I want to test the hypothesis that IV 2 (post-test value) is a better predictor for the criteria (no success vs. success) than IV 1 (pre-test value). Those IVs are indicators for static vs. dynamic learning. I conducted a logistic regression analysis for both IVs seperately and it seems that IV 2 is indeed the better predictor.

Now I would like to support this finding with the wilcoxon-test. I seperated the data into the groups "no success" and "success", but I'm not quite sure how to go from here because I have trouble to understand which procedure is correct to test my hypothesis...

Do I have to compute a variable with a hypothetic value of 0 ---> wilcoxon test with IV 1 and the hypothetic value variable, repeat this step with the IV 2

Or do I have to follow the usual procedure and just compare IV 1 and IV 2?

What is the right procedure to complement the logistic regression analysis? Thanks for any advice!


If you have both IVs in your model then 1) You are testing which has the higher odds ratio after controlling for the other one 2) Your two IVs may be colinear and 3) Your data may be dependent.

If, instead, you test the two IV's separately (still in logistic regression) then you are seeing which has the higher odds ratio on its own.

However, "better predictor" is a bit ambiguous. Do you mean "if we choose the best cutoff for (score 1 or score 2) how well does it predict success?"

  • $\begingroup$ by "better predictor" I meant, that it simply has a higher nagelkerkes R² and a higher odds ratio when I test the predictors seperately. would it be smart to test those predictors again in one model regarding my hypothesis? $\endgroup$ – Jennifer Feb 26 '14 at 14:19
  • $\begingroup$ Whether to test them together or apart depends on whether you want to control for the other one. But probably apart. $\endgroup$ – Peter Flom Feb 26 '14 at 14:50

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.