2
$\begingroup$

I am working on a model in which I would have percentages among my independent variables. To be more specific, I would include in the model the variables that, together, would attain a 100 (say I want to explain Diet and the two variables I have are the share of apples and the share of pears..).

My question is: do I necessarily have a problem of multicollinearity or of too high correlation between my independent variables?

Thank you very much!

$\endgroup$
2
$\begingroup$

Because your independent variables must necessarily sum to a constant, there is bound to be correlation amongst them, as they cannot vary independently. The situation is similar to the correlation of the residuals from a linear model. They are correlated because we are imposing that their sum is zero through the first order conditions of the OLS estimator. For just two-three variables that you are describing, I would expect quite high correlation.

$\endgroup$
  • $\begingroup$ Thank you for your answer. Actually, i figured i would have had correlation since, yes, as you said, those vars are not independent the one from the other. The thing is that I sense that this may be a big issue for the result i may obtain. Right? Would it be too big an issue? Would it be less an issue if, say, i have three different shares that together amount to 100 but i just want to test the effect of two of these? My feeling is that i wouldn t have a problem with two. What do you think? Thanks again for your kind answer! $\endgroup$ – Pier Jun 7 '15 at 19:31

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.