I have just performed a multivariate binary logistic analysis on SAS and I was wondering which estimator of the confidence interval to take between Wald or by the profile-likelihood.
Wald tests/CIs are crude approximations. They were invented in times where is was too demanding to actually calculate the likelihood profile. The profile-likelihood test/CIs are better approximations.
Rule: If available at similar cost, use the better approximation.
@Jochen, Sure but in case you don't say so you have confused the reader. i am often confused on this point in reading these questions. Improvement in exposition(including my own) would be helpful. Best wishes, David