Professor Ed Spitznagel
Department of Mathematics, Washington University
 
 
Title: A Power Primer
 
Abstract: One of the most important tasks a consulting statistician can perform is to tell his or her client what sample size is needed for a proposed study.
 
In frequentist statistics, sample size is a function of the desired power, and the usual standard is to have 80% power to detect a meaningful difference at the 0.05 significance level.
 
In this talk I will introduce the idea of power with a simple example and then demonstrate four practical methods for determining power given sample size and sample size given desired power.
 
Method 1: A known-variance or large-sample case. (Why 2.8 is a fundamental constant of the power universe.) Portable power with the TI-83.
 
Method 2: Noncentrality parameters: From UNIFYPOW to PROC POWER.
 
Method 3: A handy-dandy Excel-based method for inserting graphs into proposals.
 
Method 4: When all else fails, simulate!