I am working with the package "survey". I implemented a logit regression with svyglm (pseudo ML with survey weights). I now would like to test the goodness of fit.
I use the ROC Curce and pseudo R. As well I use the hosmer-lemeshow test.
Is there already a better alternative for the hosmer-lemeshow test, which can be used with weighted survey data?
My survey sample includes n=1000 participants. All variables are either binary or categorical.
Would you recommend a minimum amount of observations in each category cell (in contingency table)?