I have data involving 3 key variables (predictors), 2 of them being randomly assigned treatment conditions, and the other denoting the sample (A vs. B). I am finding it easier to present the results doing 2 regressions each with a two-way interaction, but just wanted to check if there's any statistical difference/issue with doing that vs. putting them all in one regression and doing a 3 way interaction.

Thank you!


1 Answer 1


The issues are the same as when comparing subsets vs. a single two way interaction: You don't get parameter estimates, p values, etc. for the difference in the two subsets, but you would get these with the interaction.

Here, it's all just one level "up" - you don't get parameter estimates for the differences between the two models, each of which has a two way interaction. How big is the three way interaction? What's its standard error? P value? etc.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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