Regression discontinuity designs (RDD) are widely used in the social sciences to estimate causal effects from observational data. Scholars can choose from a range of methods that implement different RDD estimators, but there is a paucity of research on the performance of these different estimators in recovering experimental benchmarks. Leveraging exact ties in local elections in Colombia and Finland, which are resolved by random coin toss, we find that RDD estimation using bias-correction and robust inference (CCT) performs better in replicating experimental estimates of the individual incumbency advantage than local linear regression with conventional inference (LLR). We assess the generalizability of our results by estimating incumbency effects across different subsamples and in other countries. We find that CCT consistently comes closer to the experimental benchmark, produces smaller estimates than LLR, and that incumbency effects are highly heterogeneous, both in magnitude and sign, across countries with similar open-list PR systems.