Chaplin et al. (2017) tests the efficacy of regression discontinuity (RD) by comparing RD causal estimates at the treatment cutoff to those from Randomized Control Trials (RCTs) that are also estimated at this same cutoff. The study identifies 15 previously completed within-study-comparisons (WSCs) that explicitly examined this issue by assuming the RCT results are unbiased and then comparing them to RD results.

The differences between these results can be thought of as estimates of bias due to use of the RD method. The authors address the internal validity of RD by using the average estimated bias across all 15 WSCs. The study also addresses concerns about external validity by using meta-analystic methods to examine variation in estimated bias across studies. Existing theory predicts no difference between RD and RCT estimates on average, but difficulties with the implementation and analysis of RD in particular can lead to the possibility of deviations from theoretical expectations.