Thread by @gelliottmorris

🎉Polls are closed in Virginia!

I am running a live election-night model for the governor’s race (similar to the NYT “needle”) and will be posting updates in this thread as results pour in. The model turns on once ~20 precincts report e-day votes & ~5 counties report absentees.

The state elections website (which is publishing statewide returns publicly) is really taking its sweet time. Lagging the networks by quite an amount. But with 3 absentee dumps and results from 107 precincts, reporting jurisdictions show a small shift toward Dems v 2020.

I imagine the model will switch itself on any moment now and then I can send you a screenshot

"any moment now"

The hangup is that we don't have full absentee votes anywhere yet. The model needs or else we're projecting a strong pro-McAuliffe swing in election-day votes to absentees which isn't correct. Need to observe the polarization in both types.

Ok, we're running now! The model isn't going to get information from late mail votes tonight, so it won't be able to call a close race if those are a big chunk, but I'm going to try ignoring them for now and assuming swings in partisan splits are roughly equal statewide

The lay of the land right now is that McAuliffe is underperforming in absentee votes, both in terms of vote share and turnout, and it's enough to cancel out (by a large margin) any increase in Dem election-day vote versus 2020. The promise dashboard chart is coming momentarily

Here's the full model projection now. It thinks Youngkin is on track to win by 2-3 points **though there is large margin of error.** Things typically pan out as expected, but we need to see more election-day precincts from the cities and suburbs (esp in and around Fairfax

The model is stabilizing around Youngkin +3 with 32% of the vote in. Things are close enough and there are enough votes left in Fairfax where McAuliffe could win, but the model sees a pretty likely (70%) R win in Virginia tonight. And a swing to Rs of at least 10 pts either way.

Just to give you an idea of where I'm at qualitatively, I am currently writing our post-VA piece as if Youngkin has won

The only reason my model is not at 100% Youngkin probability now is because there is lot of uncertainty in the absentee ballot predictions as most have not reported yet, especially in the big cities. But the writing is on the wall here. It looks like Rs will go 2-3/3 in VA.

I don't have time to clean up these plots right now, but the story in Virginia is a clear drop-off in turnout in the strongest Democratic areas (esp in absentee voting) and a stronger-than-average swing against them in the exurban places where they have lost ground since ~2012

One thing making precinct-level analysis in VA reallllyyy hard right now is the splitting up of votes at the absentee and election-day level. Makes e-night prediction easy, but muddies up comparisons across years when partisan composition of vote method changes.

The model is still at Youngkin +2.

Confidence intervals narrowing quite quickly now.

A better model would have treated the absentee ballots a little better. I could also have imposed a Bayesian prior on vote shares; obviously, Dems aren't losing the absentee votes in big counties by 50 points.

This could end up closer to Youngkin +4 or +5 at the race absentee votes in Democratic strongholds are falling off. Just clear complacency among Dems for McAuliffe. But also real evidence of sub/exurban vote-swticthing. See scatters here (at the county level).

The model is now telling me to call the contest for Youngkin — which, uh, yeah, is pretty obvious and was from the start. Look how flat those lines are over election night, esp after we started getting returns from Fairfax.
You love to see a working model...

Someone DM'd me for a model update, so here you go. Also attaching a chart comparing McAuliffe's projected margin in the remaining absentee buckets as compared to Biden 2020.

OK, here is my final chart for the night.

Comparing 2021 and 2020 in Virginia, we see disproportionate Democratic drop-off in big cities and evidence of substantial vote-switching among swing voters (or at least among voters in swing counties). Article coming soon.

That chart uses the final predictions from the live-model, made when 98% of the estimated vote was reporting (just a few absentee buckets were outstanding). Point being the comparisons are using complete observations and are as close to final as we’ll get until tomorrow AM.

/END