Instrumental Variable with Fractional Endogenous Regressor

[Maxence Morlet]

Hi all,

I know that Stata18 has a new built-in command - ivfprobit - for instrumental variable analysis with a fractional endogenous regressor.

Does anyone know of a command, potentially user-written, that applies fractional logit or probit to instrumental variable estimation in Stata17?

My particular problem is that I have the following endogenous regressor:

input float X
        0
        0
        0
        0
 .8068182
        1
        0
        0
        0
        0
     .125
        0
.14285715
        0
        0
        0
        1

(varying between 0 and 1)

and the following instrument

input float Z
         0
         0
         0
         0
  .1618742
         0
  .0608372
         0
         0
         0
 .04500082
 .04500082
         0
 .04500082

(also a fraction, comprised between 0 and 1)

The outcome is as follows:

input float Y
 0
 0
 0
 0
 0
-1
 0
 0
 0
-1
 1
 1
 1
 0

(ordinal dependent variable varying between -2 and 2, not only integers)


In the first stage, using the user-written command xtivreg2, I get out of bounds prediction, i.e. a coefficient superior to 1 on the regression of X on Z.

Therefore I am looking for an estimation method to bound the predictions in the first stage.

Thanks in advance!

[Maxence Morlet]

This thread is partially related to the following interesting thread: https://www.statalist.org/forums/for...ndent-variable.