Model selection criterion based on the prediction mean squared error in generalized estimating equations

Abstract

The present paper considers a model selection criterion in regression models using generalized estimating equation (GEE). Using the prediction mean squared error (PMSE) normalized by the covariance matrix, we propose a new model selection criterion called PMSEG that reflects the correlation between responses. Numerical studies reveal that the PMSEG has better performance than previous other criteria for model selection.