rates. The between-effect of EHV provision, the within-effect of MIECHV-funded EHV provision, and the between-effect of MIECHV-funded EHV provision were not significant for any of the subgroups.

Figure 1 visually assesses the within-effect of EHV provision. It depicts changes in CMR rates by the following four longitudinal trends of EHV provisions from “Year” to “Year+1” (i.e., 2016-to-2017 or 2017-to-2018): absence-to-absence, absence-to-presence, presence-to-absence, and presence-to-presence. Among 1,184 pairs of “Year” and “Year+1” observations, relatively small numbers of pairs showed the absence-to-presence trend and the presence-to-absence trend of any EHV provisions (n = 39 and 19, respectively) and MIECHV-funded EHV provisions (n = 17 and 10, respectively). Both the unadjusted means (i.e., simple arithmetic means) and adjusted means (i.e., predicted values by the adjusted models in Table 2) of CMR rates decreased with the increase of any EHV provisions (i.e., absence-to-presence) and increased with the decrease of any EHV provisions (i.e., presence-to-absence). On the other hand, for all longitudinal trends of MIECHV-funded EHV provisions, the unadjusted means of CMR rates increased over time and the adjusted means of CMR rates showed no significant change over time. It is worth to note that for the adjusted means of CMR rates, the absence-to-absence and presence-to-presence prediction lines are flat and overlapped to each other because they are statistically held constant while allowing only the within-effect of EHV provision to vary for prediction lines.

We conducted sensitivity analysis using negative binomial models because some less frequent dependent variables, such as Black, Latino, physical abuse, and sexual abuse report rates, were slightly positively skewed (skewness = 1.3 to 1.5). We maintained our linear model findings as they were consistent with those of negative binomial models. Table S3 in the Supplement reports the coefficients and confidence intervals of the control variables for those interested in the relationships between the controls variables and CMR rates.

To our knowledge, this work is the first effort to examine the county-level impacts of EHV provisions on CMR rates. We found that longitudinal changes (i.e., within-effects) of any EHV provisions were significantly associated with CMR rates in U.S. county-level data from 2016 to 2018 when a range of controls were considered. That is, after starting any EHV provisions in communities, their CMR rates decreased. After stopping provisions in communities, their CMR rates increased. This association did not significantly differ by urbanicity, county, or state. This association was significant overall and within male, female, age 0-5, and age 6-11 subgroups, but not significant for other subgroups. The between-effect of any EHV provision and both the within- and between-effects of MIECHV-funded EHV provision were not significant overall or within any of the subgroups.

Within-effects can capture only short-term effects (within 1 year for this study). Some may wonder whether EHV provisions can change county CMR rates within such a short timeframe. However, studies do exist that identify changes in community processes (e.g., collective efficacy) immediately after a few months’ training or therapy, as well as after several months follow-up (e.g., Aguilar-Raab et al., 2018; Ohmer, 2016). In addition, intervention effects are generally well identified during and immediately after the intervention. It is also common that intervention effects fade out over time during the post-intervention period (Bailey et al., 2017; van Aar et al., 2017). Therefore, it is possible that community changes occur in a relatively short period of time after changes in EHV provisions.

The significant within-effects of any EHV provisions for children aged 0-5 is expected as the EHV programs target this age group. Yet, the significant finding among children aged 6-11 is a somewhat unexpected, exciting finding. Several speculations are possible for this finding. First, the social processes discussed in the Introduction section might be protective of child maltreatment risk for children in both age groups. Second, the finding can be due to the presence of older siblings (aged 6-11) in the same household of EHV-served younger siblings (aged 0-5). EHV programs are often not limited to first-time parents. The majority of families with a child(ren) have more than one child (Pew Research Center, 2015), and the median interpregnancy interval is 2-2.5 years (Copen et al., 2015). EHV-served families with younger children (aged 0-5) therefore may also have older children (aged 6-11), and thus participation may lower CMR rates among children aged 6-11. Finally, interfamilial interactions between children aged 0-5 and aged 6-11 may be common occurrences in a community (Ellis et al., 1981), which may facilitate spillover effects of EHV programs. While further research is needed to draw implications from this exciting finding, it provides suggestive evidence that EHV provisions have impacts on individuals in a community beyond service participants.

These significant findings on the within-effects of any EHV provisions have several important implications. Within-effects are methodologically equivalent to fixed effects, which can control for unobserved inter-county heterogeneity. The within-effect findings therefore have greater implications for causality than cross-sectional findings. This study’s findings also make a contribution to the existing evidence base. Prior research has mostly focused on individual-level impacts of EHV (Avellar & Supplee, 2013; Curry et al., 2018; Levey et al., 2017; MacMillan et al., 2009), and research has been sparse on community-level impacts. The within-effect findings help fill this knowledge gap by confirming the EHV-CMR relationship at the community level. Another implication has to do with prevention efforts. Public health approaches, especially community-based strategies, have become a vital part of prevention efforts for social and public health problems (Graaf & Ratliff, 2018; Merzel & D’Afflitti, 2003), including child maltreatment (Lo & Cho, 2021). This study provides evidence supporting potential benefits of integrating EHV provisions into community-based strategies to reduce child maltreatment incidents and reports at the whole community level.

Some aspects of the within-effects findings need further discussion to understand their limitations and where we can go from here. The first limitation is that the effect sizes we identified are small. This is not surprising for two reasons: 1) Each year, the EHV programs serve only about 1% of young U.S. children (NHVRC, 2019). Considering this low provision, small effect sizes are expected and likely to grow if provision is increased. 2) High child maltreatment rates in the United States are closely related to a range of formidable risk factors (e.g., poverty, crime, mental health, and substance abuse problems). It may be impractical to address them solely by EHV services, which are low-intensity and focus largely on individual-level behavioral changes (Finello et al., 2016; Gomby et al., 1999). Despite this limitation, increasing community EHV provision may be worthwhile given the enormous social and economic burden of child maltreatment, including for example costs of child welfare, health care, criminal justice, special education, productivity losses, and quality of life losses (Peterson et al., 2018). Yet, it may be naïve to hope EHV can be a cure-all. As others have noted, communities should adopt EHV as a part of a comprehensive response to child maltreatment (Fernandes-Alcantara, 2018; Finello et al., 2016; Gomby et al., 1999). The second limitation is that we could not examine long-term mechanisms of the community-level relationship between EHV and CMR because the data were only available for 3 years, from 2016 to 2018. It is entirely possible that some community changes that EHV may promote (e.g., establishing social ties) take a long period of time (Sampson et al., 1999). Home visiting also may have long-term effects (e.g., Olds et al., 1998). More research with a longer timeframe is required to understand how fast community-level changes can occur, how they are accumulated over time, and how long they last. Another limitation is that a relatively small number of counties have shown longitudinal changes in EHV provisions in our data. Although this is simply what has occurred among U.S. counties, this limits the generalizability of our findings. Further research is required to confirm our findings. Given the limited but best-available data, nevertheless, our findings have implications for both counties that have EHV provisions and those that do not. The first group is larger than the second, and our findings support maintaining EHV provision. While a smaller number of counties do not have EHV provisions, they are still a substantial group, and many low-populated counties that were excluded from analysis due to confidentiality restrictions likely also do not have EHV programs, given most are rural. Preventive services are generally less available in rural areas than urban areas (Maguire-Jack & Kim, 2021). In line with this, the overall levels of provisions in our data were higher than the known coverages (about 50% for any EHV and about 33% for MIECHV-funded EHV). Our findings suggest that launching EHV provisions could protect a considerable number of families.

The within-effect of any EHV provision was not significant in any subgroup of race/ethnicity and maltreatment type. First, it was not significant for any of White, Black, and Latino CMR rates. This was perhaps due to our use of the overall measure of EHV provisions. Even if services are available in a community, accessibility and utilization of services can be widely different between racial/ethnic groups (Allard et al., 2003). Race/ethnicity-specific measures of EHV provisions may be required to identify the potential impacts of EHV provisions on race/ethnicity-specific CMR rates. Second, the within-effect of any EHV provision was also not significant for neglect, physical abuse, or sexual abuse report rates. It is possible, for example that EHV services focus more on neglect in some communities but physical abuse in other communities, based upon each community’s specific needs. This would explain why the impact of EHV provision is less consistent for type-specific CMR rates than for overall CMR rates across communities. Given the small effect size for overall CMR rates, the impact of EHV provision on type-specific CMR rates may be subtle and thus difficult to detect.

We found that the between-effect of any EHV provision was not significant overall or within any of subgroups. A possible reason is that between-effects are methodologically less rigorous than within-effects. Between-effects are therefore vulnerable to inter-county heterogeneity, especially uncontrolled service needs. We controlled for a range of known community risk factors such as socioeconomic (median household income and % single-parent), demographic (% Black, % Latino, and % foreign-born), residential instability (% moved), and care burden factors (% children among residents and % children with disabilities). However, other important risk factors, such as mental health, substance abuse, and crime, were not adjusted for. Concentration of such problems can increase service needs, as well as child maltreatment incidents and reports in communities. EHV services might be provided more often to communities with higher service needs and thus with higher CMR rates. This might offset the protective between-effect of any EHV provision. A more rigorous approach, such as randomization of communities (e.g., Prinz, 2009), may help address the potential heterogeneity in service needs.

Our findings indicate that neither the within- nor the between-effects of MIECHV-funded EHV provisions were significant. We offer several possible explanations for this with no definitive answer. First, the between-effect of MIECHV-funded EHV provision was not significant perhaps because of the abovementioned heterogeneity in service needs. MIECHV also requires states to provide EHV services in at-risk communities with high rates of adversities, including child maltreatment, based on a statewide needs assessment (Fernandes-Alcantara, 2018). This legislative requirement might mean that MIECHV-funded EHV provisions are more concentrated in high risk communities than EHV provisions funded by other sources. Second, MIECHV appears to have reached maturity prior to the study years in that the program did not grow (DHHS, 2020), and indeed in our data, fewer counties showed longitudinal changes in MIECHV-funded EHV provisions compared with any EHV provisions. This lack of longitudinal change might lead to the non-significant within-effect of MIECHV-funded EHV provision. Third, the MIECHV funding is a relatively small part of the entire funding for EHV programs. Among the families served by EHV programs in 2016-2018, slightly over one quarter were served through MIECHV (NHVRC, 2019). The impacts of EHV provisions might not be identifiable when the focus was limited to such a small subset (i.e., MIECHV-funded EHV provisions). Given that both MIECHV and other funding sources support the same EHV programs, we do not view the non-significant within-effect of MIECHV-funded EHV provisions as evidence that EHV does not affect communities. On the contrary, the significant within-effects of any EHV provisions suggest that expansion of MIECHV is likely to increase overall EHV provisions and thus may have benefits in child maltreatment prevention.

This study found significant community-level associations between longitudinal changes (i.e., within-effects) of EHV provisions and CMR rates among U.S. counties from 2016 to 2018. These findings make two principal contributions to the existing evidence base. First, it expands our understanding of the EHV-CMR relationship from the individual level to the community level. Second, it expands the scope of data from a single-state setting to many counties across most states, while generalizability of our data is still limited for many low-populated (mostly rural) counties. For policy and practice, our findings suggest that expanding EHV in all counties is a promising strategy to further reduce CMR rates at the community level. Considering that MIECHV-funded EHV provisions are a small subset of all EHV provisions, community-level impacts of MIECHV should be evaluated within the context of overall EHV provisions. More research is needed to confirm our findings, examine longer-term impacts, establish causality, and support generalization to low-populated, rural areas.