I run a binary logistic model and had a results showing significant in deviance and non-significant in Pearson GOF test. I also tried HL stat and showed non-significant.
How do I decide on that model whether it is fit or not fit?
What will be possible problem for that and
what will be possible solution for that?
Anyone who can answer that problem, please help me.
How many covariate patterns/ profiles do you have in your model?
You could try adding interactions or higher order terms to your model and make some diagnostics (influential observations).
I think you are OK with the GOF because two tests showed nonsignificant.
How about an R-squared value and AUC?
Here is a relevant recent thread you might have a look
https://www.researchgate.net/post/Is_it_ok_to_use_binary_logistic_regression_when_two_goodness_of_fit_tests_give_a_p_value_of_0000
I did not understand the last sentence in your reply. Do not bother with GOF in your case.
Goodness of Fit test is very sensitive to empty cells (i.e cells with zero frequencies of specific categories or category). Cut down on cells with high percentage of zero frequencies if possible. Note that in this situation as yours, neither of the chi-square tests of deviance and pearson provides reliable test for goodness of fit. A more efficient method is the Likelihood ratio test which compares the complete model to the "intercept only model".Check the following references supporting the above arguments (Schendera, 2014, page 194; Backhaus et al. 2011). All the best.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis