Ask me a weird question, "what is the weakness of logistic regression?" I did not know it. So I answered frankly, "sir, I do not know this question. Can you suggest a book, or a paper about it so that I can learn it?" He refused to answer and finished the interview. Later I asked several professors in department. One professor said "I guess he means the logistic regression does not work well if the data contains too many zeros (something like perfect separation)". Well, if this is the answer, is it the weakness of logistic regression, or the flaw of dataset itself? I think It is not a weakness of a statistical method, it is that we fit a wrong model.
