No announcement yet.
  • Filter
  • Time
  • Show
Clear All
new posts

  • Graph Regression Discontinuity with controls

    Dear all,

    i'm implementing a regression discontinuity design. The sample is not stricly balanced, so I am controlling for the covariates significantly unbalanced between treatment and control groups (Calonico et al., 2018). When producing graphs I run the following command:

    rdrobust depvar runvar, c(0) p(1) covs(covars) kernel(triangular) bwselect(mserd) vce(cluster clustervar) all
    local bw=e(h_l)
    rdplot depvar runvar if inrange(runvar, -`bw',`bw'), c(0) kernel(triangular) genvars nbins(10 10)
    tw (scatter rdplot_mean_y rdplot_mean_x) (lpolyci depvar runvar if runvar<0&inrange(runvar, -`bw',`bw'),fcolor(none) degree(1)) (lpolyci depvar runvar if runvar>=0&inrange(runvar, -`bw',`bw'),fcolor(none) degree(1)), xline(0)

    Yet, the above - other than for the local bandwidth - does not account for the covariates. I am wondering how Stata can draw the linear regression line (and the associated lines of the confidence interval) adjusted for my covariates.

    Many thanks,