For binary experimental data, we discuss randomization-based inferentialprocedures that do not need to invoke any modeling assumptions. We alsointroduce methods for likelihood and Bayesian inference based solely on thephysical randomization without any hypothetical super population assumptionsabout the potential outcomes. These estimators have some properties superior tomoment-based ones such as only giving estimates in regions of feasible support.Due to the lack of identification of the causal model, we also propose asensitivity analysis approach which allows for the characterization of theimpact of the association between the potential outcomes on statisticalinference.
展开▼