Yesterday's shocking defeat of House Majority Leader Eric Cantor by Randolph-Macon professor Dave Brat has raised several questions, one of which I will address here. How much did Democratic efforts to defeat Cantor actually contribute to his loss? There were several public encouragements for Democrats to vote for Brat, as there was no Democratic primary district (fellow Randolph-Macon professor Jack Trammell was nominated Monday), including one by former Georgia congressman Ben "Cooter" Jones. Brat won (without provisional votes counted) with 55.55% of the primary vote, a large 11.1% margin. It is an absurd notion to think that 11.1% of the primary electorate was composed by Democrats, let alone ones who would solely vote for Brat (as many moderates and liberals prefer Cantor's extremism to even further-out extremism, regardless of the black eye that Cantor's loss gives the Republican Party). It is, however, feasible that a small number of politically informed Democrats did help add to Brat's margin of victory. As I will present below, I believe there is significant evidence to show that Democrats added at least 0.4% to Brat's vote share, but I concede that there is almost assuredly a small but undetectable extra addition to Brat's margin. While I have no empirical evidence for this number, given the small 0.4% margin that I find with two measures of vote-skewing, I believe that that undetectable number is somewhere around 0.1%.
My data is gathered from the Virginia Board of Elections website; they have better than usual transparency with election statistics. I use results at the precinct level of Congressional District 7; they exclude the 1475 counted absentee votes, as they are reported on the county level, not the precinct level (without these votes, Brat's vote share increases to 55.62%). Three pairs of precincts were merged between the 2012 and 2014 elections and are thus combined in my data: Chesterfield Co.'s Midlothian and Midlothian North, Goochland Co.'s Centerville and Shallow Well and Louisa Co.'s Elk Creek and Fredericks Hall. Excluded from my graphs below are Spotsylvania Co.'s Riverbend and and Summit precincts, whose 57 registered voters in their District 7 portions cast zero votes in the primary.
Evidence of Democratic crossover voting for Brat
Red: solid Obama (55%+), solid Brat (55%+)
Green: solid Obama (55%+), not solid Brat (<55%)
Blue: not solid Obama (<55%)
Without 55%+ Obama precincts: 55.24% Brat
All precincts: 55.62% Brat
Difference: 0.38% for Brat
It is clear that strongly Democratic precincts were much more consistent in giving Brat a large share of their vote. As you would expect Democratic voters to prefer the less extreme Cantor and non-Democratic voters to be more moderate in Democratic areas than Republican areas, it seems that this 0.38% swing is no coincidence and is the result of Democrats trying to defeat Cantor. Given the reasons I just mentioned, it is possible that this margin would be slightly larger if Democrats voting for Cantor were discounted, but that does not add to that 0.38% because of the Democratic Cantor voters' countering effect.
This compares the vote shares of Obama and Cantor's 2012 general election opponent, Wayne Powell, with Brat's. As you can see there is little difference other than Powell performing uniformly worse by a few percent in essentially all precincts (not true in some of the most Republican and white areas).
2014 Republican primary turnout as a percent of Romney's vote total in 2012: 31.38%
Actual result: 55.62% Brat
Normalized result if all precincts had turnout of 31.38% of Romney votes: 55.19%
Difference: 0.44%
It is clear that, at least in a few very Democratic precincts, significantly more votes were cast than would be expected given the 2012 presidential results. This is a small number of voters, but it is also clear that the most swingy precincts were much more likely to have average turnout (by this measure) than more conservative areas. I am therefore fairly confident that this 0.44% difference is explained in part by Democratic voters. I can see two potential caveats, one being that Republicans who bother to vote in very Democratic areas may be more motivated voters and the other I will mention below.
Showing how small the Democratic voters' effect was
It is clear from this that Brat performed best in areas that gave President Obama less than 38% of their vote and that the slight impact of strongly Democratic areas was very small and hardly distinguishable from statistical noise.
So how do you explain this (somewhat) seemingly conflicting evidence?
Turnout is the key. As this is a Republican primary, even with Democratic efforts to influence the results, Republicans were still much, much more likely to vote. If turnout was normalized to the districtwide 12.56% turnout, you might expect Brat to improve with the greater turnout in Democratic areas, but that isn't true. His vote share would decline from 55.62% to 54.96%, primarily because of the caveat I alluded to earlier: Brat benefited significantly from high turnout in areas of Hanover County. If my estimate of 0.4-0.5% of Democratic influence are correct, the higher turnout in Hanover was more than three times more important for Brat than were the votes of Democrats. Note that there is little difference whether using the Board of Education's numbers for 'active voters' or registered voters.
Any thoughts? Anything I missed? Any other simple analyses?