Inference Properties of QIC in the Selection of Covariates for Generalized Estimating Equations

Abstract

The Quasi-likelihood information criterion (QIC)which results from utilizing Kullbacks I-divergence as the targeted discrepancy is widely used in the GEE framework to select the best correlation structure and the best subset of predictors. We investigated the inference properties of QIC in variable selection with focus on its consistency, sensitivity and sparsity. We established through numerical simulations that QIC had high sensitivity but low sparsity. Its type I error rate was approximately 30% which implied fairly high chances of selecting over-fit models. On the other side,it had low under-fitting probabilities. The statistical power of QIC was established to be high hence rejecting any given false null hypothesis is essentially guaranteed for sufficiently large N even if the effect size is small.