We took a shot, tried to be transparent about what we did, and ended up giving Youngkin a 33% chance of victory. Those things happen 1/3 of the time. Forecasting and modeling is hard.
Some success for the model: as @Thorongil16 said, we did appear to catch the fact that there would be significantly more college-educated voters in this electorate than in 2020, and modeled the electorate decently well. Where we missed was in margins among demographics.
I don't feel bad for saying polls like Fox were bad. The electorates illustrated by those never materialized. We always said Youngkin's path to victory would consist of flipping a ton of Biden voters, including white suburbanites. And that's what he managed to do.
At the end of a long campaign, here's where the model made by me and @Thorongil16 stands. The race is rated as Lean Democratic, with a forecasted margin of D+3.6 and a win probability of 67%. An interactive map is over at striderforecasts.com.
The consensus on this race is all over the place, and we've seen polls in the last week ranging from D+3 to R+8. Absolutely nobody can agree on what an electorate looks like or who's voting. What we all can agree on is that Democrats have a large early vote (EV) lead...
But even then, nobody knows exactly how big, and the margin here matters. We know via L2 that the EV electorate is roughly 2:1 Biden, but the crossover vote matters here. If it knocks EV down to D+19, Youngkin's probably favored. D+27 and it's likely McAuliffe's race to lose.
On the day before the race, the #VAGov forecast from @Thorongil16 and me remains Lean Democratic, with a D+3.5 margin forecasted and a 67% win probability. An interactive map with county margins is over at striderforecasts.com.
Both have a chance, but I'd rather be McAuliffe.
Youngkin appears to have momentum, and this was a key reason cited by @JMilesColeman and @kkondik in their Lean R call. It's a respectable one you can make a good argument for. I'd disagree primarily for a few reasons...
(1) the overall crosstab splits don't show as many danger signs for Democrats as the toplines do (2) the polling average has been flooded by a host of pro-GOP firms as of late (3) there *appears* to be significant herding and nonresponse impacting the results too
Democrats are still forecasted to win by a touch over 3.5. Race remains Lean Democratic with a 68% win probability, in our eyes.
Some definite late movement in our model to Youngkin, but in our eyes, this is not enough.
Firstly, Virginia is a highly educated Biden +10 state. We begin with a heavier McAuliffe prior due to turnout dynamics that, based on education and propensities, would be more Dem-leaning
I do think some polls have drifted towards heavily R samples; Fox poll yesterday was an example, and based on some of the margins, I think the Echelon poll oversamples Rs. Hard to tell, but I do not think there has been this much movement re: turnout dynamics in the 2 weeks
I can't find crosstabs for this, but on the surface, this is probably not what you want to see if you're Terry McAuliffe. A continuation of the "late movement" theme; whether that's an artifact of LV screens being wonky is unclear, but it goes in the model (D+4.1 now).
pre-emptively to those saying McAuliffe's margin has plunged in our model: in the last 3 days, we've added in the Fox, Washington Post, and Echelon polls. Doesn't take a genius to figure out that it'll change the forecast!
Something in this poll is that enthusiasm is at comparable levels across party lines, so this would indicate Youngkin is pulling a *lot* of Biden voters over to his side. Whether that's a reasonable estimate or not, I don't know, but that's his actual path to victory.
A pretty reasonable poll and electorate screen that tells us more of what we knew: a close race, where more Biden->Youngkin crossover is happening rather than McAuliffe/Trump. Anyways, something interesting is how downballot races like the AG/LG appears to be 3-4 points more Dem.
Youngkin outperforming Trump by 15 or so with noncollege whites would be interesting. I think his path to victory is going to involve some gains with noncollege whites and more gains with college whites, especially if given this level of ticket splitting, but let’s see.
Yeah it’s only a 43% educated electorate but also, if you want to play that game, dragging it up to 50% or so would only give you like a D+3 result at best, so…still a very close race, once again.
Every poll has two parts: the electorate screen and the margins among the demographics. The electorate screen is where you give a representative sample of what you think the electorate will look like. The crosstab margins are the margins among those groups. 🧵
If the electorate screen is broken, there’s a problem. It means that the poll will not be representative of the state. Because fundamentally, you’re giving the results that would occur in an electorate different from the one that actually exists. Allow me to construct an example.
Say that a state is 40% Black, 20% Latino, and 40% White. Your poll would probably want to match up in terms of demographics — You cannot poll all 3M voters in the electorate, so you need a good sample to allow you to approximate what you think the electorate resembles.
Given recent polling trends, the model's converging to ~D+5, in my read. That'll likely be what our forecast ends up as (and thus on the border of lean/likely Dem too). The confidence interval here would be D+2 to D+8.
If I had to be forced to pick, I guess I'd take the under? Just because it's not really statistically safe to rely on a polling miss of >4 points even accounting for the flaws I've discussed. But honestly, I'd just go with the forecast, because my punditry has never been great.
I think a lot of the polls aren't great, but part of analysis is setting your rules beforehand and then following them. We add in just about every poll we can *unless* it's from a banned agency or violates our pre-set rules on internals/data quality/data collection procedures.
#VAGov polls have largely been junk, but Youngkin can win this race. But the early vote suggests that turnout for both sides will probably be fine, so if he wins, he wins by swinging significant amounts of college-educated voters (specifically college-educated whites).
No polls have been showing a significant swing right with college voters, though. They've either been showing the race close because Black voters swing right by 35 from 2020, because non-college whites swing right by 25, or because their partisan screen is ridiculously R-heavy.
I find the "non-college whites swinging right by 25" part a bit questionable because Trump's support is already very, very high in most of the rural white areas, so unless you've seen a catastrophic collapsed with Virginia urban whites (questionable?), I don't see the math.
Model updated for #VAGov with the new @Suffolk_U poll. The forecast from @Thorongil16 and me is D+5.8. Suffolk is the second high-quality pollster to show a close race, so let’s dive in 🧵
I will say this: if you firmly believe crosstab examination is not a useful exercise, please stop reading. Nothing I say will be of any utility to you, because we have fundamentally different ways of going about this. Save yourself the time and stop reading.
Suffolk’s poll only has an N=500. You need to be careful with the window of results you consider as a result. Anyways, the racial crosstabs are okay; McAuliffe wins Black voters 81-8, and I would wager that undecideds here are expected to break Democratic.
Emerson has been banned by us for a month now because of MTurk data quality issues. Not including it regardless of what the poll says.
Also, the n=145 among Black voters here. A 36 point swing right would be well outside the margin of error for the Black subsample.
I need to stress that *I COULD VERY EASILY BE WRONG*. All I'm saying is that if Youngkin wins, the path to victory will look *nothing* like what these polls suggest. Data points to a McAuliffe win, but if D turnout is low or Youngkin swings suburbanites, it could change.
Heading into the final week or so before the election, here’s how @Thorongil16 and I have things shaping up for #VAGov. We still have this race as fairly stable and think McAuliffe should win by about six-and-a-half points. His odds of winning are a bit over 80% (likely D).
Democratic early vote has really surged recently, as folks like @samshirazim, @ZacharyStarbuc1, and @WinWithJMC have noted. This week will be crucial for Democrats, but they had a very good weekend and continue to see their base areas go up in share of statewide votes cast
As folks like Zachary and Sam have noted, NOVA will likely be fine for Democratic turnout. Areas like Richmond are lagging a bit, but that probably isn’t enough for Rs, who currently have their own troubles with getting whites in Southwest VA to remember an election exists.
Points in favor of a Youngkin victory: Biden's approval is horrific, so independents might break R, and and Democratic turnout is undeniably lagging currently, so you could see him eke out a 1-2 pt win.
Points against: Virginia 2012-2020, and he *still* can't clear 46% in polls.
That last part is especially telling for me, because in every poll, the undecideds have been disproportionately Democratic, and we know from a lot of polling data that split ticket voting is quite low in this race right now. So the median scenario shouldn't be a tossup IMO.
Polling already seems to price in a Republican turnout/enthusiasm advantage, and even with that, Youngkin isn't crossing 46%, which is a problem. People are eventually going to decide, so he'll have to get a good bit of crossover votes he's not currently getting.
Here are all public, nonpartisan polls of likely voters conducted in October for #VAGov. Polling points to a D+4 race, but it really depends on the screen used. This stuff is really hard to model!
CBS/YouGov had two LV models, so both are included (but weighted by half to ensure that double-counting isn't a problem), and Monmouth's LV model was D+2, so we used that; our rule is to use LV results wherever possible.
That doesn't really favor a party; sometimes we put in things like D+2 instead of the tie among registered voters in Monmouth, but then we've also omitted the Fox D+11 poll of registered Virginia voters, so...
We've updated the model made by @Thorongil16 and I with the Monmouth poll.
Our forecast is at D+6.7, with an 84% chance of Democrats winning, making our rating Likely Democratic. You can view the county margins at striderforecasts.com. 🧵[1/]
The Monmouth poll is absolutely not a good one for Democrats. Let's not dance around it. Even if you take the educational splits and weight it with our screen, which is 52% college-educated, you get a D+2 margin. This is not good.
What's going on, though?
From every poll, we see that Biden voters choose McAuliffe while Trump voters choose Youngkin, with little middle ground. But independents have taken an 18 point swing *in a month* in this poll. They were McAuliffe +9 last month, Youngkin +9 this month.
Educational polarization means it is genuinely way more plausible for Democrats to win 53% of the two way vote than people think — it’s just that one of the key ways to do it lies in getting minority turnout to stay at presidential levels in off-years.
I’ll put it this way: a month ago, I wrote a piece about educational turnout differential and its electoral implications. But the upshot is that if Democrats can keep their minority turnout numbers high, then the midterm electorate is a point or two *more* favorable to them.
It means that if Black voters were 11% of the electorate in 2020, you’d want them to be around 11-12% of the electorate in 2022 — we’re speaking in terms of electorate share here. Essentially, the drop among minorities cannot be as stark as the drop among white noncollege voters.
FWIW I still believe a default is not the most likely event; between the $1T coin, 14th amendment, and a filibuster carveout that Manchin refused to rule out, I think there's enough paths to prevent it.
But if it happens, voters will blame the party that controls all 3 chambers.
Nobody cares about the filibuster or Senate procedure.
If a voter loses their social security checks while a Democratic trifecta was in office, the response will be to blame the party that was in control for letting it happen. And that's basically all there is to it.
Want to make #VAGov one where you can credibly call the Republicans favored? you have to see an event of such groundbreaking proportions that it shifts the national mood 5 or 6 points to the right in less than 4 weeks. Something like a default.
Point that Nate brings up here, but it's more plausible for them to sustainably drive up their average popular vote margins and hope that carries them. A state being R+2 relative to the nation is ok if the nation's median is D+4, for example.
I'm in the minority, but I am not convinced that Democrats are long-term completely done in the Senate if this happens. Biden won 25/50 states. Let's say WI slips, but they gain two of NC/TX/AK. If you keep your vote share high enough and midterm penalties fade, you'll survive.
The messaging needed to appeal to white working-class voters often directly conflicts with the strategy needed to keep current Dem voters engaged. And because of where the center of power is concentrated in the Democratic party (donors/staffers/etc), a pivot is near-impossible.
Parties aren't monoliths. You can't just press a button and shift focus. But if you did try to force a message shift at the top, the odds that you break through without some serious change in social/racial rhetoric are very minimal in today's day and age.
And because those WWC voters *no longer see themselves as Democrats*, the odds that you get them back are minimal, but the odds you manage to lose your primary by angering current voters are way higher as a result. So what's the incentive?
Short of an extraordinary surge with working-class whites in southwest Virginia, I do not see where Glenn Youngkin is getting the votes needed to beat Terry McAuliffe.
Every poll shows him losing college voters by ~20 points. And he's not even trying to cut that margin down.
College voters are anywhere between 47-53% of an off-year electorate in Virginia (this stuff is impossible to estimate precisely but we can be decently confident it's in that ballpark). In polls of likely and registered voters both, McAuliffe is keeping that margin near Biden's.
The next lane for Youngkin would be to hope Black turnout craters while his base of rural whites turn out to an extraordinary degree. That's plausible, but it's not really the strategy you want to rely on, especially considering his base isn't exactly reliable with voting.