At Daily Kos Elections, we try hard to verify the accuracy of the Pres-by-CD numbers we give you by calculating the numbers ourselves. We didn’t do this in a few states where the legislature provided these numbers for us. Apparently, we should have—at least in Virginia.
In its calculations, the Virginia legislature, in addition to providing too few significant figures, apparently also failed to include absentee votes in its calculations. (As a humorous sidenote, this all came about because David Nir and I were looking into the Virginia results, which showed VA-10 being “50% - 50%” and us wondering who actually won the district.) From their perspective, it could be reasonable to exclude early votes—not much research has been done into how early votes can be allocated to a given precinct (or area) when early votes/absentees are presented in aggregate across the jurisdiction (this doesn’t excuse, of course, their exclusion of absentee votes even when an entire county is contained within a given district). We’ve ran into a similar difficult preparing data for the DRA, and of course several years back in the 2008 Pres-by-CD project.
At DKE, though, we are fairly adamant about splitting absentees and early votes properly—after all, an early or absentee voter votes for Congress, just like everyone else. We’ve been applying a fairly consistent set of methodologies, depending on how detailed early voting data is presented (or, for that matter, not presented; over the flip, we detail the exact methods that we’ve used). In Virginia, absentee votes are certainly important—there were about 500,000 such votes, and Barack Obama won 62 percent of them. (Put differently, without absentees, Obama won the state by 3.4 percent; he actually carried the state with a 6.3 percent margin.) We used both methods here to compare to the legislature’s numbers (and our attempt to reproduce the legislature’s numbers). At the district level, the impacts were as follows:
(The results for the subdivision method, which we're adopting moving forward, are available
here.) Based on the similarity of our calculations with no absentees and the legislature’s numbers, we’re fairly confident that that is indeed the exclusion in how the legislature performed its calculations. Further, there are substantial differences once absentees are added into the mix—for example, Scott Rigell is actually in an Obama district. (It does turn out that Obama won VA-10, even without absentees allocated.)
While we had been using the old set of numbers in our judgments of the race, these revised numbers will be in our information set moving forward. Further, we will also be verifying that a similar issue does not exist in the other states. This yields us two benefits—first, we can confirm or question the accuracy of numbers presented in other states, but second (and perhaps more importantly), we can show actual vote totals (and not just percentages), as well as breakouts by county. As always, any redistricting-related work that we do is available from our consolidated redistricting resources page.
Across the various states, we’ve used the same two methodologies, depending on whether early votes are presented at a county level (“county method”) or a sub-county level (“subdivision method”). (The one exception to this is Georgia, where we were given an extra piece of information—the number of people that voted from a given precinct, either early or on election day; this did affect our calculation method.) Otherwise, the guiding principle remains the same: At the level at which early/absentee votes are presented, allocate the votes for each candidate across the districts proportional to amount of election-day votes that that candidate received across the districts.
Put in an example, the county method does the following: When a county is split between New Districts A and B and, for example, 2/3 of Obama's precinct votes were cast in New District A (and 1/3 in New District B), 2/3 of Obama's absentee vote total gets allocated to New District A. The same is done, independently, for McCain's votes and third-party votes. The rationale underlying this is that the relative partisan leanings of the two district fragments are consistent between precinct votes and absentee votes, and that the relative turnout of the two district fragments are consistent as well.
The subdivision method is similar to the Countywide Method, but takes advantage of reporting of absentee ballots at a more detailed level. In Virginia’s case, this was by old CD. The county-wide method is repeated at the [county] x [old CD] level. Take the same county split between New Districts A and B above, but the part of New District A is composed of two distinct sections: Half that was in Old District A, and half that was in Old District B. (All of New District B was in Old District B.)
The two batches of absentees—one from Old District A and one from Old District B—are then treated separately. All of the absentees from Old District A are allocated to New District A, since all of the votes that were cast in Old District A would also have been cast in New District A. Old District B's votes are allocated proportionally across New District B and only the parts of New District A that were in Old District B.
The subdivision method is more "detailed," since the scope of the central underlying assumptions are made more narrowly—at the county-Old District combination level, rather than at the county level. In states that split precincts across districts, the Detail method does require additional assumptions, since votes cast from split precincts will need to be allocated between old districts first. Virginia does not split precincts across CDs, however, meaning the Detail method is strictly preferable here. Further, as you can see, the results do not differ materially under the two methods.