Hi @cluhmann,
Thanks for your prompt reply. Indeed, I could verify that ij from your code follow a normal distribution whose result is reasonable (centered on 4, probably because the prior ij is much larger than the observed 3).
I must conclude I misinterpreted the meaning of the log_likelihood field, from which I had read out total. It must be instead the actual log-likehood of the “total” variable, not the variable itself. Seems obvious now that I read your reply.
I was also confused by the warning:
The number of effective samples is smaller than 25% for some parameters.
Thanks much for your time replying, sorry for the kind of basic issue reported here. At least this gives me more confidence dealing with the actual, more complex issue I am set to resolve.