Over at the interestingly named Marc F. Bellemare.com, there is a pro-IV post.
A very good point is actually raised, namely that in some very cool cases you can get random assignment of the instrument. Here I totally agree that you are on solid ground.
I can actually argue further against my former self and point out that Fuzzy RD models ARE IV models.
Whoever writes at Marc F. Bellemare.com also appears to somewhat agree with me saying that,
Don’t get me wrong: If you are going to use an observational IV, you do need to think very carefully about how and why it meets the exclusion restriction. And if it does meet it, you need to pray that it will be a relevant IV. But there are clear cases where IV works, and that is especially the case in a setting where you randomly assign the IV, or in quasi experimental settings where people are assigned to some treatment at random (e.g., Angrist’s famous Vietnam draft lottery setting).
Again, I agree these are clear cases. But they are a tiny minority of the cases where IV is used.
Look at a typical dynamic panel paper. it uses a test for no second order autocorrelation, generally accepting if the P level is worse than 0.10, so all variables lagged twice or more can be instruments. Then a second test, Sargan or affiliated, of OVERidentification again accepting the null with a P worse that say 0.10, and then claim to have validated their identification strategy.
Two consecutive filters with little to no power to fail to reject a false null, a test that doesn't test what you are claiming, and voila, SCIENCE.
In other news, Me and Mungowitz are looking into legally changing our blog's name to Marc F. Bellemare.org.
Wish us luck!