Hi Ben, it took me quite a while to answer and I apologize for that. thanks for your helpful suggestion!

I just listened to your you tube talk about rstanarm and then realized and applied what you suggested here.

up to now I focused my efforts in writing a the model in Stan, which I uploaded here today (bellow) and hope to receive some feedback on it (my group-level params have low n_eff). I took your suggestion and wrote the following model in rstanarm, and no such problem occurred.

```
options(mc.cores = parallel::detectCores())
post <- stan_glmer(Resp13 ~ (Distance * Context ) +
(Distance * Context || Subject),
data = myData, family = binomial(), QR = TRUE)
```

comparing my model and the rstanarm model, the apparent differences are:

- I had low n_eff for the group parameters and the rstanarm model didnāt have them. all params had high n_eff
- I used one sigma for all betas and rstanarm model implemented sigma for each interaction between group params and subject params (is that true? can you explain that?).
- yet, the posterior predictive check looks very similar

Do you have any suggestion regarding these essential differences? and how should I improve my model? in parallel I will work with rstanarm, which is very efficient for meā¦

again, many thanks !