Experimental Design based on Compound Optimization Criteria: Essays on Design Construction, Analysis, and Applications

Abstract

This thesis focuses on the construction of experimental designs having multiple objectives. It explores theoretical and practical applications of compound design construction and analysis in the context of problems involving statistical and practical optimization criteria. Different essays of this dissertation focus on linear or non-linear models characterizing different experimental settings. The overarching theme of the separate essays is the idea of constructing experimental designs for initial exploration of a given phenomenon - such as in the case of screening experiments - which are created with goals that are not only shortsightedly related to the initial exploratory phase of the experimentation, but that account for the potential subsequent goals of further phases. In this fashion, screening experiments are created with the double goal of initial factor screenings, but also protection from bias induced by higher-order terms not included in the initial model. This idea is applied to both linear and non-linear experimental designs construction and involved in the analysis of such types of designs when doing variables selection. Similarly, in the applied context of two-round procurement auctions, the experimental design created for bidding strategies on the first round of the auction accounts for the impact on profit maximization related to the bidding strategies involved in the second round.