Finite Mixture Models: Know What You Don't Know*Funda Gunes, SAS Institute Inc.
David Kessler, SAS Institute Inc.
Keywords: mixture models, latent classes, overdispersion
Every statistical practitioner knows that you can't always measure everything. This can be due to the difficulty of making reliable measurements or to the inherently unobtainable nature of some characteristics. If you ignore hidden structure, you can miss important effects that vary between the populations that these qualities define. This issue of "known unknowns" invites the application of the finite mixture modeling technique, which uses a flexible structure to address the latent class problem directly. This poster presents an epidemiology example motivated by the work of Nierenberg et.al. (1989) to explain the approach. We will illustrate the flexibility, utility and appeal of finite mixture models, using methods implemented in SAS. We demonstrate how you can model the components of the study population, sharpen the analysis after accounting for these divisions, address overdispersion and develop useful predictive models.