Closed larryshamalama closed 1 month ago
I think what we can do is as follows: we can test if the Bayesian analysis is correct from a purely code perspective by including $U$ in the outcome model and see if our gamma and beta parameters converge (across a few simulations with appropriate burn_in
and n_samples
, maybe 10). This is already completed in this PR
Because of non-identifiability, it would be then important to investigate situations where we can better identify the treatment effect will converge to its true value. This can either be done using a restrictive data-generating procedure or, hopefully, using informative priors. This can be done in another PR
Closes #5