Small Area Estimates of the Average Treatment Effect with Applications to Generalizability

Policymakers are increasingly interested in the extent to which experimental results generalize from a sample to a population of inference. When the sample is not randomly selected, propensity score methods are used to reweight the sample. Subclassification by propensity score is commonly used in which the population is partitioned into strata using the propensity score distribution. However, subclassification methods are limited whenever the sample size is small or when there are problems with distributional overlap in the propensity scores. This study extends small area estimation methods to the problem of "sparse" strata under subclassification. The performance of two small area estimators is compared with two design-based estimators in a simulation study based on a completed cluster randomized trial.