I am trying to creat a multiple regression model with a forward stepwise procedure. Predictors are air temperature, soil temperature, PAR and snow depth. I also want to see if there are some interaction effect between snow depth and other factors that can also be included in the model. Should I include the interaction as factors (e.g. Ta:Snow depth) when going through the stepwise procedure (main effect may not be included) or make another model to illustrate the interaction effect separately (include both main and interaction effect)?

PS:I am using R to do this.

Thanks a lot in advance!


(1) No one here likes stepwise. Again...just to be clear. No one here likes stepwise.
(2) In this example, unclear why you wouldn't use backward stepwise if you want a stepwise procedure. Usually preferred and makes interactions easier to deal with (examine).
(3) If you have an interaction, you want the main effects to be included.
(4) You can either reduce to main effects model, add interactions of interest (with main effects if needed) and examine significance or ... add interaction of interest and removed using backwards (again only removing main effects if interaction already removed).
(5) Interactions are tricky. One is often underpowered to detect interactions and they raise multiplicity (multiple testing) issues. Everyone has an opinion about this, but can get rather longwinded. Worth looking into if your interested.

| cite | improve this answer | |
  • $\begingroup$ I am new here and not sure why people don't like stepwise here. But thanks for the useful and clear suggestion! $\endgroup$ – Bin Feb 4 '14 at 8:56
  • $\begingroup$ Glad it was helpful. Good luck. $\endgroup$ – charles Feb 4 '14 at 17:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.