Efforts to tweak model and tuning parameters to squash divergent transitions have not been successful. Let the following measure be tried to help solve this issue:
Simulate data from the model (logodds_glfp) of a size that permits faster fitting
20 groups, 100 units per group, censoring approx. 50%
"True" hierarchical distributions can be determined by setting hyperparameters to the prior mean (or at some other reasonable value, e.g. tau_ls1 between .1 and .5.)
Efforts to tweak model and tuning parameters to squash divergent transitions have not been successful. Let the following measure be tried to help solve this issue:
Simulate data from the model (logodds_glfp) of a size that permits faster fitting 20 groups, 100 units per group, censoring approx. 50%
"True" hierarchical distributions can be determined by setting hyperparameters to the prior mean (or at some other reasonable value, e.g. tau_ls1 between .1 and .5.)