A regression discontinuity design (RDD) is a quasi-experimental design where treatment assignment is made by treating participants who meet or do not meet a cutoff on a numeric assignment variable. It requires that data is available on the assignment variable and the outcome variable for both participants and non-participants. A common problem that prevents a regression discontinuity design is that researchers may have scores on the assignment variable for both participants and non-participants, but only have outcomes for participants. An RDD is a powerful design because the mechanism of assignment is known (that is, the assignment variable), and therefore, the local average treatment effect (LATE) can be estimated without confounding. However, the RDD provides less power to detect a treatment effect than an experimental design. Furthermore, the LATE only generalizes to individuals around a specific cutoff.
example regression discontinuity analysis
| simulated_rdd_data.rdata |
| guide_to_analysis_of_regression_discontinuity_design.rmd |