0
$\begingroup$

I'm looking for group differences in a variable (Power in a frequency band) that depends on age (i.e. increases linearly with age). If the two groups are matched for age, do I still need to include age in my model (e.g. Power ~ Group + Age)?

Also if the relationship of Power with Age differs between the two groups (e.g. the slope is steeper for Group 1 than for Group 2), should I include an interaction term in my model to take it into account (e.g. Power ~ Group * Age)?

I guess I could generalize the second part by asking when do we need to include an interaction term in a linear model to correct for a confounding factor?

$\endgroup$
1
$\begingroup$

First, it is almost always bad practice to have an interaction in your model unless both variables are in your model.

Second, even if you have matched on a variable, the variable could be important.

$\endgroup$
  • 1
    $\begingroup$ I apologize, that was not clear on my part. When I say to include an interaction term I mean of course to also include the main effect of Age so the model would be Power~Group+Age+Group*Age. $\endgroup$ – d myl Mar 22 at 19:38
  • $\begingroup$ Would you please elaborate on the second part of your answer. What do you mean by "the variable could be important"? Many things could be important but not all of them are included in our model, unless we have reason to believe they are confounding our results. $\endgroup$ – d myl Mar 25 at 15:36

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.