First, a huge Hat-Tip and Thank You to m247 for relentless ground reporting (and equally relentless fundraising… heh) making us all feel like honorary progressive Georgians for a few weeks. m247 introduced me to the runoff’s early voting summaries on Georgia Votes. Numbers seem legit vs. official numbers, and are presented beautifully and usefully.
Hailing from the statistician branch of political junkies and having some time on hand during the holiday, I took these numbers and merged them with the GA Prez results Wikipedia page, whose #s also seem legit.
Both sources have data at the Congressional district and county levels. I consider the former more useful since they are similar-size and also “closer to independent” than the 159 counties (although of course all of Georgia is interconnected).
We’ve seen in October-November how hard it is to read the tea leaves of early voting in an unprecedented pandemic year. But the runoffs provide a rare opportunity with a benchmark from the very same group — early voters — in the very same state and pandemic year.
As the headline image shows, regions with higher Biden % have generally come closer to matching their November early-voting tallies. But how much edge could that translate into, if any?
Well, a simple regression of the Jan/Nov early voting ratio vs. Biden % actually has a very useful interpretation:
It estimates how much more (in %) Biden voters have turned out.
This estimate comes down to nearly 16%. Per the model,
A hypothetical “100% Biden district” would early vote at ~82.6% of November early voting, while a “100% tRump district” would only get to ~66.7%.
This comes with a margin of error, of course. The nominal 95% confidence interval around the 16% goes from 8% to 23%. (If one accounts for lack of “perfect independence” between CDs, and adds a tiny bit of pessimism — then maybe 5% to 25%.)
Assume for now it’s 15%. Interestingly the EV and total voting #s in GA have been very round. In November the EV was almost exactly 4M and the final total was 5M. In January, the EV came out almost exactly 3M. Since the final Nov. Prez result was nearly 50:50 (well, with some going to other parties), these 15% translate to a ~5% net advantage in total actual votes for the final tally. Put some wiggle around them and you get the 3%-7%. Or 2%-7% if you want to extend the uncertainty in the pessimistic direction.
Still a sizable edge.
Now caveats (besides the usual stats ones):
- This is a temporary edge. If all those early-shy R voters (i.e., those who voted early last time but didn’t this time) do come out on Tuesday, then the edge is gone. It’s probably wiser to assume at least some of them will (half?).
- Dems are starting as underdogs in both races, I’d gauge it at -5% for Ossoff and -4% for Warnock. Remember, Ossoff was down 1.8% in Nov., and another 2.3% went to a Libertarian: closer ideology to Rs, and perhaps preferring a deadlocked Fed government over a Dem trifecta So it’s -3% or so, then I’d add at least -1-2% for the runoff effect which — if not for GA’s amazing new Dem turnout operation — would have already had us buried.
Anyway, I did the same analysis at the county level (plot below); it yields a somewhat lower estimate of 14%, still in the ballpark though.
UPDATE: fixed the previous version of the county plot, which had inadvertently excluded 32 counties. h/t ukga in the comments!
Clik here to view.
