December/January 2018Vol. 18, No. 9Predictive Risk Modeling in Context

Child maltreatment, both chronic and acute, is a growing concern within the field of child welfare. Of the 6.6 million children referred to child protective services (CPS) in the United States in 2014, 702,000 were found to be victims of abuse or neglect. However, an understanding of how to best screen for maltreatment and serve those affected by maltreatment is still a work in progress.

An article in Children and Youth Services Review assesses the strengths and weaknesses of current risk assessment tools used by CPS professionals. The article then describes how predictive analytics or predictive risk modeling (PRM) can ameliorate the shortcomings of these current approaches.

Currently, two general categories of tools have been developed to help standardize CPS risk and safety assessments. Theoretical tools, which are guided by a theoretical approach, look at child maltreatment identified by experts through clinical experience or research. The risk factors are then combined into an instrument or scale that CPS workers and others can use to help them gather information during assessments. Actuarial tools examine risk factors that are empirically related to maltreatment. These risk factors are then validated statistically. Although both tools have been adopted by CPS agencies, they carry limitations, such as being prone to operator error during application and interpretation as well as often containing subjective measures that require clinical judgment to score.

The article describes PRM as the application of data mining, modeling, and analytical techniques to existing data to find patterns and make predictions. As such, PRM can address the shortcomings of other assessment tools in the following ways:

• The vast amounts of data used in PRM can identify previously unobserved relationships between variables.
• PRM models are learning models that can adjust to new relationships within the data.
• PRM models can use existing data on the population for which it is being used.
• The PRM approach is more consistent because its variable selection is mathematical with no arbitrary selection of predictors.
• Unlike operator-driven assessment tools, which can lead to operator error, PRM does not rely on worker training and compliance.

The authors conclude that PRM, when used in combination with careful clinical practice, shows promise in the field of child protection as an effective and reliable risk assessment tool.

"Risk Assessment and Decision Making in Child Protective Services: Predictive Risk Modeling in Context," by Stephanie Cuccaro-Alamin, Regan Foust, Rhema Vaithianathan, and Emily Putnam-Hornstein (Children and Youth Services Review, 79), is available at http://www.sciencedirect.com/science/article/pii/S0190740917300452#!.