The myth of continuity-corrected sample size formulae

Abstract

In this paper, we consider methods for determining sample sizes when inference on two proportions is required. It is assumed that the proportions axe from independent samples. Fisher's exact test is regarded as the relevant test for sample size determination in this case. However, the computation involved is extremely laborious, and so good approximations are extremely useful. One of the best-accepted approximations is due to Casagrande, Pike, and Smith (1978, Biometrics 34, 483-486). This approximation is supposed to be derived from applying the correction for continuity when a normal distribution is used to approximate a discrete distribution. However, a consistent application of the corre..