There's no such thing as binary instrument or binary treatment. Both instrument and treatment are continuums. Kirk, erase this sexist thread now!
Counting Defiers | Amanda E. Kowalski
-
Why didn't she try simulating data and checking to make sure her algorithm could recover the correct parameters? This would take, like, half an hour given what she must have already coded. If her algorithm works, such an exercise would demonstrate as much, and possibly help overcome the other glaring omission from the paper: why no attempt at an explanation of the intuition?
The acknowledgements suggest the paper was put up as an NBER wp without ever being presented, even internally, and without comments from anyone. The paper should be withdrawn, and this should act as a lesson for all of us: the economics culture of circulating work prior to publication really is beneficial, and the NBER wp series --- which is in effect a peer-reviewed publication with no peer-review --- should be overhauled or simply abandoned.