We want to model \beta \sim \text{Normal}^+(\mu,\sigma), a Normal distribution truncated below at 0, where \mu and \sigma are fixed (and do not depend on parameters). Is the below what you recommend? So that things in the parameters
block have unit scale?
parameters {
real<lower=-mu/sigma> beta_raw;
}
transformed parameters {
real<lower=0> beta;
beta = beta_raw * sigma + mu; // want ~ N^+(mu,sigma)
}
model {
beta_raw ~ normal(0, 1);
}