I calculated a binomial logistic Regression with R and I am surprised about the influence of one of 8 independent variables.

The two groups have got almost the same average value regarding this variable, but nevertheless, the variable is highly significant and show relatively high estimates in comparison with the other independent variables.

To understand why this could be as it is, I plotted the distributions of this independent variable for the two groups.

The two groups are unequal, but the ratio should be ok:

Group one: N=1252 Group two: N=372

The result of the analysis says, the higher the variable is, the more likely it is, to be in group two.

Do you think this is realistic, when you look at the plots? Or might there be any other Problem?

There are no problems with multicollinearity.

  • 1
    $\begingroup$ The two distributions do not look at all similar to me, by eye. On the basis of the remaining details you give us there seems no reason to suspect anything wrong with your model. $\endgroup$ – mdewey Apr 20 '16 at 12:37
  • $\begingroup$ Thank you for you response! To put in a nutshell: Even if the average value of an independent variable of group two is lower, a positive estimate which say that a higher value of this variable results in a higher probability of being in group two, is not necessarily wrong? Because of the special underlying distribution? $\endgroup$ – Ole Apr 20 '16 at 13:15
  • 1
    $\begingroup$ It is impossible to be sure since we do not have any more details. I would question whether comparing your two groups on the basis of an average is likely to be helpful when the distributions are so different. $\endgroup$ – mdewey Apr 20 '16 at 14:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.