Some methods on distributions for p-values and their applications
Chang Yu, PhD Vanderbilt University Medical Center
In our data analyses, we often encounter situations in which we have a set of p-values that we need to make inferences from. For example, one may have a handful of p-values that we need to combine to infer an effect size in a meta-analysis. Or, we have a large number of p-values from an omics project where we need to infer which ones are discoveries. To effectively analyze these special data, we need to know what distributions the p-values follow. Using distributions that we recently derived for them, we present two applications to demonstrate their utility. The first application is a meta-analysis with a sample of p-values. The Fisher's combination procedure provides a chi-squared test of whether the p-values were sampled from the null uniform distribution. After rejecting the null uniform hypothesis, we are still faced with the problem of how to combine the assembled p-values. We present the maximum likelihood estimate of the standardized mean difference from the p-values using our derived distributions. The second application is a microarray gene expression study in which we would like to assess how two multiple comparison procedures behave, also based on distributions for p-values.