Oaxaca with instrumental variable

[Adi Weider]

Dear all,

I am trying to use an Oaxaca- Blinder decomposition to see if a variable of interest contributes to the gender wage gap.
However, I would like to use an instrumental variable in place of my variable of interest in the decomposition.

1) Is there a command that combines the oaxaca command with use of an IV?

2) Alternatively, can I create the instrument variable and then put it directly instead of my original variable in the oaxaca command? If I do this, I suppose I will have to somehow change my standard errors to correct for the fact that my instrument is estimated to begin with. How might I change my standard errors?

3) Is there a reference to a paper that has done something similar? Does anyone know if this is even econometrically sound and if yes/ no, why?

Many thanks!

Best regards,

Adi

[Dylan Webster]

Hi Adi,

Its been a few years but I am looking into something similar. Have you ever found out the answers to these questions?

Kind regards,
Dylan

[Frode Andre]

Here is a post on a similar topic ("Instrumental variable decomposition): https://www.statalist.org/forums/for...-decomposition

[George Ford]

This straps on Jann's oaxaca decomp calculations to ivreg2.

Not ssc approved. Built with Claude Code. Use at your own risk. I've performed a crude test to check the calculations and it replicates Jann. Also ran in python and it replicates.

Let me know if you run into a problem.

If I can find the time, then I'll go through it and clean it up any errors and verify.

clear all

use http://fmwww.bc.edu/ec-p/data/hayashi/griliches76.dta

** ivreg2 results
ivreg2 lw s expr tenure rns smsa (iq=med kww age mrt)
estat summarize

** straight oaxaca
oaxaca lw s expr tenure smsa  iq , by(rns) omega detail vce(robust)

** check the calculations
cap drop u
g u = iq + rnormal()/200
ivoaxaca lw s expr tenure smsa  (iq = u) , by(rns) omega detail vce(robust) ivopts(small)

** ivoaxaca
ivoaxaca lw s expr tenure smsa  (iq = med kww age mrt), by(rns) ///
   omega detail vce(robust) ivreg2 first

** bootstrap
bootstrap, reps(500) strata(rns) seed(123): ///
ivoaxaca lw s expr tenure smsa  (iq = med kww age mrt), by(rns) ///
   omega detail vce(robust) ivreg2

[George Ford]

I'll add factor variables and e(sample) fix.

[George Ford]

factor vars and e(sample) fix.