B1574
Title: Empirical likelihood for designed experiments
Authors: Eunseop Kim - Eli Lilly and Company (United States)
Steven MacEachern - The Ohio State University (United States) [presenting]
Mario Peruggia - The Ohio State University (United States)
Abstract: Experimental likelihood provides a framework that extends the use of likelihood from heavily structured parametric problems to those with minimal restrictions. The formulation of empirical likelihood allows one to focus inference on targeted quantities. We consider the application of the methods to the analysis of designed experiments. In this context, we address issues that arise due to blocking and multiple testing. Technical results identify the appropriate limiting distribution for a set of comparisons of interest. These same results suggest computational strategies that can be used for finite samples. The effectiveness of the method is demonstrated through simulation and analysis of an experiment on a commonly used pesticide. The method is shown to be robust to violations of the standard assumptions for designed experiments. The method extends to the linear mixed model.
