Power Analysis for Categorical Methods

This article shows how to assess and control the power of chi-square tests for categorical data. It discusses both the case of simple null hypotheses (i.e., the category or cell probabilities are completely specified a priori) and the case of composite null hypotheses for parameterized multinomial models requiring parameter estimation (e.g., log-linear models). Power calculations are illustrated for chi-square tests of association and chi-square goodness-of-fit tests using standard examples from behavioral research. Keywords: statistical power; categorical data; power divergence statistic; likelihood-ratio chi square; Pearson's chi square