I am doing multilevel logistic regression for cluster effect (student from 15 schools). My null model ICC is 0.10, but when I added some predictors, it's ICC increased to 0.12. In this situation, can I add predictors or not?
Thank you
Please note Prof Kelvin Jones response here: Why does the intraclass correlation coefficient (ICC) increases as fixed components are added in a multilevel logistic regression model? (researchgate.net) . Yes you can. This effect is exactly what you hope will happen. Best, D. Booth
You can always add predictors, but should you?
I suggest you make a scree plot of ICC with increasing predictors. Once you establish the best predictors, you should see that adding more predictors, no matter which one, all produce about the same increase in ICC.
There are versions of which ICC penalize for too many predictors, similar to adjusted R-square. Look to see if this is available in whatever software you are using.
David Eugene Booth and Dan Eastwood sir, thank you very much for replying and suggestion.
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation