2
$\begingroup$

in the programs I know (i.e. R statistics) the default numbers of groups to use in Hosmer and Lemeshow test (goodness-of-fit for models, especially logistic regression) is set to 10. I wonder why or if I should define another number for my data. Because I also found that the number of groups is really important for the p-value which results from this.

Does it depend on the amount of data, the degrees of freedom, or something else?

$\endgroup$
  • $\begingroup$ Please see comments about this now obsolete test elsewhere on the site and what replacements exist. $\endgroup$ – Frank Harrell Dec 11 '15 at 13:48
2
$\begingroup$

The Hosmer and Lemeshow test is really sensitive on the group number and sometimes the change of group will lead to a disaster of conclusion. My suggestion is still use other test for comparison. And be careful with HL-test.

The issue of this is presented in this paper: https://support.sas.com/resources/papers/proceedings14/1485-2014.pdf

And another published paper proposed alternatives and adjustment of HL-test, and also gave recommendations for different situations. Here: http://onlinelibrary.wiley.com/doi/10.1002/sim.5525/full

$\endgroup$
  • $\begingroup$ I think the links help especially the first. Do you really think a slight change in the number of groups can be a disaster for the HL test. If so why? The second one could be the best for addressing the question. But there is not free access to the Statistics in Medicine paper. $\endgroup$ – Michael R. Chernick Jan 12 '17 at 5:35
  • $\begingroup$ Actually, I don't think HL test is reliable due to its sensitivity of groups number that need to pre-specified and the problematic power. These two papers point out the issue of HL test. And I also simply use HL to test a binary variable generated by logistic regression and a random error series between 0 and 1 representing predicted probability, the HL cannot reject the null hypothesis that the model fits the data, which is ridiculous. I can send the second paper to you if you want. $\endgroup$ – Xiaorui Zhu Mar 21 '17 at 21:32
  • $\begingroup$ Please leave your email I can send you that paper. $\endgroup$ – Xiaorui Zhu Mar 22 '17 at 21:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.