Description:
In an RDD study of the incumbency effect, observations somewhat away from the threshold separating winners and losers in an election are necessarily employed. We consider how incorporating the vote volatility of elections into a preferred index of electoral competitiveness or closeness, in contrast to the often used unadjusted vote share margin, affects the estimated incumbency effect through this route for Liberal party candidates in Canadian general elections, with emphasis on the post-1950 period. Estimation is by local linear nonparametric regression with a data driven bandwidth. We also consider how allowance for the competitiveness and outcomes of prior electoral contests alters the estimated incumbency effect. Comparisons of our results with previous work on incumbency in Canadian elections by Kendall and Rekkas (2012) in this journal are presented, along with a reproduction of their model, for the cases we consider, based on combining our refined and extended electoral data with their (volatility unadjusted) index of electoral closeness and different estimation methodology.