Regression discontinuity designs aren't susceptible to regression to the mean. I'm not quite sure why, though. The best I can come up with is that regression to the mean is a phenomenon that occurs because you're dealing with a sample measured at different points, which have different underlying distributions at each point. With RD designs, however, you're measuring the difference in intercept along a column separating two smooth regression functions within a single bivariate distribution.
Can anyone shed further light on this?
