When studying Linear Regression, I remember that Multicollinearity is something that impacts inference and not prediction, at least not always. Also, I noted that the assumptions tends to be neglected by those who are only interested in predictions.

Is there anything like that for Logistic Regression? I couldn't find any thread related to this question, except for this one. It helped me a little, but it doesn't quite answer my question. I mean, I'm not actually talking about causality, since that is not achieved by these regressions by themselves, but about coefficients interpretability and estimators.

Well, I hope the question is good enough, it doesn't break any rules, but I'm always skeptic when it's something that broad. If I missed something, I have no problem in deleting it. And I apologize in advance for that.


1 Answer 1


So as you point out in your question, individuals interested in only predictions are often not interested in verifying assumptions of their models. Another way to put this is that they are not interested in looking at (or computing, for that matter) standard errors corresponding to the coefficients they estimated for prediction. Standard errors are what allow an analyst to go beyond interpreting the coefficient itself and make inferences. Large coefficients and small standard errors signify confidence that a true (population) effect exists. Small coefficients accompanied by large standard errors suggest no true effect in the population.

See the link below for a more detailed conversation on standard errors and logistic regression.

Understanding standard errors in logistic regression


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.