Please Do Not Cite or Circulate Without the Authors ’ Permission

The purpose of this study is to extend the potential-outcomes causal framework to encompass multi-level observational data. We handle the multiplicity of potential outcomes associated with each treatment for each unit by invoking the exchangeability assumption. We propose a joint multi-level design for addressing a complex policy issue that involves multiple sets of treatments and is relevant to multiple subpopulations. Using the national Early Childhood Longitudinal Study kindergarten cohort data, we apply the extended framework and adapt the propensity score-based methods to an evaluation of the kindergarten retention policy effects on children’s cognitive development. The results show no empirical evidence supporting the practice of retaining children in kindergarten