I am running a multiple linear regression using SPSS to test the effect of ethnicity and ethnic/racial attitudes or perceptions on political predispositions.

One model, as an example, looks like this:

DV: 'support for democracy' (1-7 scale) Controls: age, gender, education etc. Predictors: Ethnicity, and various attitude/perception vars

My question is what to do with these attitude/perception variables when they are correlated; i.e. whether to leave all of them in the model, or remove one or more. A particular case is the pair of variables 'common national culture' ('Bolivians share many common values that unite us as a nation; 1-7 disagree-agree scale) and 'strength of national identification' ('To what extent do you identify as a Bolivian citizen?'; 1-7 scale).

As you might expect, 'common national culture' and 'strength of national ID' are correlated (Pearson coefficient=.268 and is significant). But they also both have significant coefficients in the multiple regression, and adjusted R-squared for the model decreases substantially when either one is removed. In this case - i.e. when both correlated vars have significant coefficients -, should both be kept in?

Many thanks in advance.

VIF is between 1.2-1.4 for each.

Condition indices have values of around 12, but there are 15 variables in the model, so perhaps this is not so remarkable? Looking through the variance proportions, however, neither have any values >0.2


1 Answer 1


You should first do a search on the issue of multicollinearity and how to best understand if it is occurring. If you still have further questions then following up might be appropriate.

Edit: You should be okay. A VIF over 2 worries me a bit. According to this reference 2.50 should be cause for alarm. The more important distinction is how you interpret those two variables in the context of theory. I see this a lot in my field, where often it seems like predictors are included without any kind of theoretical meaning.

I hope that helps. If so, feel free to up vote.

  • $\begingroup$ I have updated the original post with the collinearity diagnosis info for the two vars in question. From what I understand about multicollinearity (which is not much, admittedly!), this should be ok. $\endgroup$
    – Sam
    Commented Jul 15, 2014 at 13:51
  • $\begingroup$ But I am not very experienced with this, so hopefully someone can confirm $\endgroup$
    – Sam
    Commented Jul 15, 2014 at 13:52
  • $\begingroup$ @Sam I edited my post for you. $\endgroup$
    – bfoste01
    Commented Jul 15, 2014 at 14:08
  • $\begingroup$ Thanks for your help (can't up the vote due to low 'reputation', sorry!) $\endgroup$
    – Sam
    Commented Jul 15, 2014 at 14:19

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.