What diagnostics for random effects logistic regression?

Question

I'm wondering what checks/diagnostics you usually calculate and report if you do a random effects logistic regression.

• C-statistic/ROC curve
• Check multicollinearity?
• Check heteroscedasticity?
• McKelvey & Zalvoina R-squared
• ...

There is a lot of information about standard logistic regression and what you should check for, but I'm missing an overview about what applies to logistic regression on panel data.

Thanks

Answer 1

It is true that there is not much information about diagnostics for random effects logistic models.

• R package influence.ME provides tools for detecting influential data in mixed effects models, e.g. DFBETAS, Cook's Distance.
• There is a similar function in Stata, gllamm, including DFBETAS and Cook's Distance to detect influence points, empirical Bayes (EB) prediction of higher-level residuals, and detecting outliers by cross-validation.
• For population average models, e.g. generalized estimating equations (GEE) models, there is a section "13.3 Residual analysis and diagnostics" in the book Applied Longitudinal Analysis by Garrett M. Fitzmaurice, Nan M. Laird, James H. Ware. There is also an example of using cumulative sums of residuals to assess the adequacy of the model.