I was wondering if it's possible to perform a 2SLS regression where I first run a regression for my endogenous variable, obtain the predicted values and then use them in my second stage regression (a logit)?
Would this approach yield consistent estimators?