Hi,

yes that makes it clearer.

I am unfortunately a `data.table`

person, which means that I use `subset_draws`

to extract parameters of interest, which I then put into a data.table, before I use melt, cast, and other functions to calculate what I need. So I can’t help much with `gather_draws`

(which is from tidybayes) and the like.

Here is how you could do this with posterior and data.table:

```
theta_a1 =
draws %>%
subset_draws("theta | a_1", regex = T) %>%
as_draws_df()
%>% data.table()
theta_at_a1is0 =
theta_a1[a_1 == 0, theta]
theta_stats_by_a1 =
theta_a1[, list(mean = mean(theta),
sd = sd(theta)),
by = "a_t"]
```

The reason I asked about the joint estimation is that I was wondering how one would deal with the fact that one can’t calculate basic indices like Rhat for these conditional estimates. I also don’t see how one could use posterior or tidybayes functions to get statistics for conditional thetas. Lastly, I am uncertain how to deal with the fact that all alpha value will not have the same frequency in the posterior.