Matching and weighting

Tricks for handling a non-random sampling problems to adjust for the causal graph . Common in e.g. survey data.

MRP, a.k.a. Mister P, is one method for correcting for non-response bias and other such sampling nastiness. Dan Simpson explains MRP with structured priors.

See also the generalized version, RPP.

What I can do is link to my reading list of examples and explainers. Bob Carpenter’s worked example is interesting reading in this context, explaining Thomas Lumley’s post.

There are implementations also by Adam Haber using Tensorflow Probability and Lauren Kennedy and Jonah Gabry in rstanarm.