Session: WITHDRAWN Foster Youth at Risk of (Im)Permanence: The Possibilities of Predictive Risk Modeling at the Back End of Child Welfare (Society for Social Work and Research 26th Annual Conference - Social Work Science for Racial, Social, and Political Justice)

Background. Child welfare policy and practice are increasingly attending to the needs of youth who do not experience intended outcomes of reunification or adoption. As many as 24,000 youth had a case plan goal of long-term foster care or emancipation in FY 2019. There are good reasons to be concerned about this group. Three decades of research describes challenging prospects in adulthood. Though independent living policy expanded safety nets, youth without permanence may have high levels of exposure to childhood adversities - before and during foster care, that erode development without trauma-specific intervention(s). The social work research community is in a unique position to offset risks. Social work professionals have long recognized the roles of childhood adversities, oppressive institutions, and unjust practices in the production and maintenance of disparities, but trauma remains neglected in child welfare. When the trauma that families bring to systems and transmit to their children is ignored or made worse in systems, there is a potential for greater harm. This symposium challenges the field to think about possibilities. Could advances made in analytical approaches be used to identify children at risk of not achieving permanence? If so, at what age could reliable predictions be made? If children at risk can be identified, what interventions have the power to strengthen healthy child and family development?

Methods: This collection of studies uses data from two states with large child welfare populations. The first presentation offers an orientation to predictive risk modeling - a set of tools that identify which variables can best predict the outcome that the model is trained on and create a predictive model that results in a risk score for each case. The second presentation explores whether machine learning methods – making predictions for newly observed cases by learning the patterns found in the relationships between features and outcomes in existing data – can be an aid to identify youth at risk of exiting foster care without permanency. The last two presentations are part of a mixed methods study in Illinois, which present findings from age-specific predictive analytics and life story interviews.

Results: Using child welfare administrative data in California to identify foster youth between ages 13-18 with referrals, the first presentation showcases innovative ways to leverage risk scores at multiple decisions points. The second study suggests that for youth in foster care in California between ages 12 and 14, a simple model can be useful for early identification of risk. The findings from the analysis of children born between 1994-1998 with at least one entry in Illinois indicate that acceptable predictions of risk could be made as early as age 2. The findings from life story interviews indicate that addressing trauma histories may be critical to strengthening developmental functioning.

Discussion: This collection of studies asks questions about how data harnessed for administrative purposes can be analyzed to assist with decisions in child welfare.