A0286
Title: Collaborative design of controlled experiments in the presence of subject covariates
Authors: Qiong Zhang - Clemson University (United States) [presenting]
Abstract: In some cases, researchers may run multiple, separate, controlled experiments where subjects participate in more than one experiment. Due to subjects participating in multiple experiments, there is a correlation among the responses across experiments. Taking account of the correlation across experiments, a recent study proposed the collaborative analysis framework and demonstrated that their framework can provide more precise estimates of treatment effects than if one were to analyze the experiments separately. The experimental design problem of allocating subjects to treatment or control within each of the multiple experiments when subject covariate information is available is considered. The goal of the allocation is to provide precise estimates of treatment effects for each experiment to further improve precision gained through collaborative analysis. Using D-optimality as the allocation criterion, semi-definite programming-based randomized algorithms are proposed, which provide solutions to the D-optimality problem. The performance of the algorithms is showcased in a simulation study, demonstrating their effectiveness over pure randomization methods when the number of subject covariates is large.
