Predict on unseen group in hierarchical model

Thank you very much for your reply @lucianopaz.

If I understand correctly, this is the explanation why I needed to specify > 1 instances. However in my last post, I had only used 1 instance and this did not lead to an error.

Also, do you have an idea why the posterior predictions don’t seem to pick up the learned sigma_a from the trace? (Described in my last post)