Risk Assessment and Ongoing Involvement with Child Protective Services

Background/Purpose: Over the past decade the debate concerning the utility of child welfare risk assessment models has gained momentum. There is a need to advance knowledge about the validity of actuarial and clinical risk assessments used in child protective services (CPS).This study examines assessment relationships with ongoing CPS involvement.

Study Objectives: Objectives of the current study are 1) to determine the relationship between safety and risk assessment for families involved with CPS,  2) to determine the relationship between risk and ongoing involvement with the child protective service system , 3) to determine the relationship between ongoing involvement and recurrence of maltreatment at 6 and 12 months, and 4) to explore the predictive ability of actuarial risk assessment compared to subjective (i.e., caseworker/clinician informed) family risk assessment.

Participants: Data included cases from 1417 families involved with CPS from one midwestern state.

Procedures: This study is a secondary analysis of data that were gathered as part of a large risk assessment evaluation titled, The Comprehensive Assessment and Planning Model—Interim Solution (CAPMIS). The original study had the following primary goals:  implementation of the CAPMIS model, reliability of the use of the assessment tools, validity of the tools, and the relationship between needs, services, and outcomes.

Measures: The CAPMIS model was developed to provide a structured, reliable, and valid set of tools for assessing safety and risk and for working with families who have come to the attention of the county child welfare agency. The current study uses the CAPMIS Safety Assessment, the CAPMIS Family Assessment, and the Actuarial Risk Assessment.

Analytical Technique: Preliminary analysis involved a comparison of risk scores and recurrence. We also conducted preliminary logistic regressions to explore the relationship between safety factors, risk contributors, and recurrence of child maltreatment over time. Primary analysis with Mplus 5.21 involved simultaneous path modeling with count and binary variables with Monte Carlo integration.  Models are run in consideration of Poisson distribution of count variables; thus, chi-square and related fit statistics are not available. Akaike and Bayesian Information Criteria are used to compare competing risk assessment models.

Results: With respect to the first objective, we found that safety was a significant predictor for subjective child, adult, and family level risk as well as actuarial risk level- intense (p<.000). There was an inverse and significant relationship between safety and low risk (p<.000). For the second objective, we found that subjectively defined adult risk and family risk significantly predicted ongoing involvement (p<.001 and p<.05 respectively). Actuarial low and moderate risk inversely predicted ongoing involvement (p<.001). Intense risk significantly predicted ongoing involvement (p<.05).  Relative to objective 3, we found no relationship between ongoing involvement and recurrence of maltreatment. Lastly, we found that the actuarial model is a better fitting model than the subjectively defined risk assessment model.

Implications: Results indicated predictive value in both types of risk assessment. However, practitioners are encouraged to be critical about assessment utility.  While generalizability is limited, this study adds to the growing body of evidence about risk assessment validity.