Title: Bootstrap model selection for linear mixed models
Authors: Garth Tarr - University of Sydney (Australia) [presenting]
Alan Welsh - the Australian National University (Australia)
Samuel Mueller - University of Sydney (Australia)
Abstract: Linear mixed effects models are widely used in applications because they provide flexible models for a variety of types of clustered data. Model selection, which often aims to choose a parsimonious model with other desirable properties from a possibly very large set of candidate statistical models, is a key part of many applications. We discuss the use of bootstrap model selection in linear mixed models. Bootstrap model selection was originally developed for simpler models with independent observations. It is an interesting approach because of the flexibility it allows in permitting the use of measures of fit different from those used to define the estimators used to fit the models. We discuss statistical properties as well as computational issues and present both theoretical and simulation results.