Help!! Negative (count adjusted) R2 in logit regression!

[Angelo Cozzubo]

Dear Statalist friends,

I am having troubles with the fitstats command since the R2 count adjusted is giving me negative results. Is there any intuition behind this?

Here I paste the Page Imprint

Thanks in advance!

[William Lisowski]

UCLA provides a good reference on goodness-of-fit measures in logistic regression models:

http://www.ats.ucla.edu/stat/mult_pk..._RSquareds.htm

Examination of the description shows that the adjusted count r-squared will be negative when you could have done better by just always guessing the most common outcome, regardless of the covariates. All these measures are to be taken with a grain of salt.