Developing a spatial risk prediction model for child maltreatment

Abstract

While predicting child maltreatment risk at the household level is useful for allocating limited child welfare resources, significant privacy, data integration, data governance and legal hurdles make such an algorithm economically and politically difficult to put into production. In this project, we take a different approach to child maltreatment risk prediction, developing machine learning models that predict, not for a household but for a small spatial areal unit, such as the block. The only private health data required for this use case are geocoded maltreatment events. We present the results of a machine learning analysis in Richmond Virginia, including exploratory analysis, feature engineering, model development and validation. We then interpret our models in a resource allocation context.

Young adulthood is a time of transition which poses particular challenges for youth who are homeless or at risk of homelessness, including those exiting foster care. The instability of being homeless puts youth at greater risk of many poor outcomes. Connection to relevant housing resources and services are critical to ensure that young adults have the opportunity to succeed. Better aligning youths’ needs with relevant housing resources can help young adults become and remain stably housed, leading to better lifetime outcomes. This study presents a typology of young adults who exit foster care and residential programs for homeless young adults, including emergency shelters and transitional living programs. The study uses administrative data to follow a cohort of 8,795 young adults, including young parents and unaccompanied young adults from ages 18 through 21, who exited foster care or homeless services. Using sequence analysis, subsequent service use after exit, including utilization of homeless services, hospitals, jail, subsidized housing, and supportive housing, was used to build three-year trajectories of service use patterns of youth. These patterns were then grouped together based on similarity using cluster analysis to form six distinct groups of youth: (1) Minimal Service Use, (2) Later Homeless Experience, (3) Earlier Homeless Experience, (4) Consistent Subsidized Housing, (5) Consistent Supportive Housing, and (6) Frequent Jail Stays. Profiles were developed for each typology to comprehensively, but concisely, describe differences in the characteristics of each group of youth. Models were also developed to determine factors that were predictive of each typology. This typology is being used to inform prioritization processes for housing resources and to better understand how to target programs based on potential pathways of youth.