Restriction of Candidate Modelsin Orthogonal Regression

A Monte-Carlo method to identify excessively underfit and overfit models is presented for orthogonal regression models. We propose using the distribution in the reduction in SSE for adding one variable as a means of identifying underfit and overfit models. Once identified, these models are eliminated from the list of candidate models. Our approach is used with current variable or model selection techniques. A Monte-Carlo study is presented illustrating our procedure.