GEPWREG: Stata module for percentile weights regression: the effect of the covariates at a quantile of the outcome

gepwreg estimates the effect of the covariates on an outcome for the units located at a target quantile of the outcome, at the initial quantile (the outcome before the contribution of chosen covariates), or along the rank of an observed covariate, with the percentile weights of Araar (2016): Gaussian kernel weights on the percentile scale. The effect at a quantile of the outcome is estimated in two steps: unit-level effects from a heterogeneity model fitted on the whole sample (a conditional quantile regression, or interactions with observed variables) are averaged with the percentile weights on the rank of the outcome, and the estimator converges to the unconditional quantile partial effect. Standard errors come from the influence function of the whole procedure, with Taylor linearisation under a complex survey design declared by svyset, or from a bootstrap that re-runs the whole procedure; the bandwidth is the MSE-optimal plug-in. ref() gives the effect for a given household, which holds the composition of the group at the quantile fixed, and graph() draws the profile of the effects across percentiles with a simultaneous confidence band and a test that the effect is constant across percentiles. merr corrects the interaction model for classical measurement error in the heterogeneity variables. gepwreg_setable compares the standard errors; a dialog box is included, and the examples of the help run from clickable links (gepwreg_examples). The paper that documents the estimators is included as an ancillary file (gepwreg_paper.pdf).