1
$\begingroup$

I have two logistic regression models, using the same data set and same dependent binary variable but with different sample sizes due to different IV's. How would I go about comparing the two models aside from using a classification matrix?

$\endgroup$
  • 2
    $\begingroup$ Do you want to compare the model fits, or test for differences between the terms across models (which may or may not be possible if you have different sample sizes). Have you considered using list-wise deletion to make the sample sizes comparable (if you have missing data issues rather than different samples due to non-overlapping IVs)? $\endgroup$ – mCorey Aug 7 '12 at 2:54
  • 2
    $\begingroup$ What is it that you want to find out? You could compare them in many ways (there are lots of statistics associated with a logistic model). Which one is relevant to you depends on what your question is. $\endgroup$ – Peter Flom - Reinstate Monica Aug 7 '12 at 11:09
1
$\begingroup$

Don't just compare the classification matrices--compare the entire ROC curves.

$\endgroup$
  • 1
    $\begingroup$ Hi, Michael, welcome to the site. I think you make a good point here, but I wonder if it might be helpful for the OP if you provided a little more information. Can you say what a ROC curve is & why it might be better; perhaps even provide a link to some internet resources that will allow the OP to begin exploring further? $\endgroup$ – gung - Reinstate Monica Aug 7 '12 at 13:57
  • $\begingroup$ When you say compare the entire curve do you mean the entire AUC or the AUC for some cut off line? $\endgroup$ – Stats34534 Aug 7 '12 at 18:43
0
$\begingroup$

to find the best fit model do frequency test like chi square if ur using R here is the sample code


anova(fit.reduced, fit.full, test="Chisq")


$\endgroup$
  • 1
    $\begingroup$ This doesn't work, the OP said the two analyses don't even use the same samples. The two models are certainly not nested. (Plus, even if the samples were equal, nothing in the OP's question says that the IVs of one analysis is a subset of the IVs of the other analysis.) $\endgroup$ – Patrick Coulombe Aug 6 '14 at 7:04

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.