Forward screening and post-screening inference in factorial designs

Ever since the seminal work of R. A. Fisher and F. Yates, factorial designs have been an important experimental tool to simultaneously estimate the treatment effects of multiple factors. In factorial designs, the number of treatment levels may grow exponentially with the number of factors, which motivates the forward screening strategy based on the sparsity, hierarchy, and heredity principles for factorial effects. Although this strategy is intuitive and has been widely used in practice, its rigorous statistical theory has not been formally established. To fill this gap, we establish design-based theory for forward factor screening in factorial designs based on the potential outcome framework. We not only prove its consistency property but also discuss statistical inference after factor screening. In particular, with perfect screening, we quantify the advantages of forward screening based on asymptotic efficiency gain in estimating factorial effects. With imperfect screening in higher-order interactions, we propose two novel strategies and investigate their impact on subsequent inference. Our formulation differs from the existing literature on variable selection and post-selection inference because our theory is based solely on the physical randomization of the factorial design and does not rely on a correctly-specified outcome model.