Robustness Check for 2SLS Estimation

[Ming Ding]

Hello,

I'm wondering if there is any regular method to do robustness check for 2SLS?

I also have another concern about relevance condition of IV. One of my profs told me I should run regression of outcome variables on control variables plus IVs directly, to see if there is direct effect of IVs on outcome variables. But I don't feel this right...I have never seen this done in any study.

Really appreciates any suggestions!

[George Ford]

You professor is telling you to run the reduced form equation. Standard fare.

Say Y is outcome, X is endogenous, and Z is IV.

Y = b0 + b1X + e

X = p0 + p1*Z + v

Y = a0 + a1*Z + u

Note that b1 = a1/p1.

If a1 is significant and p1 is not, then the IV is affecting Y directly and not through X. Not good.