Dear all, thanks in advance.
I have an issue that freaks me out.

I wonder if I'm wrong to set up the cross-structure of my data.

I have a study in which I have 500 surgeons (ID variable 1-500) in which a binary "success" outcome is measured.
"Success" is measured for each surgeon in a condition (surgery schedule: morning, afternoon, evening Variable "schedule") crossed with the variable "assisting by another surgeon" (variable "assisted". Three levels: no assistance, one surgeon, two surgeons).
In the model I also correct for hypothetical variables predicting the outcome.

After transforming the dataset to long and assuming an exchangeable matrix between random effects, the command I use is:

xi: melogit success i.schedule##i.assisted continuouspredictor i.categoricalpredictor || _all: R.schedule: || assisted:, or cov (exc)

It does not converge in any way or in any case takes a full day.

If, on the other hand, I consider the factors "schedule" and "assisted" nested and not crossed (ie I remove _all: R.) The model converges and gives me interpretable results.

Does anybody have ideas?

Thanks again!
Gianfranco