Date of Award

Spring 1-1-2015

Document Type


Degree Name

Doctor of Philosophy (PhD)


Psychology & Neuroscience

First Advisor

Leaf Van Boven

Second Advisor

Charles Judd

Third Advisor

Gary McClelland

Fourth Advisor

Matt Jones

Fifth Advisor

John Lynch


Research involving samples of participants responding to samples of stimuli is ubiquitous throughout psychology and neuroscience. This dissertation is concerned with the optimal design of such experiments, and more generally with experiments that involve multiple random factors (i.e., factors whose observed levels are a random, exchangeable sample of all possible levels of theoretical interest) beyond the conventional random participant factor. I first consider experiments with two crossed random factors--participants and stimuli--and a single fixed Condition factor with two levels. I describe procedures for statistical power analysis in such experiments and demonstrate how these procedures can be used to answer difficult questions of optimal design in the presence of stimulus sampling. Next I discuss the implications of these power analysis results for how direct replications ought to be conducted for studies that involve stimulus sampling. Finally I discuss power analysis for general ANOVA designs that can consist of any number of fixed or random factors, each with any number of levels, and with any valid pattern of nesting or crossing of the factors. I present a web application I have developed (PANGEA: Power ANalysis for GEneral Anova designs), discuss the statistical theory underlying the application, and illustrate how it can be used to answer questions of optimal design in a very general experimental setting.