Convergence issues with GLLAMM

Hi all,

I am trying to run a multilevel model on Stata 15 using gllamm. Our data comes from multiple waves of a repeated cross sectional survey conducted in many countries, so we have individual respondents nested within country-years (CY) which are nested within countries (COUNTRY).

The outcome variable is a binary, so the model is logistic. In the example below we have four individual-level covariates and one level two covariate.

Code:

xi:gllamm drinker C01 i.C49 i.ESPAD_Year parentdis PC, i(CY COUNTRY) family (binomial) link (logit) nip(3) adapt trace

We have managed to get very basic versions of these models to converge (e.g. cutting down to just 3 waves of the survey for 4 countries and one or two covariates), but anything more complex iterates for hours with “(not concave)” after the log likelihood for each iteration.

Eventually model output is produced, but with two error messages:
“convergence not achieved” and “Condition Number could not be computed”

Parameter estimates are still provided, but I’m not confident that they are meaningful.

Any suggestions would be highly appreciated.