336 – Computational Techniques for Mixtures
An Analysis of Categorical Injury Data Using Mixtures of Multinomials
Minglei Liu
Medtronic
Jorge Morel
The Procter & Gamble Company
Nagaraj K. Neerchal
University of Maryland Baltimore County
Andrew M. Raim
University of Maryland Baltimore County
Finite mixture models are useful for data that exhibit heterogeneity from unobserved sources. Such models can assign observations into a set of latent classes, and may be helpful in understanding the nature of the heterogeneity. In this paper, the finite mixture of multinomials model is applied to an injury dataset in order to study the probabilities of several injury types common among emergency service providers. Computational techniques from (Raim et al., 2012) are used to determine the number of mixing components, obtain estimates, and compute standard errors and confidence intervals. We find that three classes provides an adequate model for the data, and that the class compositions differ by geography and gender.