A strategy for the selection of regression models with qualitative variables

In this methodological paper, a strategy for selection of regression models is proposed. The application is possible in situations with quantitative regressand and two qualitative regressors. The proposal is based on a minimization of an estimator of the Mean Square Error of Prediction in certain predefined class of models. Some structures for interaction (as in Mandel, Johnson and others) are considered. Some graphical displays for the form of interaction are also proposed.