|Friday, February 19|
|PS2 Poster Session 2 & Refreshments||
Fri, Feb 19, 5:15 PM - 6:30 PM
Multiple Independent Hypothesis Testing of Discrete Data (303228)Naomi S. Altman, Penn State University
*Stefanie Rose Austin, Penn State University
Isaac Dialsingh, University of the West Indies
Keywords: Hypothesis testing, multiple testing, false discovery rate, FDR, discrete, null hypotheses, p-values
Multiple hypothesis testing remains an important area as data sets containing a high number of variables continue to surface in many fields. P-values of continuous test statistics are uniformly-distributed, but the p-values of discrete test statistics come from a discrete distribution with finite support and a null distribution that may depend on an ancillary statistic that varies among the test statistics. This work addresses the latter case, namely estimating the number of true null hypotheses (pi0) and the control of the false discovery rate (FDR). A simulation study was conducted to compare pi0 estimators and FDR procedures. We generated RNA-seq and SNP data, varying the values of pi0 and of m, the number of total simultaneous tests. Eight methods for estimating pi0, including a new regression estimator, and eleven FDR-controlling procedures were considered. We identified several useful pi0 estimators, including those developed for continuous test statistics, and found that adaptive FDR methods suggest traditional procedures can be improved by using an estimate of pi0 rather than m in the algorithms. Filtering out zero-powered tests shows also to be a useful technique.