Evaluating the Accuracy of Approximate Power and Sample Size Calculations for Logistic Regression

Abstract

This master’s thesis evaluates and implements power, sample size and effect size calculations for logistic regression. The earlier sections set up an ordinary logistic regression model, review the current approaches including those of Whittemore, Hsieh and Schoenfeld & Borenstein, and illustrate comparisons of the existing approaches. Schoenfeld & Borenstein’s method exhibits general superiority and, with slight modifications, is implemented in a Shiny web application and an R package. We give examples to demonstrate its use, and make recommendations about when its results can be considered accurate enough for applications.