Treatment effects in Stata: Inverse-probability weighting