Improving Child Welfare Practice Through Predictive Risk Modeling: Lessons from the Field

Abstract

Predictive risk modeling (PRM) utilizes existing administrative data sources, including those integrated across multiple systems, to automatically generate risk scores for a range of adverse events. These methods are in contrast to other risk-based assessment strategies, such as actuarial approaches or Structured Decision Making methods commonly used in child welfare services. PRM tools draw on existing historic data to construct and validate latent estimates of risk on the population for which they are being used. PRM offers a number of advantages over traditional means of risk assessment in child welfare including the ability to leverage large, population-level data systems to improve predictive capacity, the ability to customize and validate risk assessment tools to specific groups and populations, and the ability to monitor predictive accuracy in real time. Given the reliance on administrative data, PRM is also limited by the need for the initial effort of cleaning data and building the model rather than an off-the shelf set of questions. This chapter presents a series of case studies demonstrating the application of PRM to a number of different contexts as a means of improving child welfare social work practice, supporting front-end child protection screening decisions, and identifying children at risk of child maltreatment based at birth to support access to preventive services.