By building a simple model I show that Regression Discontinuity Design estimates of incumbency advantage should be negatively correlated with partisan support in an election district if higher quality candidates are more able to take advantage of the office in terms of future election prospects. This negative relationship between incumbency advantage estimates and past party performance holds empirically in 11 out of 14 single member district election offices from 10 different countries. The RDD estimate of party incumbency advantage is higher in samples restricted to constituencies where a party had a negative winning margin in previous election and lower in constituencies where winning margin was positive. In order to show that this is not driven by period specific shocks, different measures of past party performance are used. In addition to that an open party list setting is exploited in order to have a quasi-random variation for candidate quality composition of the RDD samples. If a party wins more seats in an election constituency it has marginal candidates lower on the party list (i.e. lower quality marginal candidates) compared to marginal candidates where the party won fewer seats (i.e. marginal candidates are higher on the list and thus of a higher quality). Using Norwegian municipality election data negative relationship between number of seats party list gained and incumbency advantage is shown as predicted by the theory. Moreover this result holds when RDD is used to estimate the causal effect of change of number of seats to the incumbency advantage estimates.