Generalized linear model selection using R2R2

The problem of model selection in generalized linear models amounts to selecting a subset of useful covariates from a set of possible covariates and choosing a link function from a set of possible link functions. A model selection procedure based on a modified R2R2 statistic is proposed. Like in linear models, R2R2 statistics in generalized linear models are used to quantify the proportion of variance in the response explained by covariates. Model selection using R2R2 statistics is natural for investigators who are familiar with the use of R2R2 statistics. The modified R2R2 statistic is obtained by introducing an extra penalty term on the complexity of the candidate model. Under weak conditions, the proposed procedure is shown to be consistent in the sense that with probability tending to one (as the sample size increases) the selected model equals the optimal model between the response and covariates. Simulation results are presented to demonstrate the effectiveness of the proposed procedure in finite sample applications.