Information criteria - conflicting advice for hierarchical models

Thanks, this is useful. When you earlier said just “hierarchical models”, I just guessed the most common simple hierarchical model. The network in you model makes A, B, C, D non-exchangeable and cross-validation more difficult. Could you provide a bit more details? Different levels you mention are

  • in general no individuals
  • arms
  • trials
  • treatments (A, B, C, D)
  • studies

Can you tell a bit more how these are related (I can guess, but I’m not expert on these kind of studies, so I could make mistakes) and how many each of those you have, and which are assumed to be exchangeable given some other information?

1 Like