Dear all statlists,

I have a problem when I try to interpret my results from a regression discontinuity. I am using Stata 16 at a Mac. I am using the rdrobust plugin, with uniform kernel and a fixed bandwidth of 25000 and two controls. I have a hard time interpret the results as they show a conventional negative and significant, effect while at the same time show a positive, and significant, robust effect. Does anyone know why this might be the case?

I tried with several placebo-cut offs and they all seem to be significant in a different direction so I do not think that my results are any stable, but this switch between significant positive and negative between conventional and robust I've never seen before.

Code:

Code:
rdrobust percentage_sup distance, kernel(uniform) h(25000) covs(DIST_FDR turnout) p(1)

Covariate-adjusted sharp RD estimates using local polynomial regression.

      Cutoff c = 0 | Left of c  Right of c            Number of obs =    5781202
-------------------+----------------------            BW type       =     Manual
     Number of obs |      3081       10204            Kernel        =    Uniform
Eff. Number of obs |      1785        2265            VCE method    =         NN
    Order est. (p) |         1           1
    Order bias (q) |         2           2
       BW est. (h) | 25000.000   25000.000
       BW bias (b) | 25000.000   25000.000
         rho (h/b) |     1.000       1.000

Outcome: percentage_sup. Running variable: distance.
--------------------------------------------------------------------------------
            Method |   Coef.    Std. Err.    z     P>|z|    [95% Conf. Interval]
-------------------+------------------------------------------------------------
      Conventional |  -2.827       .369   -7.6615  0.000   -3.55027     -2.10383
            Robust |     -          -     2.9830   0.003    .716598      3.46233
--------------------------------------------------------------------------------
Covariate-adjusted estimates. Additional covariates included: 2
Don't mind the unusually high number of observation, that is a bug in the plug-in. It doesn't affect the results.