A better stopping rule for conventional statistical tests

Abstract

The goal of some research studies is to demonstrate the existence of an effect. Statistical testing, with p less than .05, is one criterion for establishing the existence of this effect. In this situation, the fixed-sample stopping rule, in which the number of subjects is determined in advance, is impractical and inefficient. This article presents a sequential stopping rule that is practical and about 30% more efficient: Once a minimum number of subjects is tested, stop with p less than .01 or greater than .36; otherwise, keep testing. This procedure keeps alpha at .05 and can be adjusted to fit researchers' needs and inclinations. Copyright 1998 Psychonomic Society, Inc.