You might indicate what your research is about, but I would not recommend H-L for much of anything. Best wishes, David Booth

For more details about why the H-L should not be used see limitation section of the attached Wikipedia entry by Prof Frank Harrell. In addition you will find out why H-L is not working for you. Best wishes, David Booth

Hello Mutlu,

The issue of the Hosmer-Lemeshow test is whether the model behaves consistently/coherently across multiple (usually between 3 and 10, as implemented in some software packages) levels of the IV. So, that process obviously was envisioned for continuous or pseudo-continuous IVs.

If your IV is nominal, the H-L test isn't going to be helpful, as a nominal variable must be re-expressed as a set of k - 1 dichotomous variables (where k is the number of nominal categories). Within each dichotomous variate, the correspondence of "actual" to "expected" cases is perfect. So, for any variable that has only two levels in the data set (even if ostensibly continuous, or a re-expressed, more-than-two category nominal variable), the Hosmer-Lemeshow test doesn't inform as to model consistency.

Good luck with your work.

Dear Dr Gulbay: I would suggest you read Frank Harrell's course notes on the subject of logistic regression. They are attached. Please note that Dr Harrell's opinion on the H-L test is given in the Wikipedia article I attached previously verbatim. While it's nice that SPSS included it that really doesn't mean that you may use it for everything in logistic regrrssion.As Dr Morse indicated there are some situations in which it may be helpful. I don't find it in my software and that's ok with me. With respect to your new question I have no idea what the answer may be. I do not know exactly what you mean by that question. What for example can a nominal parameter be? In any case I would like to send you a paper that may give you a full example of a common use of logistic regression. The paper is currently in press. But this is the accepted copy. Hopefully with the attached references some common uses of the procedure will be available to you. All of the software is available as is indicated in the text. I would further suggest you download the Lander references from our paper from the z library and consider the logistic regression programs there. I believe it might be a better approach than the current SPSS approach you are taking. Examples are available in Harrell's notes as well as our paper. This may be helpful to you as you press forward with a somewhat different project. Any questions that you have are welcome here. Best wishes, David Booth

yes ofcourse!