Abstract
The estimation methods of regression discontinuity (RD) designs with a single assignment variable have recently been acknowledged to have a wide range of applications in social science. While treatment assignment is often determined by one threshold value, many empirical studies have shown the pervasiveness of RD designs with more than one assignment variable. Moreover, the literature has focused on the average treatment effect and overlooked the interesting perspectives provided by treatment effects at different quantiles of the outcome distribution. In this paper, we propose new approaches for RD designs with multiple assignment variables. The approaches allow nonparametric estimation and could be applied to estimating average treatment effects and quantile treatment effects. Based on our Monte Carlo simulation study, we suggest that the performance of the existing approaches is sensitive to the interaction terms in data generating processes as well as large variations in assignment variables. Our new approaches produce robust and more accurate estimates compared to the existing approaches with respect to all scenarios.