To be specific, I want to conduct an IV regression in stata with the following three equations.
EQ1: Y1 ~Y2 Y3 X
EQ2: Y2 ~ Z1
EQ3: Y3 ~ Z2
Both Y2 and Y3 are endogenous. However, Z1 and Z2 are not the same set of exogenous variables.
In stata, all exogenous variables will by included in the first step estimation with the command ivreg. However, I want to estimate the reduced form of Y2 and Y3 with different sets of IVs. Does anyone know how I can achieve this goal? Thanks a lot.
Dear Joy Yang ,
I think you can easily estimate it as follows:
ivregress 2sls Y1 X (Y2 Y3 = Z1 Z2)
Here you have an example you are able to reproduce:
syuse auto, clear
ivregress 2sls price mpg trunk (weight turn = displacement length)
Best,
José-Ignacio
You can do this in two ways:
(1)
Stata command for IV 2sls
ivregress 2sls Y1 X (Y2 Y3=Z1 Z2)
(2)
You can also estimate with OLS.
Stage 1:
Y2=a0+a1Z1 +Control Variables
Y3=b0 +b1Z2 +Control Variables
Stage 2:
Y1 =p0 +p1Y2_hat+p2Y3_hat +Control Variables
Where ai, bi, and pi are the estimated parameters. i=0,1,2,...
Hope you find these information helpful.
I am not aware of a built in Stata command that will achieve what you want. I think the recommendations above will include all exogenous variables in your first stage.
You could run reduced form regressions for Y2 and Y3 and use the predicted values in your Y1 regression. The standard errors will be incorrect, but the coefficients should be what you are after.
I am more concerned about why you want to do this. The reason there is no built in Stata command is because it is not a good way to use IV. The danger is that Z1 and Z2 may be correlated with one another, and this may make both your coefficient estimates bias in the IV regression.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression