Model Selection On Interactive Effects

Since the 1960’s.model selection has always been the hot spot of the study by s-tatistics. Model selection is often accompanied by multicollinearity problem and choos-ing the important independent variable, no matter what kind of reason, the final purpose of model selection is to make the effect of model explaining and predicting be enhanced.Our paper is to study the joint action of main effects and interaction effects to response variables. Elastic Net is a relatively new model selection method and it has many good properties, it is suitable for p>n and n>ρ model, but the combination of the theory and practical application research is insufficient. Meanwhile, the dimension (p+p(p-1)/2) of our model in this paper will be bigger than the sample size, and the general model selection method is not applicable, so our paper adopts the Elastic Net method and using the simulated data and empirical data for interaction effect model selection, on this basis, we estimate parameters by OLS.This article verifies that the Elastic Net method has a good effect on the interac-tion effect model, and verifies that the interaction effect is common in our daily life, it deserves more scholars to mining and research.