I’m currently working with the following model structure as part of a mixed effect multilevel survival model with group level predictors
The model fits decently well, but when I start to scale it up an include more units (where a unit is called ‘style’), the models efficiency starts declining dramatically.
Looking at the pair plot below, it looks like the model has trouble exploring higher values of log lambda. My guess is that as I include more units, there end up being more cases of extreme log_lambda values and the MCMC sampler has more and more trouble exploring the posterior - any thoughts on how to improve that?
