In paper titled "Using Heteroskedasticity to Identify and Estimate Mismeasured and Endogenous Regressor Models", Lewbel proposes an approach based on heteroskedasticity to create an instrument when a good candidate is not available.
He also released the Stata command (ivreg2h) implementing his approach.
For example, in the model: Y1=a0+a1*Y2+a2*X+e，X is exogenous regressor, Y2 is endogenous variable,
the stata code is:
ivreg2h Y1 X (Y2=)
If I want to know how Y2 and M interact to affect Y1, and the model is: Y1=b0+b1*Y2+b2*M+b3*M*Y2+b4*X+e,
can Lewbel instrumental variable approach be used in IV estimation of regression modelsincluding endogenous interaction terms?
Are lewble(2012)'s assumptions violated because of the appearance of the interaction term between an endogenous and an exogenous regressor?
If I run the stata code:
gen MY2=M*Y2
ivreg2h Y1 X M (Y2 MY2=)
can the estimation results help recovering the true parameter estimate of the interaction term between an endogenous and an exogenous regressor?
Can anyone help? Thanks
He also released the Stata command (ivreg2h) implementing his approach.
For example, in the model: Y1=a0+a1*Y2+a2*X+e，X is exogenous regressor, Y2 is endogenous variable,
the stata code is:
ivreg2h Y1 X (Y2=)
If I want to know how Y2 and M interact to affect Y1, and the model is: Y1=b0+b1*Y2+b2*M+b3*M*Y2+b4*X+e,
can Lewbel instrumental variable approach be used in IV estimation of regression modelsincluding endogenous interaction terms?
Are lewble(2012)'s assumptions violated because of the appearance of the interaction term between an endogenous and an exogenous regressor?
If I run the stata code:
gen MY2=M*Y2
ivreg2h Y1 X M (Y2 MY2=)
can the estimation results help recovering the true parameter estimate of the interaction term between an endogenous and an exogenous regressor?
Can anyone help? Thanks
