Last updated on July 6, 2017
Back in January, I wrote two posts about dealing with imperfect instruments. One went over Conley et al. (2012), which deals with instruments that the authors deem plausibly exogenous; the other went over Nevo and Rosen (2012), which deals with instruments that the authors deem imperfect.
In a new working paper, Damian Clarke and Benjamin Matta discuss both methods, and they present the Stata commands –plausexog– and –imperfectiv– they created to implement them.
If you are dealing with an instrumental variable that can sort of, kind of meet the exclusion restriction but might not (in other words, almost any observational IV), I suggest checking out their working paper and, if you use Stata, the add-on commands they wrote.