2
$\begingroup$

I have several independent variables & one dependent variable for a regression model.

One of these IVs have a much better curve fit (with the DV) as quadratic regression.

I was thinking about transforming the IV (by squaring it) and then adding both the IV and its squared value in the multiple linear regression model.

However, I've heard that if I do that, I can only keep either one or the other (i.e. either the IV or its squared values) in the regression model because they're co-dependent.

What would be the correct way to go about this? Can I add quadratic regression to a multiple linear model at all?

$\endgroup$
  • $\begingroup$ Yes you may add the IV and it's squared value if it is supported theoretically and diagnostic shows that it is needed. $\endgroup$ – tatami May 16 '17 at 3:12
3
$\begingroup$

It's totally fine to include both the linear ($X$) and quadratic ($X^2$) terms in the model, and I recommend you do so. Whoever you were talking to is right that you can run into problems with multicollinearity when adding polynomial terms, though, if you don't center your predictor before making the higher order polynomial terms. If you center $X$ before making $X^2$, then that problem will be greatly reduced and chances are you'll have two relatively independent variables in your model. This is even mentioned in the wikipedia article on multicollinearity:

Mean-center the predictor variables. Generating polynomial terms (i.e., for $x_1, x_1^2, x_1^3$, etc.) or interaction terms (i.e., $x_1*x_2$, etc.) can cause some multicollinearity if the variable in question has a limited range (e.g., [2,4]). Mean-centering will eliminate this special kind of multicollinearity.

You can always check the level of multicollinearity for your predictors by getting the tolerance or variance inflation factors for your predictors after you estimate the model.

$\endgroup$
  • $\begingroup$ Thank you very much! This helps a lot. Should I mean-center all of my IVs or just the one I'm squaring? $\endgroup$ – ms_caalis May 16 '17 at 20:38
  • $\begingroup$ @ms_caalis This issue is only related to the IV you'll be squaring, so definitely mean-center that one, but for the others it doesn't matter either way. $\endgroup$ – Rose Hartman May 17 '17 at 15:06

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.