Dr. Juras will start by summarizing the original Bloom, Hill & Riccio (2003) conceptualization of how to estimate the impacts of variation in policy implementation.  Since this work, researchers have become more cognizant of the assumptions required for and limitations of such an analysis.  Bloom, Hill & Riccio (2003) had a large sample of individuals and sites, as well as the benefit of randomized experiments in each site (although not across sites).  Many research projects will not have these benefits but may still choose to use a multi-level analysis to estimate program impacts.  Juras will discuss the sample size requirement for similar analyses, especially the cases where data are purely observational, and will also extend the idea to the case of double random assignment (i.e., both sites and individuals randomly assigned).