Alphabetized Optimality in Experimental Design by Isabelle Ahn ’19, Wednesday, February 13, Statistics Colloquium, 1:10 – 1:50 pm, Stetson Court Classroom 105
Abstract: Contrary to popular belief, experiments are not just used in science labs. Everything from your allergy medication to the email marketing you receive goes through some level of experimental testing. A full factorial design, one that tests all possible combinations of factors, often requires hundred trials, which is both a costly and lengthy process. Thus, researchers often seek to limit the number of trials performed to reduce the costs of experimentation. While this risks information loss, optimal designs allow full statistical models to be estimated with fewer experimental runs. Optimal designs determine the combination of factors that should be run according to the optimal value of a specific statistical criterion. This talk will focus on Alphabetized Optimality, a popular sect of optimality criteria that aims to minimize the variance of estimators, corresponding the maximizing the information matrix. I will first introduce properties of experimental design and then explain each criterion and the differences between them with examples in the simple regression case.