Ok, rant time about #matching. I’ve seen a lot of econ people diss matching, and it was something somewhat repeated on my different methods classes in different institutions. So I’ve got stuff to say: 1) Stop confusing matching w/ an identification strategy. 1/n
It’s an adjustment method. Of course it’s a huge stretch to think you can recover causal effects *just* by matching. Just as it’s ridiculous to think you can recover those effects by using OLS. However, if you layer it on top of an actual sound identification strategy 2/n
(e.g. natural experiment, IV, freakin’ RCTs) you can discover a whole world; 2) It provides a lot of advantages over other parametric methods that people seem to just not even think about (not gonna extend on this here, you can read the intro on my paper w/ JRZ & @J_P_Vielma).3/n
3) And I can’t stress this enough, not ALL matching is propensity score matching. Solvers have improved, computers are faster, you can do so many things! Balance covariates directly! The world is your oyster! 4/n
5) Finally, should you always use matching for adjustment? Probably not, just as you should probably give a bit more thought before using any adjustment method. But please, for the love of god, if you take anything out of this rant, let it be point (1). Rant over 5/5
