Replicated discrimination tests: Dirichlet–multinomial (DM) model
Jian Bi · 2015
The Dirichlet–multinomial (DM) model is a natural extension of the beta-binomial model. It can be regarded as a multivariate version of the beta-binomial model. One of the earliest discussions and applications of the DM model appears to have been by Mosimann. Ennis and Bi discussed its application in the sensory and consumer fields. This chapter discusses the DM distribution, estimation of the parameters of a DM model, applications of the DM model in replicated ratings and discrimination tests, testing power for DM tests, and the DM model in a meta-analysis for usage and attitudinal (U&A) data. The DM model is suitable for use with replicated ratings data, especially for the three- or five-point Just About Right (JAR) scale data and purchase intent data. It has been noted that chi-square testing can be used for both multinomial data and overdispersed multinomial data (i.e., DM data).