# Dealing with complete separation in Generalized Additive Models

I am modelling survival using a binomial mixed model (gamm from mgcv), but the model won't converge - I suspect because of complete separation in some observations. What are strategies to deal with this in the Generalized Additive Model framework?