I usually look at both and then try to figure out any substantive
differences in results, but I'm generally partial to the coefficient
estimates from rreg (I often deal with skewed distributions where the median
is noticeably lower than the mean). On the other hand, I sometimes find
rreg's std errors estimates questionable.
An additional tidbit: Hamilton (Statistics with Stata, updated for version
8, p. 246) argues that "-rreg-, -qreg- and -bsqreg- deal comfortably with
y-outliers, unless the observations with unusual y values have unusual x
values (leverage) too." He argues that -rreg- tends to deal with such
situations better, because -qreg- tries to track such outliers while -rreg-
tends to just discard them.
