Comparison of procedures for estimating sample numbers

Abstract

Published standard methods for estimating sample numbers entail hidden tradeoffs. This note will describe the nature of these tradeoffs and their consequences. A modified version of the standard method is presented. The modification recognizes that the desired answer is implicit in the t‐statistic. The modified version uses a succession of estimates until achieving convergence. Examples are presented showing that either under‐ or overestimation is possible when using the original methods compared with the modified method. The variation in estimates reflects the variation in the tolerance probability. We recommend when to select specific methods.