0
$\begingroup$

I have a question regarding binary logistic regression. My DV is dichotomous (0=no, 1=yes), and I have 9 predictor variables (dummy-coded & scale). If I do for each predictor one regression analysis, some of them are significant. But as soon as I enter one predictor with other predictors, they loose their significance levels. My question: WHY?? I now that they depend on each other, but I really do not get why this happens.

And my second question: Is it ok to run separate logistic regression for the same DV?

THANKS FOR ANSWERS :-)

$\endgroup$
  • 2
    $\begingroup$ Stepwise variable selection is an invalid method - don't even consider it for your problem. Formulate a model based on subject matter understanding and stick to that model. Estimate its parameters and especially get confidence intervals for effects on meaningful scales. The full prespecified model should be the basis for all of your inference. $\endgroup$ – Frank Harrell Sep 10 '17 at 13:06
  • $\begingroup$ Thank you Frank! So if I am only interested in the influence of one IV on the the DV (indecent of other variables), is it legitimate to run separate logistic regressions (simple binary logistic regressions)? $\endgroup$ – Sophia Backhaus Sep 10 '17 at 13:35
  • 2
    $\begingroup$ No. Pre-specify a sensible model, fit it, and draw inference on the partial effect of the one variable of major interest (adjusted for all the others). $\endgroup$ – Frank Harrell Sep 10 '17 at 13:52
0
$\begingroup$

Because when you have more than one independent variable in the model, the model is controlling for those other variables. So, you are asking different questions and getting different answers.

$\endgroup$
  • $\begingroup$ Thanks Peter, Does this mean that if I am only interested in each predictor separately (and not interested in the influence on each other) I can run separate logistic regressions? $\endgroup$ – Sophia Backhaus Sep 10 '17 at 13:38
  • $\begingroup$ It would be very unusual to not be interested in the effects of controlling for other variables. Especially, as in your case, when they are related to each other. $\endgroup$ – Peter Flom Sep 10 '17 at 14:56

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.