Fitting Finite Mixtures of Generalized Linear Regressions in strongRstrong.pdfVIP

Fitting Finite Mixtures of Generalized Linear Regressions in R Bettina Grun Friedrich Leisch ¨ Vienna University of Technology University of Munich This is a preprint of an article accepted for publication in: Computational Statistics and Data Analysis, 2006. /locate/csda Abstract R package flexmix provides flexible modelling of finite mixtures of regression models using the EM algorithm. Several new features of the software such as fixed and nested varying effects for mixtures of generalized linear models and multinomial regression for a-priori probabilities given concomitant variables are introduced. The use of the software in addition to model selection is demonstrated on a logistic regression example. Key words: concomitant variable, finite mixture, fixed effect, gener- alized linear model, R. 1 Introduction Finite mixtures of regression models are a popular method to model un- observed heterogeneity or to account for overdispersion in data. They are flexible models and in theory it is easy to modify and extend them by using more complex models for the component distribution functions and estimate the corresponding parameters, e.g., using the EM algorithm. R (R Development Core Team, 2006) features several extension packages for estimation of mixture regression models, e.g., fpc for mixtures of linear regression models (Hennig, 2000) and mmlcr for mixed-mode latent class regression (Buyske, 2006). However, like virtually all other (non-R) imple- mentations, they consider only a few particular types of mixture models and