Time dependence and unmeasured heterogeneity in administrative data analysis: Application to adoption photolisting services data

Administrative data provide an excellent source of information about policy events and interventions. Use of administrative data presents challenges in estimation, however, due to incomplete spells, time dependency, and unobserved heterogeneity. Statistical models are needed to help discern these effects over time. To address the statistical problems inherent in administrative data this paper proposes two alternative empirical approaches, hazard model and matched-sample analysis. The hazard model provides a multivariate framework within which predictors of an event or length of spell can be examined. Matched-sample analysis allows researchers to focus on the impact of particular policy intervention strategies of interest using matched sets of individuals with identical measured characteristics and elapsed observation time. The approach allows for the most precise statistical controls possible and eliminates the problems posed by time dependency. To demonstrate both the appropriateness and usefulness of these models in addressing policy questions of interest an application example in the area of child welfare is used. Implications of study findings for administrative data analysis are discussed.