Announcing 2024 election model plans

The presidential model will be available to Silver Bulletin subscribers beginning in June.

As much as I might dream of spending election night on a beach in Thailand, there’s no escaping the fact that I’m going to have a lot of my mental bandwidth devoted to the election whether I want to or not. So since I’ve teased at this long enough, here’s the headline: I’ll be running a presidential election forecast again this year — basically the same model¹ that I’ve run every year since 2008 (formerly under the auspices of FiveThirtyEight). The model will run here at Silver Bulletin: I’ll publish polling averages for everyone, and probabilistic forecasts and additional commentary and data for paying subscribers.

However, the pace will be different this year. Instead of constant updates, I’ll instead initially run the numbers once per week, but then more frequently after Labor Day. The purpose of this is twofold. First, it allows me more opportunity to put the numbers into proper context — I’m not sure there’s a lot of journalistic value in constant updates when there’s not a lot going on. But also, this way leaves me enough time to accomplish the other things I have on my list, including continuing to cover plenty of non-election topics at this newsletter.

If you’re already interested, you can sign up using the link below. Or if you want to learn more, here is more detail on what I have planned in the form of a Q&A.

When will the model launch?

June, with some chance of late May if I can get my ducks in a row. This is my typical schedule: in 2016, we launched the FiveThirtyEight presidential model in late June. And in 2020, we launched our presidential polling averages in late June but, amidst the pressures of the pandemic, didn’t turn the full model on until early August.

I’m not planning on doing too much re-engineering of the model under the hood this year, which should help. On the other hand, I’m doing this myself this time instead of with a bigger team, although I will likely hire some help. It usually takes a few weeks of dedicated work to get everything ready and I also have some other projects that are getting underway shortly. I’d love to launch by Memorial Day — but June is a safer bet.

What will a model update look like?

These NCAA tournament model updates are a good template. A typical update will include:

• National and state presidential polling averages, free for everyone. These aren’t simple polling averages but instead use a Bayesian-type process to make inferences based on national polls and polls from other states. For instance, if Arizona hasn’t been polled for a while but Joe Biden’s position has improved in other polls, the Arizona averages will take that into account.

• For paying subscribers, probabilistic forecasts of every state and the Electoral College, with various charts, maps, graphics to flesh out the numbers.²

• Spreadsheet(s) with additional detail and raw data, also for paying subscribers.

• Finally, text commentary to contextualize everything, sometimes longer but often fairly expedient. Keep in mind that there will still be more philosophical pieces about polling, the election and so forth that will run separately from the model updates, often in front of the paywall.

So the probabilities will be paywalled?

Yes, strictly. Over the years, I’ve had enough issues with people (mis)interpreting probabilistic forecasts that I deliberately want to reserve the probabilities for a more self-selected audience. However—

Will you still be writing regular newsletter updates/blog posts?

Yes, absolutely! The theory of the case is that the model updates will be mostly additive. I’ll still be writing about the typical mix of Silver Bulletin topics, in other words — that is, you should still get 2-ish non-model posts per week (about a third of which will be paywalled). I will acknowledge that the pace of non-election stuff may slip once we get past Labor Day. But I’m very much hoping to write about a variety of topics for as late into the year as possible, and then again once the election is over.

How often will you run model updates?

• Launch through Labor Day: ~1 time/week³

• Labor Day through mid-October: ~2 times/week

• Mid-October through Nov. 5: ~3 times/week or “as needed”

The idea here is to encourage more of a “slow food” approach to the election and not compulsive refreshing of the forecast page. Maybe I’ll revert back to constant updates in 2026 or 2028, but I don’t know. Unless you’re literally betting on the outcome, there’s not a lot of value in changing your priors after every new poll that comes out. Particularly for this election: the fundamentals of a Biden-Trump rematch are likely to be fairly “locked in” and polling shifts will often be noise.

Are you going to be increasing prices?

Full disclosure — yes, so that may be a good reason to subscribe now. Current prices are $8/month or $80 a year and will remain locked in indefinitely so long as you remain continually subscribed.

I plan to increase prices to $10/month and $95/year at some point before model launch. I will also temporarily turn off the option for monthly subs at some point⁴, then restore them after the election.

Although I’ve flirted with the idea of a separate pricing tier for election forecast updates, I just want to keep things simple. The aim is to build this newsletter for the long term. Even if you’re mostly subscribing for the election updates, I hope I can continue to provide you with plenty of value even once the election is over.

Plus, if you sign up, you'll get access to all other Silver Bulletin paid content: 2-3 paywalled posts a month on top of the election updates, plus the monthly subscriber questions post, plus access to next year’s NCAA tournament model.

Why should I pay for your model when I can get others for free?

It's a great question. So let me take two cracks at answering it.

There are lots of smart people working on election forecasts these days — but election forecasting is a notoriously difficult problem and this is the only model in the business with any sort of long-term track record. Historically, the model’s forecasts have been extremely well-calibrated, meaning that there is “truth in advertising”: for instance, events that are forecasted to have e.g. a 70 percent chance of occurring have in fact happened about 70 percent of the time over a large sample of data. (Even in 2016, for example, when Donald Trump won as an underdog, the model gave him a ~30 percent chance, considerably higher than other models, betting markets or the conventional wisdom held.)

The other answer makes me cringe — but you're not just getting the model, you're also getting my commentary. This is my fifth presidential election cycle and I’ve developed some pretty thick skin that helps me to stick to my process, keeping my forecast and my personal political preferences decoupled from one another. And I hope I can write about what will undoubtedly be an anxiety-inducing election in an insightful and occasionally even entertaining way. I’m also looking forward to writing for a more self-selected group and not 10 million drive-by readers coming in from ABC News or ESPN.

What about Congressional and gubernatorial elections?

So here’s the bad news. I’m not counting on running Congressional and gubernatorial forecasts this year. I say “not counting” because there’s some chance I’ll change my mind depending on what kind of help I’m able to hire and how everything else is going. But I’ve tried to carefully budget out my time between now and November, and I want to be able to give all my projects the attention they deserve. Between the regular newsletter updates, the presidential election model updates, some consulting work⁵, a new podcast that I’ll be announcing very soon (!), and public appearances related to the book, I need to triage something. With it being a presidential year, Congressional forecasts are the lower priority.

You mentioned that you’re looking for help?

Yep, look for job listings soon! This newsletter is already exceeding my expectations, and I’m past due to look for some help. I will likely be hiring a part-time assistant editor position on an ongoing basis — and depending on how a few other details work out, possibly also a second position through November for someone to help with entering data, generating charts and graphics, and perhaps running the model and writing occasional forecast updates and other commentary when I’m out of pocket. You are more than welcome to email me to indicate your interest, but I probably won’t be able to respond until I formally post the positions.

How do I sign up?

Just use the link below. As much as I’m one of the many Americans who isn’t thrilled about a presidential rematch, I expect this election to be close and competitive — and very important to the future of the country — and I hope you’ll choose to follow Silver Bulletin’s coverage.

1. There are always some minor changes and bug fixes from year to year, so you quickly get into a ship of Theseus paradox. If you’ve rewritten every line of code in a model at some point, is it still the same model?

2. As a warning, these charts will likely start out being fairly spartan but will get prettier as the election goes on.

3. I’m hereby reserving the right to take one or two mulligan weeks if I’m getting totally buried with everything else, but in exchange, there will likely also be some bonus updates when news hits or there’s a significant shift in the polling.

4. Just to be clear, if you’re a current monthly subscriber — or you sign up for monthly subs while the option is still on — this won’t affect your experience at all. But I’ll suspend the option for new monthly subs at some point.

5. This will potentially include one election-related consulting project (for a financial firm, not a political client). However, the firm would be buying my time, not the model — I’ll have free rein to run the model here at Silver Bulletin.