From the teffects documentation in Stata 14, I see that the margins command may be used to calculate the conditional ATE or ATET for a subpopulation for etpoisson and etregress.
Both etpoisson and etregress also take complex survey weights (using svy, pweight, etc.). However, the margins post-estimation command does not. It seems like my options are to re-parameterize etregress as needed, or ignore the biased standard errors.
Specifically, I want to model a treatment - covariate interaction (treatment * income). The example on page 49 of the documentation shows how to calculate the interaction, but does not use the complex survey weights.