Sample size determination in logistic regression

Whittemore (1981) and Hsieh et al. (1998) have proposed different methods for determining sample size in the context of testing the significance of a slope parameter in logistic regression. Their sample size formulas have been incorporated in some statistical software packages. In this paper, we use a variation of Whittemore (1981) method to calculate sample size. We compare these three sample size formulas to assess closeness of the nominal and observed powers via simulations. These studies lead us to propose another method, which is a combination of Hsieh et al. (1998) method and the proposed variation of the Whittemore (1981) method for calculating sample size when the covariate is normally distributed. However, when the covariate has a Bernoulli distribution, sample size calculations widely diverge between the Hsieh et al. (1998) method and the proposed method. Interestingly, the proposed method gives a better account of power than the Hsieh et al. (1998) method.