A Single-Step Modified Bonferroni Procedure for Multiple Tests

• Activity Number: 405
• Type: Contributed
• Date/Time: Wednesday, August 6, 2008 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Michelle A. Roozeboom*+ and Daniel Mundfrom and Jamis Perrett
• Companies: Iowa Foundation for Medical Care and University of Northern Colorado and Texas A&M University
• Keywords: Regression ; Type I Error ; Statistical Power ; Simulation ; Multiple Tests ; Bonferroni Adjustment

Abstract

The Bonferroni method maintains a Type I error rate less than or equal to the nominal level when multiple tests are performed. However, this adjustment reduces power. The purpose of this research was to establish a single-step method for multiple tests that controls Type I error while increasing power. Sample data (n = 25, 50, 200) were simulated to have normal distributions for hypothesis testing of 3 to 20 beta coefficients. Correlations between x and y and among the xs was varied. A new denominator related to the number of tests performed was less than or equal to k. Mean Type I error for Bonferroni, Holm, New, and No Adjustment were .031, .035 , .055, and .317 (SDs= .011, .011, .019, .115, and N=2958). Mean statistical power was .339, .344, .361, and .496 (SDs=.391, .394, .390, .363). The new method resulted in increased statistical power while correcting for inflated Type I error.