Hi, I have a question about the use of Regression Discontinuity Design (RDD) in economics research. In papers I’ve read in AER or AEJ that employ RDD, I’ve noticed different approaches to determining the bandwidth. Some papers use the same bandwidth across all outcome variables, determined by the first outcome variables, even when exploring potential mechanisms and different outcome variables later on. Others apply a different bandwidth for each outcome, often using methods like rdbwselect.
However, I’m concerned that changing the bandwidth might lead to different subsamples being analyzed for each outcome, which could affect the consistency of the results. Which approach is more appropriate? And if we are comparing two heterogeneous groups, should we use the same bandwidth for both groups to ensure a fair comparison, or should we adjust the sample range of the running variable for each group separately?
Any guidance would be much appreciated!
However, I’m concerned that changing the bandwidth might lead to different subsamples being analyzed for each outcome, which could affect the consistency of the results. Which approach is more appropriate? And if we are comparing two heterogeneous groups, should we use the same bandwidth for both groups to ensure a fair comparison, or should we adjust the sample range of the running variable for each group separately?
Any guidance would be much appreciated!
