Sample Size Determination for High Dimensional Neuroimaging Studies Controlling False Discovery Rate

Two fundamental problems that we encounter in multiple comparisons are how to control the False Discovery Rate (FDR) and provide adequate power. Often, FDR is confused with type I error rate and type II error rate is confused with non-discovery rate (NDR). In classical hypothesis testing problem, the type I error rate is fixed at 5%, no such standard has yet been developed for FDR. The open door policy of choosing the FDR raises many questions related to what we should do now. In addition, the literature of type II error is quite vague in MC, however, the best suited may be the NDR at least for neuroimaging studies. Sample size determination to achieve the target power (80%), while controlling the FDR at a certain level is a complex problem in MC, as parameters in di fferent dimensions are involved. In this dissertation, I will discuss a unifi ed approach that deals with FDR, NDR and sample size determination. We choose an FDR approach best suited for our study from the ocean of FDRs, interlink it with NDR and study the impact of sample sizes on the power curve. Results will be illustrated with a real life data set from late-life depression.