Changes in the Risk Levels of Child Protection Reports during the COVID-19 Pandemic

BACKGROUND AND PURPOSE: During the COVID-19 pandemic, there was widespread concern that the reduced reporting of child abuse and neglect during stay-at-home orders meant that serious maltreatment would go undetected. Yet, given that roughly half of reports of abuse and neglect are screened out during normal years, there was also the possibility that drops in reports were largely concentrated among those that were distinctively low-risk and should never have been reported in the first place. In the current study, we use predictive risk modeling data from two jurisdictions – Douglas County Colorado and Allegheny County Pennsylvania – to examine changes in the risk distribution of reports of maltreatment during periods in which reports declined due to shut-down orders.

METHODS: Data were drawn from administrative records and associated risk levels generated through predictive risk models implemented in Douglas County and Allegheny County. The Douglas County Decision Aide is a decision support tool built from a predictive risk model that runs nightly and helps teams make screening decisions for maltreatment allegations that do not require an immediate response. Likewise, the Allegheny Family Screening Tool is built from a predictive risk model and is used as part of the hotline screening process. Both tools have been independently evaluated; additional details concerning classification accuracy and modeling methodology are publicly available. Weekly maltreatment reporting and screen-in rates for 2019 and 2020 were documented for each county, including by age, race / ethnicity, reporter type, and alleged maltreatment. A regression discontinuity design was used to assess the causal effects of COVID-19 closures on the risk distribution of children reported and screened-in for investigation.

RESULTS: Strong reporting cyclicality was observed in both jurisdictions, with seasonal reporting lows occurring during summer and other school breaks. With the onset of COVID-19 shut-downs, statistically significant declines (p<.001) in the number of reports were consistently observed in both counties, for all age groups, across maltreatment types, and for all racial / ethnic groups. Notably, findings indicate that for those reports that were received during the shut-down, the average risk score was significantly higher than during prior weeks in 2020 – or the same period during 2019. Associated screen-in rates also increased, aligning with the changed distribution of risk.

CONCLUSIONS AND IMPLICATIONS: There is little doubt that the implications of changes to CPS reporting patterns will be the subject of continued attention and study during the years to come. Initial data from the two counties examined suggest that declines in reports were greatest among lower risk referrals. Findings also reinforce opportunities to use predictive risk modeling applied to historical data to better understand changes in the risk profile of maltreatment reports and the consistency with which systems respond to those reports.