Interpreting Hosmer-Lemeshow Goodness of Fit?

Hi all, I'm having a bit of trouble wrapping my head around the Hosmer-Lemeshow Goodness of Fit test for logit regressions. Specifically, I am confused as to whether a high p-value indicates a GOOD fit or a BAD fit.

What is the H-L test checking? Is is checking Goodness of fit or BADness of fit?


TS Contributor
If I recall, the null is good fit. Lower p-values suggest more evidence to contradict "good fit". Larger p-values just mean you don't have enough evidence of "bad" fit.


Not a robit
@ondansetron - is on the right track. Typically a smaller p-value goes in the direction of rejecting good fit. Some people have issues with this test, but I think it has its place. If I recall you are conducting a chi-square test of the predict versus observed outcomes and its plot will help you visualize the fit of your model on data across the predictions (e.g., deciles).