New JRSSB @royalstatsoc.bsky.social paper by Yang Liu and me @mrc-bsu.bsky.social
We propose
- a general definition of ⩾2 “modules” in a DAG-structured Bayesian model by partitioning observations
- corresponding cut distribution(s)
- justification via KL divergence
Paper: doi.org/10.1093/jrss...
New at JASA @amstatnews.bsky.social by @hhau.bsky.social + me @mrc-bsu.bsky.social
–Directly specifying parameter priors is often hard
–Easier to elicit a predictive distribution for observables/other marginal
–Propose new algorithm to translate elicited predictive → parameter prior
doi.org/q37b