Estimating the Number of Null Hypotheses in Multiple Testing

• Activity Number: 182
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Vasyl Pihur*+
• Companies: ASA
• Keywords: false discovery rate ; Benjamini-Hochberg procedure ; multiple testing

Abstract

Unbiased and efficient estimation of the proportion of null hypotheses p0 is highly desirable in all adaptive false discovery rate (FDR) procedures. A good estimate of p0 allows for a tighter FDR control which translates into a larger number of significant discoveries while maintaining the appropriate error rate. The method of estimation of p0 proposed in this work is based on the widely used Benjamini-Hochberg rule and produces estimates whose bias and variance are comparable to those of the best existing methods.