PSA: Do not do 2SLS mindlessly in the setting of binary outcomes.
I think the underlying issue is related to the “non-collapsibility” of odds-ratios. My understanding is that a non-linear approach or g-estimation would be the way to move forward. As of July 2024, I have not found a solid Bayesian approach to this problem. The package looks promising, but does not handle cases with binary outcomes yet. If anyone out there has this figured out, please let me know.