Using margins with factor variables after xi: command (qreg2)

[Kartik Vira]

I am regressing an outcome (y) on two categorical variables (x1, x2), with standard errors clustered by a third variable (id). I am interested in the predicted values of y at each level of x2, with the average value of x1. I therefore run the commands:

reg y i.x1 i.x2, vce(cluster id)
margins i.x2, asbalanced

This produces the results I want and works without issue.

My problem arises when I try to get analogous results, but using quantile (median) regression rather than OLS. Because my standard errors need to be clustered, I have to either use bootstrap or the package qreg2. The margins command works as expected after bootstrap, but with my data bootstrapping takes an inconveniently long time to run. I had hoped to be able to use qreg2 instead. However, qreg2 is incompatible with new stata factor variables, and requires the xi: prefix. I am able to run the regression using qreg2, but I cannot use the margins command after estimation since this depends on factor variables. Specifically, I run

xi: qreg2 i.x1 i.x2, cluster(id)

which works successfully and gives the desired coefficients, but if I then run

xi: margins i.x2, asbalanced

I get the error message

variable _Ix1__ not found
r(111);

For a solution, I'd either like to know if there is a way to replicate the command

margins i.x2, asbalanced

in a way that is compatible with xi: prefixed estimation commands, or whether there is a way to run quantile regression with clustered standard errors that allows new-style factor variables.

[Richard Williams]

Are x1 and x2 0/1 binary variables? Or do they have more than 2 values? It is easier if binary but not impossible otherwise (I think). Something like

reg y x1 x2, vce(cluster id)
margins, asbalanced at(x2 = (0 1))

[Kartik Vira]

No, both x1 and x2 have more than 2 values.