ABSTRACT

This chapter explains the model selection techniques in Generalized additive models for location scale and shape (GAMLSS). It examines model selection techniques for a GAMLSS model; stepwise selection functions for selecting explanatory terms; and techniques for selecting smoothing parameters. Model selection for regression models is primarily concerned with the problem of selecting relevant predictors from a larger set of potential predictors. The performance of a statistical model is usually related to its explanatory or predictive capability on an independent data set. Cross validation techniques can be employed as an alternative to the generalized Akaike information criterion for the comparison of GAMLSS models, particularly for predictive modelling. The training data is used for model fitting, the validation data set is used for model selection and the test data set is used for model assessment. The practice of selecting only one model ignores uncertainty due to model choice, and can lead to over-confident inferences.