In social science and statistical literature, ordinal qualitative dependent variable models have received substantial attention in terms of theory and application. However, linear models with ordinal qualitative regressors have been overlooked. In this paper, we propose an approach to regression model selection with two-qualitative regressors, which will be used to build a technique for selecting qualitative variables based on the mean square of the prediction error (MSEP). Several data sets were simulated using two models, and the technique was found to choose the best model in all cases. The findings show that significant improvements in bias and efficiency may be made when compared to other estimates. To diagnose the type of interaction, certain graphs are provided. The proposed approach achieves better accuracy with reduced MSEP, improved stability, and faster convergence compared to other models.