Coefficient Matrix in Unconditional Quantile Regressions

[Micah Baum]

I am running models with the ivqte command, which estimates unconditional quantile regressions, as described by Frolich & Melly (2010). ivqte appears to only store the treatment variable coefficient in the coefficient matrix, and not any of the control variables. Here is an example piece of code:

ivqte hhwealth_h_post14 (will), q(.25 .50 .75) variance ///
continuous(hhwealth_h_pre14 agedeath sagedeath children d_parents d_siblings)

when typing ereturn list, one sees that the coefficient matrix only stores three values: "e(b): 1 x 3" storing one coefficient at each quantile (25th, 50th, 75th). This contrasts with my conditional quantile regression models, which use the qreg command and store all coefficients in the estimates matrix. For the above specification, if I were to instead use qreg, ereturn would show the matrix as "e(b): 1 x 8".

Is there a way to identify or store estimates of all explanatory variables included in ivqte?

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

After most procedures, you can issue
return list
ereturn list
sreturn list

and you'll see what has been retained from the estimation. Most procedures retain the coefficient matrix. You may also be able to access them individually using _b[x] format.

You can also open the ado file (search for ivqte on your c: drive or whereever) and modify what is retained.

With a user written program, you may need to contact the authors.

[Blaise Melly]

Dear Micah,
You have used ivqte without a varlist such that the estimator suggested by Firpo (2006) has been used. This estimator uses the covariates to reweight the observations but they do not enter directly into a quantile regression. Therefore, there are no quantile regression coefficients to report for these covariates. If you use ivqte with a varlist, e.g. ivqte hhwealth_h_pre14 agedeath sagedeath children d_parents d_siblings hhwealth_h_post14 (will), q(.25 .50 .75) variance, then the vector e(b) will contain the coefficients for all the covariates (they will be numerically identical with the coefficients reported by qreg). But the models are different; you should have a look at our Stata Journal article if you want to have more explanations.
Best regards,
Blaise