Using Multilevel Survival Models to Assess Individual and County Contextual Factors Related to Time to Achieve Permanency

Background and Purpose: Children served by the child welfare system are frequently nested within larger groups that could influence their experiences and achievement of outcomes. When pursuing child welfare research, it is important to assess the extent and need to account for nested data when child observations are not independent, such as when children are from the same family or served by the same local children welfare agency. An individual's experience and outcomes in foster care may be impacted not only by their individual circumstances but by shared contextual factors such as family parenting practices, or common policies or practice within a county agency. This research demonstrates one methodological approach for multilevel survival analysis, which accounts for the nested nature of data with continuous-time longitudinal analysis. In particular, corrective Cox proportional hazard models used to assess how child and county contextual factors related to the rates of achieving permanency.

Methods: The study uses longitudinal administrative child welfare data from over 22,000 children from a state with a county-administered child welfare system. A corrected Cox proportional hazard model is used to account for the correlation of children nested within county child welfare agencies. Specifically, the LWA marginal model (Lee, Wei & Amato, 1992) is used to assess the main effects and cross-level interactions associated with time to reunification, adoption, and guardianship/custody. To address the violation of independent observations, the LWA model estimates marginal distributions of the distinct failure times to produce a robust and optimal estimation of the variance-covariance matrix, which is then used in the statistical calculation to correct for biases in standard errors and estimate parameters. The analysis is used to assess how individual characteristics and contextual factors of agency characteristics and county demographics relate to the rate of achieving permanency outcomes of reunification, adoption, and guardianship/custody.

Results: Study results demonstrated that multiple child, agency, and county factors were related to the rate in which children achieved permanency outcomes. In particular, child age, gender, race, ethnicity, and reason for placement into foster care were all significantly related to the speed of achieving permanency. County agency characteristics, specifically caseload size, use of relative placements, agency engagement in alternative response, and agency history of implementing reform efforts, as well as county demographics of poverty and unemployment were significantly related to timely achievement of several permanency outcomes. A comparison of results from corrective and traditional Cox proportional hazard models demonstrated consistent parameter estimates yet showed great differences in levels of significance.

Conclusions and Implications: These findings demonstrate the need to assess achievement of child welfare permanency outcomes within a multilevel context. The results show how county demographics and agency factors in conjunction with child characteristics play a role in the time to achieve permanency. Furthermore, this study provides evidence for the use of appropriate multilevel survival methods to account for nested child welfare data when assessing factors related to the time to achieve permanency.