2SLS with a binary outcome variable

[William OBrien]

Hello, apologies if this has been posted elsewhere.

I'm working on a project utilizing a shift-share IV strategy with a continuous instrument and a continuous endogenous regressor, but my DV is binary. What would the best strategy be to estimate this model? For the second stage, should I be using a typical linear probability model or something like a logit/probit?

Thank you!

[George Ford]

try control function (cfregress). it will do all the work for you.

[William OBrien]

Thank you very much!

[Manh Hoang Ba]

The question only concerns second stage, which would involve considering LPM (the ivreg2 command) and Probit (the ivprobit command). I would like to estimate both models and compare the results to check robustness under different assumptions.

[George Ford]

ivprobit would work. should get roughly the same as cfregress. the second stage LPM is suitable as well, absent the need to predict.

[Manh Hoang Ba]

cfregress is applied to continuous dependent variables; when the dependent variable is binary, the cfprobit command should be used.

[George Ford]

yep. and cfregress gives you the LPM.