Abstract
In a generalized linear regression model, we consider a problem of selection of predictors, which includes the selection of estimators' type of regression parameters in addition to the selection of variables (models). This problem is quite important to deal the situation where a prediction area is different from a sample area. Based on the evaluation of goodness of predictor on Prediction Mean Square Error (PMSE), we first modify the predictors by changing the Generalized Least Squared estimator or the Ordinary Least Squared estimator, and give a criterion for the selection of predictors when these modified predictors are used. Second, we evaluate the goodness of the selected predictor by this criterion based on the expected PMSE with respect to the selection of predictors. This evaluation is done from a simulation study.
