Cluster allocation design networks

Abstract

When planning and designing a policy intervention and evaluation, it is important to dierentiate between (future) policy interventions we want to evaluate, FT, aecting "the world", and experimental allocations, AT, a ecting "our picture of the world". The policy maker usually has to dene a strategy that involves policy assignment and recording mechanisms that will a ect the (condi-tional independence) structure of the data available. Causal inference is sensitive to the specication of these mechanisms. In uence diagrams have been used for causal reasoning within a Bayesian decision-theoretic framework that introduces interventions as decision nodes (Dawid 2002). Design Networks expand this frame-work by including experimental design decision nodes (Madrigal and Smith 2004). They provide semantics to discuss how a design decision strategy (such as a clus-ter randomised study) might assist the identication of intervention causal e ects. The Design Network framework is extended to Cluster Allocation. It is used to assess identiability when the experimental unit's level is di erent from the analy-sis unit's level, and to discuss the evaluation of clusterand individual-level future policies. Cases of 'pure' cluster (all individuals in a cluster receiving the same intervention) and 'non-pure' cluster (only a subset receiving the policy) are dis-cussed in terms of causal e ects. The representation and analysis of a simplied version of a Mexican social policy programme to alleviate poverty (Progresa) is performed as an illustration of the use of Bayesian hierarchical models to make causal inferences relating to household and community level interventions. © 2007 International Society for Bayesian Analysis.