Dear All,

I am using the ivreg2h user-written routine, which implements the procedure in Lewbel (2012) "Using heteroskedasticity to identify and estimate mismeasured and endogenous regressor models". I understood the procedure of the estimator (this presentation is pretty useful indeed). So basically, the instruments for the endogenous variable are created internally as the product between the centered-around-the-mean exogenous regressors and the error term (the latter being the vector of residuals from the ‘ﬁrst-stage regression’ of each endogenous regressor on all exogenous regressors, including a constant vector). Typically the first stage regression looks like this:

Obviously 00000A,

I am using the ivreg2h user-written routine, which implements the procedure in Lewbel (2012) "Using heteroskedasticity to identify and estimate mismeasured and endogenous regressor models". I understood the procedure of the estimator (this presentation is pretty useful indeed). So basically, the instruments for the endogenous variable are created internally as the product between the centered-around-the-mean exogenous regressors and the error term (the latter being the vector of residuals from the ‘ﬁrst-stage regression’ of each endogenous regressor on all exogenous regressors, including a constant vector). Typically the first stage regression looks like this:

Code:

First-stage regressions ----------------------- First-stage regression of pc: Statistics robust to heteroskedasticity Number of obs = 125 ------------------------------------------------------------------------------ | Robust pc | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- __00000A | -.0442455 .1093248 -0.40 0.686 -.2607383 .1722473 __00000B | .4788744 .7709396 0.62 0.536 -1.047796 2.005545 __00000C | 1.243266 1.387136 0.90 0.372 -1.503641 3.990173 y80 | .7705854 .2422635 3.18 0.002 .2908377 1.250333 lnwi | .2088213 .6806487 0.31 0.760 -1.139048 1.556691 lnnda | -6.197757 1.553207 -3.99 0.000 -9.27353 -3.121985 _cons | 12.07512 5.223086 2.31 0.023 1.731989 22.41825 ------------------------------------------------------------------------------

00000B and 00000C are the generated instruments, being 3 since 3 are the exogenous regressors. My question is: which of the exogenous regressors 00000A refers to? Is it the instrument constructed by using the exogenous regressors y80 (as it is the first exogenous regressors in the list)? What I cannot understand is the order of generating the instruments, whether the routine uses the order of appearance of the exogenous regressors in the table or any other logic. Knowing it, I can figure out if 00000A is referring to y80, 00000B is referring to lnwi and so on.

I hope my question is clear. Thanks in advance for your help.

Dario

I hope my question is clear. Thanks in advance for your help.

Dario