ST05 FDR_TEST: A SAS(R) Macro for Calculating New Methods of Error Control in Multiple Hypothesis Testing     Contributed

Dr. Jeffrey D. Kromrey University of South Florida ,Kristine Y. Hogarty
University of South Florida
Abstract: The testing of multiple null hypotheses in a single study is a common occurrence in applied research. The problem of Type I error inflation or probability pyramiding in such contexts has been well-documented for many years. General procedures for the contr ol of Type I error rates in multiple testing are the Bonferroni procedure and its' more recent modifications.These procedures partition a desired level of Familywise error across the set of hypotheses being tested. Recent work on multiple testing by Benjam ini and Hochberg (1993, 1995, 2000) has focused on controlling the False Discovery Rate (FDR) rather than rates of Type I error. Both the adaptive (BH-A) and non-adaptive (BH) procedures for controlling the FDR in a set of tests promise increases in statis tical power relative to other procedures. This paper presents a SAS macro that calculates probabilities under five decision rules that may be employed in multiple testing contexts (a per-hypothesis rule, Bonferroni, Hochberg, BH and BH-A). The macro evalua tes a set of probabilities that are supplied as an input and outputs the results of the five decision rules. The paper provides a demonstration of the SAS/IML code and examples of applications in simulation studies.

Biography:
Jeffrey D. Kromrey is a Professor in the Department of Educational Measurement and Research at the University of South Florida. His specializations are applied statistics and data analysis. His work has been published in Communications in Statistics, Educa tional and Psychological Measurement, Journal of Experimental Education, Multivariate Behavioral Research, Journal of Educational Measurement and Educational Researcher. He has been a SAS programmer for 15 years and uses SAS for simulation studies as well as for applied data analysis.

Kristine Y. Hogarty is a doctoral candidate in the Department of Educational Measurement and Research at the University of South Florida. Her primary research interests are applied statistics and data analysis. Her work has been published in Multiple Linea r Regression Viewpoints, the Journal of Research on Computing in Education, Educational and Psychological Measurement and the Proceedings of the American Statistical Association. A SAS programmer for 7 years, she uses SAS for the conduct of statistical sim ulation research, as well as for applied data analysis.