Inconsistency between rdrobust and rdplot results

[Ziyue Xu]

Hi everyone,
I use the rdrobust/rdplot package in stata to estimate a simple RD model with time as the running variable. But I found the results between rdrobust and rdplot are not consistent to each other.
For example, a is my Y variable, runningvar is my time running variable.

rdrobust a runningvar, p(1) masspoints(check)
Above command shows the estimate coeefficient is -394.72***

rdplot a runningvar, p(1) kernel(triangular) masspoints(check) nbins(200)
figure generated by above command is as follows

the magnitude and sign of the coefficient around the cutoff is totally not consistent to the rdrobust result -394.72***.

I tried to adjust kernel and nbins(##), but the results are not better. Could anyone give me some hints why such inconsistency exists between the rdrobust and rdplot commands?

I have put the sample data and my stata dofile in the following link, feel free to use the data to help me fix this issue:
https://drive.google.com/drive/folde...s3?usp=sharing

Thank you so much!!!

[Øyvind Snilsberg]

with -rdplot- you need to specify the bandwidth (also it should be -masspoints(adjust)-, not -masspoints(check)-),

rdrobust a runningvar, p(1) kernel(tri) masspoints(adjust)

local b_l = e(b_l)
local b_r = e(b_r)

rdplot a runningvar, p(1) kernel(tri) masspoints(adjust) h(`b_l' `b_r')

[Yazen Kashlan]

Hi Øyvind Snilsberg, I believe you mean

local h_l = e(h_l)
local h_r = e(h_r)

to get the bandwidths used to estimate the regression functions as opposed to the bias of the regression functions. Is that right?