0
$\begingroup$

This is a repost from stackflow forum. My purpose is to discover key relationship between variables. I have several categorical variables such as "nationality, office locations, job title" etc. In total I have 14 categorical which created about 100 dummy variables and 6 continuous variables.

My dependent variable is binary outcome, hence, logistic regression is used ( input all variables at one go). I encountered multiple warning messages that these variables are omitted due to collinearity in the result. In this scenario, are there any methods to "treat" these dummy variables to be able to include them in the model? Or I should use other methods of regression or other techniques to resolve?

Below is a sample of the result:

enter image description here

enter image description here

enter image description here

Thank you!

$\endgroup$
1
1
$\begingroup$

This may be due to complete separation, since you have a binary outcome with binary predictors. Some categories may have very close to zero events. If that's the case, one way to address this is by combining categories if possible, and if not then perhaps a penalized logistic regression may help. I'm not sure what software you are using, but both R and SAS have Firth's method of penalized logistic regression available.

$\endgroup$
3
  • $\begingroup$ I am using Stata and R. Rather new to both. $\endgroup$ – user149635 Apr 10 '17 at 2:41
  • $\begingroup$ Penalized logistic regression refer to lasso / ridge regression? Pls correct me if I am incorrect. I just done some reading and realized that these 2 methods are further enhancement on the linear regression OLS formula. Since my Y dependent will be 0 or 1, is it all right to use the lasso/ridge? $\endgroup$ – user149635 Apr 10 '17 at 9:09
  • $\begingroup$ I am referring to this: sas-and-r.blogspot.ca/2010/11/… $\endgroup$ – dwhdai Apr 10 '17 at 15:14

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.