Visualization
- All RD estimates should be accompanied by a plot showing function estimates on both sides of the cutoff
- Include both local estimate and function estimates away from the cutoff, as well as intervals if possible
- In sparse data settings, these should overlay a scatterplot of \((R_i,Y_i)\) values to allow comparison of fit to raw data
- With larger sample sizes, may be hard to see much in data, so overlay binscatter instead (Cattaneo et al. (2021))
- Within narrow \(R\) bins, compute average \(Y\) value, plot for each bin, with width defined by levels or quantiles of \(R\)
- Usable as nonparametric estimator, but goal is a low-bias high-variance guide to patterns in data, so choose narrow bins that exhibit variability
- Example from Lee (2008) study of incumbency effect on re-election