Ryan D. Enos and Benjamin E. Lauderdale, “Recovering Vote by Race in Primary Elections”
Scholars know far less about voting behavior in primary elections than in general elec- tions, yet in the many parts of the country where one party dominates, it is primary elections that determine who holds office. Because of the absence of the party cue, the roles of race, income and education in voting behavior are potentially more varied and consequential in the context of primaries. For example, elections contested by multiple black candidates–typically in majority-black electorates–can show striking patterns of racial voting despite the apparent lack of any overt cue from the race of the candidate. Unfortunately, studying primary elections is difficult because polling is sparse and of low quality. In this paper, we begin an assessment of whether the use of a local ecologi- cal inference method based on geographically-weighted kernel regressions can facilitate systematic study of these unpolled elections. Our local ecological inference technique weakens the assumptions made by standard ecological inference methods in an intuitive way that is appropriate to political geography. We demonstrate the relative robustness of our local ecological estimates of support by race in a difficult case where the correct answer is known and standard ecological inference methods give spectacularly wrong estimates: the state-level vote by race for Barack Obama and John McCain in the 2008 presidential election. We then apply these methods to a pilot study of the 2000 Illinois 1st Congressional District primary between Barack Obama and Bobby Rush.