SBSQ #11b: On the future of the newsletter — and the sports models

Plus, we’re launching a new subscriber Chat platform. And: hardcover or audiobook?

This is the long-delayed second half of what is technically the “July” edition of Silver Bulletin Subscriber Questions. The first half is here, but consists of a question about Kamala Harris’s running mate that is now pretty out of date. We’ll try to get back on schedule with the next SBSQ at the end of the month — possibly we’ll experiment with a video lightning round edition — and you can leave questions in the comments below.

Speaking of comments, today we’re announcing a new addition to this publication: Silver Bulletin subscriber Chat. You can access Chat on the web, or you can get the Substack app by clicking this link.¹ New chat threads won’t be sent via email, so you might consider turning on push notifications so you don’t miss conversations as they happen.

This is a conversation space exclusively for subscribers — kind of like a group chat or live hangout/Discord. Eli and I will post questions and updates that come our way from time to time, and you can jump into the discussion. Additionally, we’ll provide Silver Bulletin Founding Members the ability to start Chat threads.² Participation will require having a paid subscription: at this point, this newsletter has a large audience, and we’ve learned from experience that the community functions work a lot better when we limit the membership to a self-selected audience. You can sign up for a paid subscription — full access to the election model, exclusive Model Talk and SBSQ posts, ability to participate in Chat and comments — at the link below. This edition of SBSQ will run free, however. We’ll take a quick question about the book and then a longer one about the future of the newsletter and the sports models.

Devin Chellberg asks:

How many visualizations are in your new book? Will I be missing out if I choose to listen to the audio book?

So, I recorded the audiobook myself this time, which I didn’t do for The Signal and the Noise. It was a lot of work — basically 50 hours in the studio for what eventually became a 15-hour (!) audiobook. But I’m really glad I did it. I worked with a really great editor/producer, and the audio version very much breaks the “fourth wall” — we’re aware that verbalized speech and written speech differ to a larger degree than is often appreciated.

For instance, a term that appears frequently in the first quarter of the book — World Series of Poker or WSOP — reads better as “WSOP” but sounds better as “World Series of Poker”.³ And we’re aware of long sentences with a complex nested structure that read well enough on the page but don’t translate as well in vocalized speech. So some of these have been rewritten/revocalized in subtle ways. In the audio version, there’s even an ad-libbed joke or two based on developments that came after the physical book was due to Penguin.

Another issue, as Devin’s question gets at, is the various data visualizations that appear throughout the book. In addition, the book has a lot of footnotes. The good news is that audiobook listeners get access to an exclusive PDF that contains these charts. But we also called a lot of audibles. Sometimes, a chart or footnote was redundant enough that we skipped it. But others seamlessly enough (we hope) work their way into the narrative, and we talked our way through them.

Some people have a strong preference for physical books, eBooks or audiobooks, and we think On the Edge is a great product either way. The benefits of text are more obvious, I suppose. There’s the browsability of the book — there’s a very detailed glossary, for instance, and I know some readers will want to skip some chapters (it’s a long book) and focus more on others. On the Edge is a hybrid of a memoir, a reported book and an “explainer” book, and the more technical sections probably work better in text. Plus, the physical book has an extremely attractive cover — I don’t know how Penguin found the shade of green they did, but it looks really nice.

But there are some advantages to the audiobook, also. Narrative sections and sections in which I quote directly from sources probably work better, for instance — it’s easier to insert “personality” in verbal speech. With one or two exceptions, I wasn’t trying to imitate my sources, but I’ll sometimes give you a little flavor — whether someone is a fast-talker or a slow-talker and whether they tend to speak with a lot of inflection or relatively little.

Speaking of fast-talking, I tend to speak quickly myself, a leftover habit from doing high school debate, so a recommendation from a good friend who moonlights as a voice actor is that you listen to the audiobook at 1.0x speed. We retook sentences where I spoke too quickly, but my rhythms are usually on the faster side, as you can hear on the podcast.

There’s also one last anti-browsability factor. In writing On the Edge, I was aware that it was a long book where some readers will want to skip back and forth between sections. But it’s also a cumulative book where there’s a fair amount of world-building. One totally fair criticism I’ve seen in user reviews is that On the Edge should really be two or even three books.⁴ But I think if you read or listen to the book straight through, it works pretty well and you’ll see why these subsections are tied together. And that perhaps comes across most clearly in the audiobook version — although if you’re totally on the fence, Devin, my publisher probably cares the most about print hardcover sales.

Ross asks:

Question for next time:

You mentioned the NFL model as a potential first sports model that might return. Any thoughts on the ordering for the others? Do you have the capacity to having all of them (NFL, NBA, MLB, NHL, march madness and i think you had a huge soccer one) running once you get them going? Or are we unlikely to see them all return with just you and Eli supporting.

Let’s zoom the lens out here. I feel incredibly lucky for how the newsletter and everything else is going. Often when people say things like “this is going way better than I expected,” it’s just false modesty. But in my case I have some receipts. 

When my contract with Disney ended, I spent most of my days working on the book, but also took some time to figure out what I wanted to do with the models and even whether I wanted to run a public-facing election forecast at all. My agents and I explored a lot of options. Partnering with another major media brand was one possibility, and we did get one firm offer and a lot of feelers. But this was the least appealing choice to me personally — too much chance of replicating the problems at Disney — and was also happening against a bearish climate for media investment.

Getting some capital and trying to turn Silver Bulletin/The Intellectual Property Formerly Associated with FiveThirtyEight into a “real business” was another option, and we had some very positive discussions about it. We even got one back-of-the-napkin term sheet. But the timing wasn’t really right either on my end or from the investors’ standpoint.

Finally, we extensively explored exclusively licensing the election model to various financial-sector clients, meaning that there wouldn’t have been a public version of it. This required me to make some wild-ass-guess estimates of how much money we’d be giving up in subscription revenues. What we heard back fairly consistently is that, while people were interested in my consulting time⁵, they didn’t want to pay the additional amount we were asking for exclusive rights to the model.

Well, it turned out that I considerably underestimated how well this newsletter would do as a subscription product. By a lot — like, already by a factor of 3x, and we’re only in August. It was a blessing in disguise that none of those clients met our price.

In a few hours, this newsletter will cross the 167,000 total (free + paid) subscriber mark. We’ve been netting about 1,800 subscribers per day since we turned the election model on. I suppose I don’t want to reveal the paid number, but we have a high ratio of paid-to-free by Substack standards. (We’re the #4 Substack in our category, and from what I can gather these lists are primarily sorted by annualized paid revenues.) It’s all a little misleading because we’re almost certainly going to have a lot of churn after the election, but it will be churn from a high base with the ability to regrow in the future.

I really appreciate all the readers who helped us to make this happen. And by “us”, I’m no longer referring to my fancy agents but instead just me and Eli (and my partner, who helps out a lot in the background, and the friendly folks at Substack, who provide good advice too). You’re looking at the entire Silver Bulletin editorial team right here:

If you survey the other Substacks that are in our general range on the list, they fall into two rather distinct categories:

• On the one hand, there are publications like The Free Press, The Bulwark, and Zeteo, which in some ways are just traditional media startups with a subscription model that use Substack as a publishing platform. The Free Press, for instance, has more than three dozen staffers, offices on both coasts, and lots of different product lines — it would be a real stretch to call it a “newsletter”.

• On the other hand, there are traditional email newsletters oriented around a single writer, such as The Weekly Dish (Andrew Sullivan), Letters from an American (Heather Cox Richardson) and Slow Boring (Matt Ygelsias). These are relatively lo-fi publications with perhaps a small support staff. Maybe you can draw a further distinction within this category. Of the three newsletters I mentioned, for instance, Yglesias operates at a much faster velocity (five or more very detailed stories a week) and my sense (from knowing Matt and Andrew and reading about Richardson) is that Yglesias is doing a lot more to optimize his business within reasonable constraints, whereas Sullivan and Richardson prefer more work-life balance.

If I were 10 years older, I’d probably take the Richardson/Sullivan approach. Keep things on at a low boil outside of election cycles, mostly focusing on essays. And even future elections would be an open question depending on how this one went. (Were the polls way off again? Was I feeling burned out? If Trump lost, were his efforts to steal the election more successful than in 2020?) I wouldn’t do much if any further hiring, and I might do the rounds again to find a third-party buyer for the sports models.

And if I were 10 years younger, I’d be thinking about how to scale up Silver Bulletin into a business that could eventually have a shot at lucking into⁶ an obscenely high valuation. Media is a tough business, but we already have a highly positive cashflow situation and total proof of concept given how well the election and basketball models did. (The NCAA forecast tripled paid subscriptions basically overnight.) What would give me pause, of course, is the experience of FiveThirtyEight @ Disney. But more and more, I realize the burnout I was feeling at Disney was mostly because of the lack of having any business model, and the inability to capture any upside or really be in charge of our own destiny at all. We were basically just a barnacle clinging to the underbelly of Disney’s blue whale.

While things could change — if we have the forecast at 90/10 Harris on Election Day and Trump wins, or vice versa, that’s going to sting a little bit — at age 46 I’m probably inclined to split the difference somehow. I no longer have any ambitions to have “data journalism” conquer the world and I’m not even sure what I’d do with my life if there were eventually some exit to a big media company and I had a lot of cash but nothing much to do. However, I still have a lot of creative energy and feel like I’m doing some of the best work of my life. And we’re clearly off to a very good start.

The trick is that there isn’t any particularly good comp for this middle-ground option, at least within the Substack politics ecosystem. The first step is to make two further hires: 1) Eli But For Sports and 2) some sort of jack-of-all-trades editor to professionalize the operation. Hiring is time-consuming, though — at least the way I do it — so that isn’t going to happen until after the election. And beyond that, there are various other constraints to consider:

• Recognizing that a lot of you are here for me. One thing I’d want the editor to do is commission more freelance guest stories like this one, and of course we’ll want Eli and Eli But For Sports writing for the site. But there’s a threshold beyond which Silver Bulletin becomes a different product if, say, more than half the stories on the site are written by someone other than me. In real life, I get along well with most people, but I realize that editorially speaking I’m a sharp flavor that doesn’t necessarily pair easily with others. That was true at FiveThirtyEight @ Disney, for instance. On the one hand, the staff understandably wondered why I was allowed to write with more “voice” than they were. On the other hand, I often felt like I was holding myself back — hence, my hot takes were often outsourced to Twitter — and we occasionally published stuff (for instance, on COVID) that I pretty strongly disagreed with. Plus, there are already several product lines we’d be trying to manage even with a 4-person staff: paid versus free, models versus other content, politics versus sports versus “other”, opinion writing versus analysis, community-building. I just have to be honest with myself that the thread that ties those things together is me. Perhaps there are ways to scale up while maintaining that balance  — in particular, by curating a more Riverian set of topics⁷ — and I definitely need to think more about Nate-gets-hit-by-a-bus scenarios/succession planning. But a business where most of the bylines are by someone other than me is a pretty big leap and one it would probably take a couple of years to make happen, if it ever happens.

• Needing downtime. Yglesias and I both have the gift of being able to write decent copy quickly — probably the single most valuable skill for any Substacker. But otherwise he clearly has much more of a routine than I do. He publishes at 6 a.m. every weekday, whereas I’m more cyclical and have bursts of intense focus. This is both out of necessity — elections (and sports) are cyclical — and because it just suits my personality and my lifestyle. I’m a much happier person when I have time to work on long-term projects, like a book (yes, I already have an idea for another book) or a new model. And I travel a lot, for work, fun and poker. I can’t have a business model that depends on me constantly being logged on — I’ll eventually become miserable, and it will show in the quality of the work.

• Avoiding overstaffing. Management skills aren’t my comparative advantage, but I think I do reasonably well with people one-on-one. However, I strongly suspect there are inflection points past which the internal politics of an organization become a thing unto themselves, a Thing That Needs To Be Managed whether you want to or not. I don’t know where that threshold is, exactly. But it’s probably a max of 5 or 10 people, not 50 or 100.

• Maintaining quirkiness. In the early days of Silver Bulletin when I was sending emails to “only” 16,000 people, I figured there was nothing to lose with offbeat posts like this one, so I was being more experimental than I had been in a long time. Now that each email is going out to 160K+, the threshold is a little higher. You’ve probably all been to a restaurant where the chef had a special version of a classic dish that was actually a total fail, or a concert where the band played its self-indulgent new material instead of playing the hits. It’s annoying. But I don’t want the newsletter to be so polished or so audience-captured that there’s no room for experimentation. For instance, I’m low-key a big fan of the hockey team, the New York Rangers, and if the Rangers make a huge trade, I’d like to be able to write about that and have it not feel weird, even if I know it would probably only interest 2-5% of the subscriber base.

All right, so we’re almost 2000 words in, and I still haven’t answered the question about the sports models. But there are lots of other factors to consider, too. How differentiated is a model from the competition? How much demand is there for it from both current and potential subscribers? How much maintenance does it require and how easily does it “break”? Does it provide for a natural dividing line between free and paid content? Is a model out of date to the point where it’s obsolete and I need to rewrite it? Also, there aren’t just the sports models — there are also politics models other than the presidential election forecasts.

But here are the models that are pretty clearly over the threshold and that we’re likely to prioritize:

• College basketball. The March Madness predictions did really, really well — both as a subscription driver and against Vegas — and I’d be foolish not to run them back and to be a little more organized about it next time around. Also, the NCAA model depends in part on college basketball Elo ratings that I think are pretty cool and we should probably be publishing year-round — for some reason, we never did at FiveThirtyEight — and possibly expanding to women’s hoops. This will probably be Eli But For Sports’s first project and one of the first things we’ll get up on the site after the election.

• NFL. It’s by far the biggest sport, especially since this newsletter has a somewhat, uh, gambling-adjacent audience. And the schedule — NFL games happen in one big burst on the weekend, and every weekend is a big event — is great for newsletter content. (Maybe we’d even experiment with a weekly NFL column.) Our quarterback-adjusted model was pretty good, but I have some ideas in mind for further tinkering. I originally had some inkling to try to get this ready for the end of the forthcoming NFL season, but I just don’t think that’s realistic. More likely, it will be one of my big summer projects, hopefully fully ready to go for the 2025 season.

• NBA / RAPTOR. The NBA is my favorite sport, so I want to turn this back on just for my own edification. And it was one of our most differentiated models — RAPTOR wasn’t always right, but it tended to have interesting opinions. The hope would be to get this back up and running, probably in stages, for the second half of the forthcoming (2024-25) NBA season. There are also a lot more seasons of historical NBA player tracking data than when I first designed RAPTOR in 2019 and some other things I’d like to reexamine, so it probably needs a reboot at some point. (RAPTOR 2: This Time With More T-Rexes And It Doesn’t Think Nikola Jokic Should Be The Defensive Player of the Year).

• Midterms forecasts. Our Congressional and gubernatorial model is actually a lot more complicated than our presidential model — there are far more candidates to keep track of and far more data sources to wrangle — so this is more work than you’d think. But the model had a really good year in both 2018 and 2022 and it would be silly not to turn it back on unless we’re totally moving away from election forecasting.

• Presidential approval ratings. These got reasonably good traffic and the code is simple and easy to maintain, so this is kind of a no-brainer as Eli’s first post-election project. The basic approval ratings would be free, but we might look at tracking approval ratings among demographic subgroups (e.g. independent voters) as a premium upsell.

• Generic Congressional ballot. We need to calculate this for the midterms model anyway, so we’d almost certainly also turn it back on as a standalone provided that we restore the midterms forecast.

But these models are a tougher call:

• College football. Our college football playoff forecast was one of my favorite models, and the sport may be on the verge of an even bigger breakout between the new 12-team playoff and even the new video game. But the playoff format has changed, we have a lot more data now on what the selection committee cares about, and the model had involved a lot of patchwork repairs (such as a hack-ish adjustment to account for Notre Dame not being in a conference). So this probably needs to be rewritten from scratch. Zero chance of this happening for the upcoming season, but I’m pretty confident it would drive a lot of subscriptions if we got it ready for 2025 or 2026.

• Soccer. The soccer model was quite popular, but was complicated on the backend given the number of different leagues that it issued forecasts for. It’s possibly more work than we could do just with me + one Sports Eli. And while I’ve always had aspirations to get really into club football, I can’t say I have the domain knowledge of the game that I do for the major American sports. So this might take longer, and it’s also the one case where it might make more sense to sell the model to some third party, whether another media company or even a team. (Drop me a line if you’re interested.)

• Baseball. Some negatives here: I’m following baseball less than I used to, it was less of a traffic driver than football or basketball or soccer, and the baseball model wasn’t as differentiated as some of the others. However, it’s comparatively simple — Elo plus a starting pitcher adjustment — which helps. This might depend on what Eli But For Sports wants to do.

• NHL. I didn’t write this code so I don’t have any rights to the IP. I love the sport, but I’d have to write a new model, and this is probably the lowest-priority item on the list.

• Presidential primaries model. This was a very complicated model, undoubtedly the most mathematically complex model that I ever designed. It tried to account for things like the knock-off effects from one state to the next one, translating vote shares to delegate projections, and how to estimate correlated and asymmetric error terms in races with multiple candidates. I’m just not sure about this one, and I’m glad I have three years to think about what to do with it.

• Polling averages other than presidential approval. Meaning, things like approval ratings for the vice president or special elections forecasts. These are sort of low-risk and low-reward (the code is simple but the audience isn’t huge). We’re currently not trying to be a “polling aggregator of record” at Silver Bulletin like we were at FiveThirtyEight, so this depends on staffing levels and what the competition is doing. 

• Congressional vote tracking. Also known as Trump Score or Biden Score — how often each member of Congress voted with the president. I thought this had a lot of public service value — indeed, our numbers were often cited in campaign commercials — but one big disadvantage is that it can’t be fully automated: someone had to manually annotate bills on which the president had taken a clear position. Not a huge priority, but it depends on what Eli wants to do.

• Congressional forecasts in presidential years. In principle, this is a layup: it’s the same code as the midterm model. But as I said, the Congressional forecasts require a lot of marginal effort, such as keeping track of detail about the candidates in roughly 470 distinct races. And time in a presidential year is precious. So again, this is a question of staffing and tradeoffs. Probably worth it, but I’m also really glad we didn’t try to do it this year.

• Economic index. The economic index is a component of the presidential election forecast, but it isn’t a lot of work to run it continuously and I think it has some value on its own as a comprehensive gauge of how the economy is doing. It wouldn’t have a huge audience, but we could make some cool graphics, and if we’re being honest anything finance-related tends to convert pretty well from free to pay. So maybe I’m talking myself into this being a good three-week project for Eli.

OK, that was a far more comprehensive answer than Ross or anyone else was probably looking for. But the whole radical transparency thing has worked out pretty well so far. The tl;dr version is that it’s quite likely some of these models will be turned back on — nothing will be ready immediately post-election, but without wanting to overpromise, within the time frame where you might get some additional value out of an annual subscription that you sign up for now.⁸

Happy Sunday, and don’t forget to leave questions for SBSQ #12 in the comments.

1. Note, however, that charts from the election model render somewhat awkwardly on the app and are primarily designed for the web.

2. While we love our Founding Members, this is still subject to usual norms of civility. If you’re abusive, we’ll comp you a premium subscription for one year and refund the difference, but take away the ability to start threads.

3. This originally claimed that “World Series of Poker” was fewer syllables than “WSOP”, but think I was pretty generously trying to count “series” as one syllable. And maybe “world”, too.

4. A book about gambling (Chapters 1-4), a book about Silicon Valley and the battle between “the River” and “the Village” (the intro, Chapter 5 and the concluding material) and a book about Sam Bankman-Fried and his odd intersection with effective altruism (Chapters 6-8).

5. Indeed, we did sign one deal, but it’s just for my time — the client doesn’t have any access to the model.

6. If and when several of these models are back up and running, we might look again at pricing or even having different pricing tiers. But keep in mind that at Substack generally and Silver Bulletin specifically, you get to lock in whatever price that you signed up for.

7. I’ve been around the industry long enough that I do think there’s a lot of luck in this stuff: there are a finite number of plausible buyers and you want the timing of a perceived hot streak to coincide with an upcycle for media investment.

8. For instance, maybe the site covers AI as one of its beats, a topic that I find really interesting, but we take a pass on general political news, which I mostly don’t.