1
$\begingroup$

Bit of a strange question...I'm trying to put together a multiple regression model. I have daily temperature as one of my control variables. I really want to control for temperature, but interestingly, removing it seems to very slightly increase my adjusted-R squared (from 0.23 to 0.24 or similar).

I know theoretically, I should just remove it right, as it adds nothing to my model? But given that the increase is so small and I quite like controlling for it in my model, would it be totally wrong to keep it in? If I take it out, am I still "controlling" for temperature in my model?

$\endgroup$
1
$\begingroup$

Generally, a higher adj. R-square is better. In your case, you might be better off working on the representation of temperature in the model. It depends on the data, really, but you could try polynomials for temperature (squared term or so) or you could make „classes“. Plot temperature against y, to see if there is a non-linear trend in the data. If so, you need to try to represent this structure in your model, because OLS regression is parametric by definition.

Adjusted R-square balances the advantage of adding a new variable and the disadvantage of losing degrees of freedom. So in your case, the explanatory power of temperature seems to add little to your model. This may change if you change the representation of the variable, so that it better fits the data generating process. You need to try.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks, that's helpful! If I can't find a way to fit temperature into my model in a way that improves it, am I still "controlling" for it? I guess I am, as it doesn't seem to be significant...? $\endgroup$ – Crimsoneer Aug 30 '19 at 15:12
  • $\begingroup$ you can only claim to "control for" something if you have it in your model. Is your model supposed to be a causal model? Or do you aim at predictive power? $\endgroup$ – Peter Aug 30 '19 at 15:17

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.