How high will turnout be?

Plus, a 36-hour sale on annual subscriptions! And our Election Night plans.

Happy Halloween! A holiday we never really get to enjoy because it comes so close to the election. A sizable share of Silver Bulletin readers¹ consume the newsletter directly in their email client. So we always try to respect the integrity of your inbox. This is what we call a “utility post”: something where we could have blasted out two or three emails, but we think you’re better served by consolidating them into one.

A sale on annual subscriptions

Beginning now through tomorrow (Friday) night, we are running a 36-hour sale on annual subscriptions. If you’re a new subscriber during this period, you’ll get a 25 percent discount on the first year of a yearly Silver Bulletin sub. Our subscriptions are heavily tilted toward monthly sign-ups relative to most Substacks, and we appreciate your business in any form. But we also have ambitious plans for the post-election period, including reviving more sports models, and we think this is a really good deal! The sale will end at roughly 3 a.m. Eastern (Midnight Pacific) on Oct. 31.

You can sign up using the usual link below:

Election Day (and Election Night) plans

Our Election Day begins right at Midnight on 11/5 when we’ll collect any last-minute polls and run this year’s model for the last time.² We’ll run 80,000 simulations rather than our usual 40,000 and also do some extra double-checking — so look for this to post sometime between 12:30 a.m. and 1. Along with the model update, or shortly thereafter, I’ll post my final pre-election narrative overview of the race. Per usual with model updates, all of this will be exclusive to paying subscribers.

Then we’ll try to get some sleep and probably fail at it.

Then, in the evening, we have a two-pronged plan. Substack doesn’t yet have a great platform for liveblogging, so Part 1 is that we’ll hang out in the Subscriber Chat instead. It’s a user-friendly solution that worked well during the debates. The participants from the Silver Bulletin side will be (of course) me and Eli, plus Matt Glassman, a friend-of-the-newsletter and a Senior Fellow at the Government Affairs Institute at Georgetown University and the only other person to date apart from me and Eli to have a Silver Bulletin byline. I do want to warn you that, as has been the case for me for every Election Night since I started FiveThirtyEight in 2008, I always have one too many things going on, so my contributions will be more direct and straight to the point, while Eli and Matt (and hopefully some of the subscribers) will provide more color.

Part 2 is a little more involved: we’ll probably be booting up the Election Night model that I previously built at FiveThirtyEight.

The Election Night Model isn’t trying to be the New York Times needle. That would require a whole engineering team. Instead, it depends on called states — places where the networks have designated a winner. The model will draw on the logic of our final pre-election simulations, including the correlation structure in the data. If Donald Trump wins North Carolina, for example, he’ll not only capture its 16 electoral votes, increasing his odds of winning the Electoral College, but the model will also calculate what we can infer from North Carolina about other states: for instance, this will undoubtedly boost his chances in Georgia.

When we ran the Election Night Model at ABC News, I was restricted for internal political reasons to relying on “calls” only from the ABC News Decision Desk. But now we have more flexibility.

Still, the Election Night Model takes a conservative approach. With a very small team, we want to automate the decision-making as much as possible. However, we should also be able to articulate conditional scenarios, e.g. if Kamala Harris wins Arizona, her win probability will increase to X percent and, include some notes about those in the Chat.

I want to hedge just slightly here because I still need to update the code this weekend, and test it to ensure the model is spitting out reasonable-looking results given a plausible range of scenarios. If I catch the flu or something, the Election Night model will probably be a casualty. So let’s call it an 85 percent probability — lower than the New York Yankees’ odds when they were ahead 5-0 in the World Series last night. (Congratulations, Dodgers.)

How high will turnout be?

We don’t make a big deal of it because it’s somewhat detached from the rest of the model — in other words, it doesn’t affect the Electoral College odds — but our forecast does spit out a forecast for total Election Day turnout nationally and in each state. On the forecast landing page, you can find these in the spreadsheets labeled “National Topline” and “State Topline” respectively.

The turnout forecasts are fairly simple. They’re based on longstanding polling from Gallup and Pew that asks voters how enthusiastic they are compared to previous elections, plus some relatively detailed estimates of population growth in each state. The final factor is the closeness of each state: other factors held equal, voters turn out more when they think their vote matters.

Gallup just published new enthusiasm numbers today, in fact, and they are high, particularly among Democrats:

On this basis, our formula predicts a total turnout of 155.3 million, with an 80 percent confidence interval between 148.2 million and 162.5 million. That midpoint is just slightly lower than the 158.7 million votes in 2020, but is much higher than the 137.1 million voters who turned out in 2016 or in any previous election.

I think I feel OK about projecting a very slight decrease. The 2020 election was outlier-ish from a turnout standpoint, perhaps in part because people had so many ways to vote (and so little else going on) during COVID. And although the Electoral College forecast is very close this year, there aren’t all that many states that matter. Turnout also declined by about 4 percent in the 2022 midterms as compared to 2018, despite population growth.

But a record turnout is well within the confidence interval — and remember that the forecast only captures 80 percent of cases. (There’s a 20 percent chance that turnout will fall outside this range.) One difference this year, though, is that Democrats might actually be rooting for lower turnout. The party is now centered around college-educated voters who are big consumers of political news and vote in nearly every election they can, leading to Democratic overperformance in cases like ballot initiatives and special elections. It’s a cliche, but turnout — particularly whether Donald Trump can turn out his marginal voters, groups like disengaged young men who will probably either vote for Trump or not vote at all — may determine who wins.

1. Roughly 40-50 percent for a typical newsletter.

2. The only slight hedge is if there’s some extremely dramatic and unforeseen event in the final 5 days that causes pollsters to go back into the field very late. Call it a ~2 percent chance. Barring that, we’d strongly encourage pollsters to publish any surveys by Midnight.