Hello,

After reading through online sources including here with @Jeff Wooldridge, Richard Williams and Joro Kolev Kolev, there are a few commands that handle % variables (0 to 1) as dependent variables for IV estimation. For instance cmp and fracivp. However these don't seem to accept more than one endogenous var-and I have two x and x^2.

I have two questions. y independent 0->1=% var, x endogenous=% variable, z instrument =% var as well

(1) Would it be incorrect to estimate the IVprobit with a fractional dependent variable. I first estimate residuals via control function for my instrument:

reg x z $other_controls, robust cluster(panel_var)
predict double myres, resid
gen myres2=myres^2

ivprobit y $other_controls (x x2=myres myres2), vce(cluster panel_var)


(2) Alternatively, would it be better to use glm and create a more traditional control function:

glm y x x2 myres myres2 $other_controls , fa(bin) link(logit) cluster(panel_var)

In this case, do I need to include myres2?

Thank you very much,

D