Hi,

I have a hierarchical model with a beta prior for a vector of subject-level parameters, with hyperparameters w1 and w2.

I am trying with 4 chains and 15000 iterations (14000 for warmup) and still Rhat ~ 1.3 for w2.

Which is the recommended prior for the probability parameter xi, bounded between 0 and 1, to avoid sampling problems in a hierarchical model?

Best regards, Benjamin

```
data {
int<lower=0> N_SUBJECTS;
int<lower=0> N_TRIALS;
int<lower=0> choice[N_TRIALS, N_SUBJECTS];
real k[N_TRIALS, N_SUBJECTS];
}
parameters {
vector<lower=0, upper=1>[N_SUBJECTS] xi;
real<lower=0> w1;
real<lower=0> w2;
...
}
model {
w1 ~ cauchy(0, 10);
w2 ~ cauchy(0, 10);
xi ~ beta(1/w1 + 1, 1/w2 + 1);
for (t in 1:N_TRIALS)
for (s in 1:N_SUBJECTS)
choice[t, s] ~ bernoulli(Phi_approx(k[t, s]-m[s])*
(1-xi[s]) + xi[s]/2);
...
}
```