For instance, reference to an interaction term will almost always be made when Two-way ANOVA is taught. However, when considering a regression with two continuous predictors and one continuous outcome, I've found (psychology teaching context) that students are often surprised that it's possible for there to be an interaction term.

  • 1
    $\begingroup$ ANOVA is regression, mathematicall, so this has nothing to do with the methods, only with traditions of teaching. $\endgroup$ – kjetil b halvorsen Jun 9 '15 at 14:03
  • $\begingroup$ While I accept that the division between ANOVA and Regression is just a historical accident, I wondered if in this case there might be some pedagogical or other reason for the teaching tradition I have been observed. That is, is there some difference between a regression with two categorical predictors and one with two continuous predictors that justifies a more explicit consideration of interactions? $\endgroup$ – user1205901 - Reinstate Monica Jun 10 '15 at 3:46
  • $\begingroup$ One difference may be that ANOVA is often used in designed experiments, with (at least close to) orthogonal predictors, so interactions might be estimated in a stable way. with regression, more often used in observational studies, colinearity is more often an issue so that estimation of interaction may be unstable. $\endgroup$ – kjetil b halvorsen Jun 10 '15 at 8:42

I guess it depends on the field. In econometrics, interaction terms are very often included as explanatory variables in regressions, and this is taught to students.


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.