Hi all,
I want to run an Unconditional quantile regression (RIF analysis) where my treatment variable is endogenous (and continuous), and I have a continuous Instrumental variable. I have engaged with the literature on Recentered Influence Functions (RIF) regressions to estimate unconditional partial effects. However, the issue is that most of the empirical work on RIF assumes exogeneity. I came across some stata commands, such as IVQTE to estimate unconditional partial effects, but they aren't applicable in my case because they require both the treatment and the instrument have to be binary. The most promising approach is the control variable approach to estimate unconditional partial effects. It's presented very well in the paper by Gosh(2016) https://papers.ssrn.com/sol3/papers....act_id=2765035 and he provides two empirical applications. I've reached out to him for some stata code, particularly for the second example from Angrist and Krueger (1991) which uses season birth dummies as a set of instrumental variables for schooling to estimate the heterogeneous aggregate returns to education. But he hasn't responded. I have performed the analysis on my own for OLS, 2SLS, QR and UQR, using Stata commands of regress, ivregress, sqreg and rifhdreg, respectively. I'd really appreciate some guidance on performing control variable unconditional quantile regression. My first impression was that I can run the first stage regression, save the residuals, and then perform UQR using rifhdreg whereby the residuals are included as an additional control variable in the outcome model. I'm not quite about my approach. Kindly assist.
So, in this last case for cvuqr, I ran the first stage, recovered the residuals, and then used these residuals as additional controls in the outcome model using rifhdreg.
I'm quite unsure as to whether this is correct. Please assist
Thanks
I want to run an Unconditional quantile regression (RIF analysis) where my treatment variable is endogenous (and continuous), and I have a continuous Instrumental variable. I have engaged with the literature on Recentered Influence Functions (RIF) regressions to estimate unconditional partial effects. However, the issue is that most of the empirical work on RIF assumes exogeneity. I came across some stata commands, such as IVQTE to estimate unconditional partial effects, but they aren't applicable in my case because they require both the treatment and the instrument have to be binary. The most promising approach is the control variable approach to estimate unconditional partial effects. It's presented very well in the paper by Gosh(2016) https://papers.ssrn.com/sol3/papers....act_id=2765035 and he provides two empirical applications. I've reached out to him for some stata code, particularly for the second example from Angrist and Krueger (1991) which uses season birth dummies as a set of instrumental variables for schooling to estimate the heterogeneous aggregate returns to education. But he hasn't responded. I have performed the analysis on my own for OLS, 2SLS, QR and UQR, using Stata commands of regress, ivregress, sqreg and rifhdreg, respectively. I'd really appreciate some guidance on performing control variable unconditional quantile regression. My first impression was that I can run the first stage regression, save the residuals, and then perform UQR using rifhdreg whereby the residuals are included as an additional control variable in the outcome model. I'm not quite about my approach. Kindly assist.
Code:
Here's a simple Stata code of that I'm attempting ***A simplified example of "The Returns to Education*** ***OLS*** regress lnwage educ year_of_birth place_of_birth ***2SLS*** using the last quarter of birth as an IV ivregress lnwage year_of_birth place_of_birth (educ = q4) ***CQR** sqreg lnwage educ year_of_birth place_of_birth, quantile(.1 .25 .5 .75 .9) reps(50) ***UQR*** ridhdreg lnwage educ year_of_birth place_of_birth, rif(q(`i')) vce(bootstrap, reps(500)) ***THIS IS WHERE I'M STUCK!: The control variable UQR regress educ q4 year_of_birth place_of_birth predict xHat, residual ridhdreg lnwage educ year_of_birth place_of_birth xHat, rif(q(`i')) vce(bootstrap, reps(500))
I'm quite unsure as to whether this is correct. Please assist
Thanks
