Our best and worst takes of 2024

Even in a good year, we didn't get everything right.

Judging by my inbox, you're probably getting a lot of Happy New Year messages, whether from your favorite Substack writer or the muffler shop you stopped into 13 years ago.

Well, Happy New Year anyway from Seoul, where I’ll get to celebrate 2025 earlier than most of you. I'll try to provide a little bit of substance here by taking you through some of our highlights and lowlights from 2024.

I’m not sure I’ve ever written so much: we published 164 Silver Bulletin newsletters in 2024, including this one. Considering that I, uh, tend to write long, that’s the equivalent of about three books. On top of that, I also did publish a book, On the Edge. Silver Bulletin readers were vital in helping On the Edge to reach bestseller lists, for which I remain deeply appreciative.¹

I'd say my main regret is how much the newsletter became dominated by the 2024 election from June through November — but that would be disingenuous since this was a conscious choice. Once I turn the election model on, it’s like some extra appendage that always gets in the way of everything else. In the long run, the only real options are to staff up beyond just two people — which we’ll almost certainly be doing in some form next year — or to accept that as the price of doing business.

Still, even ignoring the election subscription spike — there’s already been some predictable churn — we’re exiting the year in a much better position than where we started it. Thank you so much for being a part of it, especially if you're one of those subscribers who isn't mainly reading for election coverage.

The rest of this newsletter is basically one big ICYMI: a link to 26 noteworthy Silver Bulletin posts, with some color commentary about our process along the way.

Our 5 most popular newsletters

For this list, I’m going strictly by the Substack algorithm, which ranks “top” posts based on some combination of pageviews, comments, and other things. It’s in ranked order based on the Substack secret sauce, while the rest of the lists are in chronological order. The dates may be off by a day in some cases since Substack is defaulting me to Korea time.

• A shocking Iowa poll means that somebody is going to be wrong (Nov. 3). We'd planned to write more about the Selzer poll, but President-elect Trump’s ridiculous lawsuit against Selzer and the Des Moines Register complicated some reporting we had going on.² This particular post holds up just OK, I think. It did conclude the poll (showing Kamala Harris leading in Iowa; she lost by 13 points) would “probably be wrong”. But the poll maybe still got too much weight in my “mental model” of the election. It didn't influence our statistical model much, which took it more in stride as a probable outlier.

• 24 reasons that Trump could win (Oct. 21).³ Obviously, posts that were bullish on Trump will tend to look smart. I do think we were relatively prescient, though, in understanding that Democrats had deep problems and there wasn't any One Neat Trick that was going to solve them. I thought about doing a follow-up “24 reasons that Harris could win” post but couldn't make it past 8 to 10 items — maybe that revealed something.

• A random number generator determined the “favorite" in our forecast (Nov. 5). Our final pre-election Model Talk highlighted that Kamala Harris had become the “favorite” by 12 simulations out of 80,000 (!), a margin so close as to be literally more random than a coin flip. The model remained within the vicinity of 50/50 — albeit not that close — for nearly the entire period after Harris took over for Joe Biden. People got extremely excited whenever the streams crossed (i.e., one candidate pulled slightly ahead) but I think we mostly did a good job of avoiding overhyping those changes, with a couple of exceptions I'll get to later.

• There’s more herding in swing state polls than at a sheep farm in the Scottish Highlands (Nov. 2). I hope this post has a lot of “evergreen value”, as we call it in the news business. I expect pollster herding to be a big issue going forward, and the aforementioned Trump lawsuit may have a further chilling effect on pollsters’ willingness to publish seeming outliers.

• A tour of the 7½ key swing states (Sept. 22). A fairly standard post, but I'm cringing because the “½” state in 7 ½ was Florida, which polled closely for parts of the race but which Trump wound up winning by 13 points. I still think you'll be better off in the long run not trying to outguess the polls and taking your wins and losses — but you’re going to have your fair share of Ls.

My 5 favorite newsletters

• Go to a state school (Apr. 23). Taking a trollish-seeming tweet and showing that there actually was some deeper thought behind it is one of my favorite modes for the newsletter.

• Against revisionist history on Biden 2024 (Aug. 22). This is one of the posts I remember triggering a fairly negative reaction. That’s not to say the balance of sentiment was necessarily bad — it got a lot of likes — but it was polarizing. One theme here is that you'll sometimes get in trouble when there’s a sense that you're raining on someone's parade: this post was published in the middle of the DNC at a high point for Democrats. But precisely because there's a psychological price to be paid for touching the third rail in those circumstances, I think there's a lot of value in doing exactly that. Literally, part of what I hope you're paying me for is to present a well-evidenced unpopular opinion and ignore the vibes when the situation demands it. You’ll see some examples later of newsletters where I succumbed to the vibes. But this post holds up well, I think. Despite seeming like something of a hot take, there's a lot of empirical and historical analysis buried underneath the headline.

• Kamala Harris is not going back to the failed politics of 2016 (Aug. 24). In retrospect, Harris’s acceptance speech was the high-water mark for her campaign. I thought it was an excellent speech, but oddly, the campaign would later somewhat abandon the themes it covered. Then again, maybe the fact that I liked the speech was a bad sign: I’m not your typical voter, and Harris didn’t get much of a convention bounce. Still, I was happy with this newsletter. It's hard to come up with interesting things to say under deadline pressure after events like convention speeches and debates. I like that this post honed in on a particular theme — the way Harris contrasted herself with Hillary Clinton — instead of taking a paint-by-numbers approach.

• Is it crazy to pay Juan Soto $765 million? (Dec. 15).

• Save Daylight Savings Time (Dec. 18). Two recent posts indicative of the sort of analysis I hope to leave time for outside of election peaks, where I can cover a slightly unexpected topic in a relatively comprehensive way but also come to a reasonably punchy and evidence-driven conclusion.

5 slightly overlooked newsletters

• How culture trumps economic class as the new political fault line (Mar. 28). “Overlooked” is a relative term: we hope not to publish stuff that will totally bomb. But a few items felt a bit lost in the shuffle. This post holds up well, I think, after an election where there was actually some racial depolarization but where Democrats lost further ground with the multiracial working class.

• The long, strange political shadow of 2020 (May 10). Written back when Biden was the nominee (there was a more Harris-centric version later). Indicative of how Democrats’ problems were baked in fairly early and how they couldn't just Etch-a-Sketch their way out of them and were actually going to have a tough time persuading voters that Biden's presidency had been better than Trump's.

• Steph Curry is Magic (Aug. 11). A celebration of an all-time great player that was too much fun not to write up even in the midst of election season.

• SBSQ #12: Will the polls lowball Trump again? (Sept. 1). Not for the headline item about polling error but for the second beat about my tenure running FiveThirtyEight @ Disney that got somewhat buried underneath it.

• Always. Be. Blogging. (Nov. 23). Like the recent Las Vegas guide, a narrowly targeted post — mostly meant for people who are actually writing Substacks or otherwise writing for public consumption. I think it's important to do this sort of thing once in a while — a Substack should never get so big for its britches that it can’t geek out.

5 posts that look prescient

• It's time for the White House to put up or shut up (Feb. 19). I'll be quick in this section to avoid doing too much I-told-you-so-ing. But can I be honest? We took a lot of grief for our suite of posts about Biden’s age dating back to last year. So was the way it played out was edifying? Well, does a bear shit in the woods?

• Democrats are hemorrhaging support with voters of color (Mar. 16). The mileage varies slightly here: Harris did somewhat better than polls projected among Black voters but even worse among Hispanic and Asian ones. Still, one of the Ws from taking polls more or less at face value.

• Biden has a weak hand (July 10). I really dug in on this Biden stuff, not just saying that Democrats should replace Biden but also that his position wasn’t tenable after the debate.

• Tim Walz is a Minnesota Nice choice (Aug. 6). She shoulda picked Shapiro.

• The polls are close, but that doesn’t mean the results will be (Nov. 4). Great Eli post, and one of a couple of times we pointed out before the election that one of the candidates sweeping all the swing states was actually a fairly likely outcome. Indeed, the actual map that occurred — with Trump winning all 7 swing states but everything else going to form — was the modal (single most likely) outcome in our forecast.

1 weird newsletter that could arguably fit into either the previous or the next category

• Google abandoned "don't be evil" — and Gemini is the result (Feb. 28). In the short run, this newsletter seemed highly prescient: Google CEO Sundar Pichai admitted later that day⁴ that his company’s weirdly woke AI engine — depicting, for instance, multiracial German WWII soldiers — had been a major misfire. But it was more of a blip on the radar in the long run. Google stock quickly recovered, and although they remain behind OpenAI and Anthropic in the AI race, I underestimated the goodwill they'd built up with consumers and investors.

5 posts that don’t hold up well

• Kamala Harris can't meme her way to victory. Or can she? (Aug. 10). And finally, our worst takes of the year. I got caught up here in the Brat Summer vibes/Number Go Up mode when the moment called for a more rain-on-the-parade take. I'll give myself partial credit for course-correcting a week later before there had yet been any slippage for Harris in the polls. But I think that only takes this post from an F to an F+.

• We removed RFK Jr. from our model. But it didn’t hurt Kamala. (Aug. 25). I was pleased with myself for how quickly we were able to remove RFK Jr. from the model without any downtime. But in that self-congratulatory mood, I think this was my worst post of the year. It confused the absence of evidence — there wasn’t yet enough data to say exactly what effect Kennedy’s dropout would have — for the evidence of absence.

• How big will the bounce be? (Aug. 27). Actually a very good and judicious Eli post that I ruined with a question-begging, presumptuous headline. What bounce Harris did or did not get is slightly ambiguous because of the timing of the RFK news — which came the day after the DNC and I think probably did help Trump. The headline jumped to conclusions — and headlines count as content.

• Alaska, Alaska, Alaska (Sept. 15). A fun post but put too much emphasis on a single poll, and the Klondike had been fool’s gold for Democrats in the past. To be fair, there hadn't yet been any other polls of the state.

• Trump's dominating the news again. Maybe that's good news for Harris. (Oct. 29). Following controversial remarks a comedian made about Puerto Ricans at Trump’s Madison Square Garden rally, something of a last hurrah, both for me and the rest of the media, for getting caught up in the “politically correct” Beltway narrative about how voters would think about Trump. Seemed smart when Harris did get a slight uptick in the polls in the race's closing days. But that proved to be fake: exit polls suggest that Trump probably won late-deciding voters. Trump even won Osceola County, Florida, the county with the country's largest share of Puerto Rican voters, with a 15-point shift (!) from the 2020 results.

1. I think the book holds up well eight months after I sent the final, irrevocable version to my publisher — especially the parts about the increasing political conflict between The River and The Village, which anticipated Silicon Valley’s rightward turn, a theme I hope we’ll hit on next year as cracks start to emerge in the new GOP electoral coalition.

2. This is very much not to imply that we’re caught up in any legal drama, just that it turned from a polling story into a legal story and the latter is very much not in my wheelhouse.

3. After the election, re-headlined as “24 reasons that Trump won”.

4. I think I’m getting the timeline right on this, but slightly uncertain because of the Korean time zone shift.