Eight Schools example: Questions

Dear all,

just getting my head around STAN, and came across the Eight Schools example as described here [1] and have some questions:

  1. Coming from the BUGS language, I wonder about the advantage of the “target +=” notation vs “~” sampling notation?
  2. I would have expected that in the “standardized parametrization” also the overall treatment effect parameter mu and the parameter eta would require the definition of a prior distribution, but these are missing. How does Stan deal with this, since the model still works? Where is this behaviour documented?

Kr, SR

[1] RStan: the R interface to Stan

Differences between target += and ~ are discussed here

Not specifying a prior over a parameter means that stan will treat it like the random variable is uniformly distributed. This will be an improper prior unless you bound the parameter in the parameter block. I can’t find the exact place in the Stan reference manual it exists today, but here is a StackExchange quoting an earlier version of the manual.

1 Like