An extension of the regression discontinuity design to settings where a unit's treatment is an average or sum across many discontinuity events, such as an outcome measured at a higher level of spatial or temporal aggregation than the events, or spillovers onto neighbors. Two estimators are developed — an upper-level instrumental-variables regression using aggregated RD shocks as the instrument with aggregated local-linear controls, and a lower-level fuzzy-RD 'stacking' estimator over the discontinuities — and both identify the same local average treatment effect under standard RD continuity assumptions.
Regression discontinuity (RD) designs estimate causal effects by comparing units just above and below a cutoff of a running variable, assuming potential outcomes are continuous at the cutoff. This pa…