stepwise regression on nonlinear data

yuhanmc — Fri, 09 Jun 2023 00:55:03 GMT

Hi,

I have a datasets with a research question: Is X contributing additional variance in Y in addition to Age? Based on the data, Age account for significant variance in both X and Y, but the relationship is nonlinear (When Age increase, Y decrease exponentially, and X increase exponentially). So when I fit Y by X (using 0 degree correlation), the relationship between X and Y looks linear. So to answer my research question, should I use step-wise regression, with Y as dependent variable, and X and Age as effects? Knowing that Age vs. X is nonlinear, and Age vs. Y is nonlinear, and X vs. Y is linear?

Thank you!

Re: stepwise regression on nonlinear data

dale_lehman — Tue, 27 Sep 2022 13:17:28 GMT

I don't see why you would use stepwise regression in this situation. Stepwise regression, aside from having problems (particularly with correlated variables) is a way of reducing the candidate factors to the most important ones. It sounds like you only have 2 factors, X and Age, so stepwise regression would not do anything meaningful for you. You need to consider both variables as factors influencing Y, and include potential interactive effects. You should also consider multicollinearity between X and Age (the VIF is a good indicator). There are a variety of ways to incorporate nonlinear effects, but I would start with visual displays to see how important they are and to suggest ways to model these. But I don't see stepwise regression as having a place in the analysis.

topic stepwise regression on nonlinear data in Discussions

stepwise regression on nonlinear data

Re: stepwise regression on nonlinear data