Inferring the order of the choice process using consumer purchase histories

The small-sample properties of the multinomial runs (MR) test, which is often used to test for effects of previous brand choices on current brand-choice probabilities, are explored in this article through a Monte Carlo study of different buyer-level brand-choice probability models. The power and the actual size of this test, under a null hypothesis of no previous brand-choice effects, are shown to depend on the nature of the true underlying model, length of purchase history, and skewness in baseline choice probabilities. We provide some guidelines for when the MR test is more useful in empirical research.