Placebo test using an RD design

[Kathrin Me]

Hi all,

I have a question regarding the performance of a placebo test when using a regression discontinuity design.

With reference to Angrist and Pischk (2008) one can center the running variable around the cutoff to ensure that the treatment effect is still the coefficient to D_i. When performing a placebo test for a pseudo cutoff of +0.03 I conduct it as follows (Let's call the running variable x):

gen D = x > 0.03
gen xtilde = x-0.03
gen xtilde_2 = xtilde^2
gen xtilde_3 = xtilde^3
gen xtildeD = xtilde * D 
gen xtilde_2D = xtilde_2 * D
gen xtilde_3D = xtilde_3 * D
reg outcome D xtilde xtilde_2 xtilde_3 xtildeD xtilde_2D xtilde_3D, robust

My question is what to do if I want to test for a pseudo cutoff of -0.03? Do I then specify it as follows:

gen D = x > -0.03
gen xtilde = x-(-0.03)
gen xtilde_2 = xtilde^2
gen xtilde_3 = xtilde^3
gen xtildeD = xtilde * D 
gen xtilde_2D = xtilde_2 * D
gen xtilde_3D = xtilde_3 * D
reg outcome D xtilde xtilde_2 xtilde_3 xtildeD xtilde_2D xtilde_3D, robust

Thank you!
Best, Kathrin

[Øyvind Snilsberg]

To test presence of discontinuity above threshold at 0.03,

g above = x >= 0.03
reg outcome above x 1.above#c.x if x>=0

to test presence of discontinuity below threshold at -0.03,

g below = x >= -0.03
reg outcome below x 1.below#c.x if x<0

the reason for restricting the above (below) regression to treated (untreated) subjects is to avoid contamination from real treatment effects.

[George Ford]

It looks like you are estimating a spline function.

help mkspline