While the population average treatment effect has been the subject of extensive methods and applied research, less consideration has been given to the sample average treatment effect: the mean difference in the counterfactual outcomes for the study units. The sample parameter is easily interpretable and is arguably the most relevant when the study units are not representative of a greater population or when the exposure's impact is heterogeneous. Formally, the sample effect is not identifiable from the observed data distribution. Nonetheless, targeted maximum likelihood estimation (TMLE) can provide an asymptotically unbiased and efficient estimate of both the population and sample parameters. In this paper, we study the asymptotic and finite sample properties of the TMLE for the sample effect and provide a conservative variance estimator. In most settings, the sample parameter can be estimated more efficiently than the population parameter. Finite sample simulations illustrate the potential gains in precision and power from selecting the sample effect as the target of inference. As a motivating example, we discuss the Sustainable East Africa Research in Community Health (SEARCH) study, an ongoing cluster randomized trial for HIV prevention and treatment.



Included in

Biostatistics Commons