Subscribe now

This is the landing page for Silver Bulletin’s 2026 generic congressional ballot polling average. We’ll regularly update the charts below as new generic ballot polls come in, and eventually, this average will feed into our 2026 midterm forecast model.1 This page is designed to be viewed on the web rather than in our email client or in the Substack app.

Click here for more information on how the average works. The Silver Bulletin average weights more reliable polls more heavily — you can find our latest pollster ratings here.

Who is favored to win the House in 2026?

Our default average reflects a combination of all polls, whether conducted among adults, registered voters or likely voters. If a pollster releases multiple versions of the same survey, we use the likely voter version before the registered voter version.2 That’s because for this average, we’re interested in people who plan on voting in 2026.

Every generic ballot poll in our database

Each poll gets an “influence” score based on its pollster rating, its sample size, its recency, and how often a pollster is publishing numbers. You can find that in the table below. Sometimes, surveys with mediocre pollster ratings have more weight in the model just because they were conducted very recently or polled more people.3

Inevitably, there’s a lot of disagreement from survey to survey, not just because of statistical variation, but because some polling firms consistently lean toward Democrats or Republicans. By clicking on the “adjusted results” tab, you can see how the “house effects” adjustment that corrects for these predictable differences works in our model. You can also click here to download every generic ballot poll in our database — including some additional details not shown in the chart below.

State benchmarks and every generic ballot poll since 1994

We also have two cool features we’re reserving for paying subscribers:

• Benchmarks in each state. Which party would we expect to be ahead in, say, Georgia or Michigan or Ohio, given the current generic ballot?

• And generic ballot averages going back to 1994.

You can find all of that, plus downloadable generic ballot data for the past 30 years, below.

Read more

🕒 The latest on Trump’s approval rating

Updated August 16, 2026

The Silver Bulletin 2026 midterm election forecast launched today! Check out the new landing page to see our predictions for every Senate, House and gubernatorial race this cycle.

The forecast does incorporate Donald Trump’s approval rating. With a current net approval of -20.4, the president isn’t doing Republicans any favors. -EMD, 8/11/26

See also: Midterm election forecast, generic congressional ballot dashboard, and Iran War support dashboard.

Subscribe now

This is the landing page for Silver Bulletin presidential approval ratings. We’ll regularly update the charts below as new polls about Donald Trump’s approval rating come in.1 This page is designed to be viewed on the web rather than in our email client or in the Substack app.

Click here for more information on how the average works. The Silver Bulletin average weights more reliable polls heavily — you can find our latest pollster ratings here.

The topline: So, just how popular is Trump?

Our default average reflects a combination of all polls, whether conducted among adults, registered voters or likely voters. If a pollster releases multiple versions of the same survey, we use the all-adult version of the poll before the registered voter version.2 This is because all Americans have a say in how popular the president is — whether or not they vote.

The polls: What do the surveys say?

Each poll gets an “influence” score based on its pollster rating, its sample size, its recency, and how often a pollster is publishing numbers.3 Sometimes, surveys with mediocre pollster ratings have more weight in the model just because they were conducted very recently or polled more people.

Inevitably, there’s a lot of disagreement from survey to survey, not just because of statistical variation but because pollsters have long had trouble pegging down Trump’s popularity. By clicking on the “adjusted results” tab, you can see how the “house effects” adjustment that corrects for these predictable differences works in our model. You can also click here to download every Trump approval poll in our database — including some additional details not shown in the chart below.

The issues: What do Americans think of Trump’s policies?

Pollsters don’t just ask about Trump’s overall job approval, they also ask about how well he’s dealing with different issues. We’re tracking how the public feels about Trump’s handling of four topics: the economy, immigration, trade and tariffs, and the cost of living.

The deep dive

We also have a few features we’re reserving for paying subscribers:

• How do Trump’s approval ratings compare to every past president since Truman?

• How many voters strongly approve or disapprove of Trump?

• What are Trump’s numbers in only polls of adults or only polls of likely and registered voters?

You can find all of that, plus downloadable data on every president’s approval rating every day since 1945, down below.

Read more

Subscribe now

This is our landing page for polls about the Iran War1, tracking support for U.S. military action in Iran since March 1, the day after the war began.2 We’ll regularly update this average as new polls come in.3 This page is designed to be viewed on the web rather than in our email client or the Substack app.

Click here for more information on how the average works. The Silver Bulletin average weights more reliable polls more heavily — you can find our latest pollster ratings here. Our process for calculating support for the Iran War is similar to that for calculating President Trump’s approval ratings.4

The topline: So, just how popular is the Iran War?

This average includes polls that ask about support for the Iran War, strikes in Iran, or U.S. military involvement in Iran. Compared with tracking approval ratings, questions about war can be less straightforward. Our goal is to limit the analysis to polls that ask neutral questions rather than introduce assumptions about how the war is going or its objectives. (See below for more details on how we make this assessment.) Our default version of the ratings reflects a combination of all polls, whether conducted among adults, registered voters, or likely voters.5

Share

The polls: What do the surveys say?

You can see all the numbers for each poll that feeds into our average below. You can also click here to download all the Iran War numbers.

We exclude questions that:

1. Make presumptions about the reasons for or against the war and/or offer an assessment about whether these objectives have been met (even if respondents are offered arguments on both sides);

2. Ask how the war is going relative to the respondent’s expectations (i.e., “Is the war going too far”?);

3. Ask about President Trump’s approval on Iran, as opposed to support for the overall war operation6, or;

4. Ask about some particular aspect of the war (e.g. targeting nuclear facilities) rather than the overall situation.

Also, in line with longstanding Silver Bulletin policy, surveys that fail to provide the specific question wording or other key details about the poll (dates, sample frame, sample size) can’t be included.

Polls are adjusted for house effects (i.e. persistent differences between a pollster’s findings on the war and the consensus) and are weighted based on each firm’s pollster rating and the poll’s sample size and recency. The “influence” score in the table below shows the relative importance of a poll in our current average given these considerations.7 In practice, recency is often more important than the other factors in our weights.

As a reminder, we update this database about once per day. If a poll seems to be missing, we may not have added it yet, we may be awaiting more information from the pollster, or it may contain question wording that doesn’t meet our standards for this project.

Subscribe now

In case you missed it, we launched our 2026 election forecast — now named FLIPR — on Tuesday. On balance, the model is bullish on Democrats’ chances in November. But while it currently rates the House as “likely D”, the Senate is still in true tossup territory.1

Why is the upper chamber so competitive? Well, it’s pretty simple. True, the GOP has few solid pickup opportunities, but they already have 53 senators and a tie-breaking VP. Democrats need to net at least four seats, mostly from states that Donald Trump carried handily in 2024. If the national popular vote comes in anywhere under D +6, Republicans are the clear favorites to retain the Senate.

But Democrats are on track to beat that number, perhaps by a wide margin. As of today, our forecast for the popular vote is D +7.2. That’s a big enough blue wave to give Democrats a real shot at winning enough competitive seats.

Some of the data can also seem contradictory, however. Just this week, Fox News released polls showing Democrat Sherrod Brown leading by 8 points in Ohio — but Abdul El-Sayed trailing by 4 points in Michigan. In the context of a presidential race, that wouldn’t make much sense, since Michigan is the considerably bluer state. Senate races are different, however. Particularly if you believe that moderation matters, El-Sayed is arguably a weaker candidate than Brown.2 Brown is also a more familiar face, having served in the Senate from 2007 through 2025.

Outcomes in congressional races are correlated with one another, although considerably less so than in presidential races. Still, the more advanced versions of FLIPR are a little bit skeptical about taking these polls at face value. It’s not that they’re necessarily wrong about what might happen in an election held today. But El-Sayed has several months to repair the damage from a bruising Democratic primary, while Republicans have time to bring the race more in line with fundamentals in Ohio.

So for my first model talk of the 2026 cycle, I’m bringing back a popular article format from 2024 and taking you on a tour of the seven most likely Senate tipping-point states (plus some honorable mentions at the end). We’ll look at where polls and fundamentals are in tension with one another, like in Michigan, and where they’re more aligned. Combined, these races have a 71 percent chance of deciding which party controls the Senate after 2026.3

Full disclosure: we’re cheating just slightly here. In terms of tipping-point probabilities, there’s really a Big 6 rather than a Big 7, with a drop-off after the 6th-most-likely tipping point state, Alaska. Then there’s another tier of roughly half a dozen races, each with around a 4 percent chance of being the tipping point.

For purposes of this article, we’ve promoted North Carolina from that tier into the Big 7. Our model doesn’t think NC is a particularly likely tipping point because it’s pretty confident that Democrat Roy Cooper will win, perhaps comfortably. Still, nobody should take North Carolina too much for granted: it has only voted for Democrats once for president since 1976 (it went for Barack Obama in 2008) and has frequently been a close-but-not-quite state for Democrats in federal races. So it deserves some attention as we explain why FLIPR thinks this might finally be the year for a Democratic breakthrough.

But let’s begin with Democrats’ white whale: Texas.

Read more

This is the landing page for FLIPR1, Silver Bulletin’s 2026 election forecast. We’ll regularly update the charts below as new election polls and other data come in.2 This page is designed to be viewed on the web rather than in our email client or in the Substack app. If you’re curious about how the model works, and what’s new for 2026 (quite a lot, actually), we also have a very detailed methodology article.

The polls

The Silver Bulletin polling averages are the first step in FLIPR, and they include various adjustments. They account for whether polls are conducted among registered or likely voters, and for house effects. They weight more reliable polls more heavily. They also include a “timeline adjustment”, where outdated polls are revised to reflect trends in the generic ballot. Speaking of the generic ballot, you can find a version of it with our likely-voter adjustment below; this is the version that FLIPR uses.

We draw most of the election polls used in our model from The New York Times poll database3, but collect generic ballot and congressional approval polls ourselves.

Share

The forecast

Here are today’s topline numbers for Senate and House control. They are typically based on 40,000 simulations and include a small amount of sampling error. FLIPR models House, Senate and gubernatorial elections jointly, meaning that we can calculate the probability of things like Republicans holding the Senate but losing the House.

Read more

Read more

Subscribe now

This the landing page for the Silver Bulletin average of Elon Musk’s favorability polls, tracking him back to the start of 2024.1 We’ll regularly update this average as new polls come in.2 This page is designed to be viewed on the web rather than in our email client or in the Substack app.

Click here for more information on how the average works. The Silver Bulletin average weights more reliable polls more heavily — you can find our latest pollster ratings here. Our process for calculating Musk’s favorability ratings is similar to that for calculating President Trump’s approval ratings, although we use slightly more conservative settings as Musk is polled less often than Trump.3

The topline: So, just how popular is Elon Musk?

Our default version of the ratings reflects a combination of all polls, whether conducted among adults, registered voters or likely voters. If a pollster releases multiple versions of the same survey, we use the all-adult version of the poll before the registered voter version.4

Share

The polls: What do the surveys say?

You can see all the numbers for each poll that feeds into our average below. You can also click here to download all the Musk favorability numbers — and some additional details not shown in the chart below — for every Musk favorability poll in our database.

This is a free feature, but it requires work to update and maintain. We very much appreciate your support via paid or free subscriptions.

Our midterms model, which we’re now calling FLIPR 🐬 — Forecast with Leading Indicators, Polls and (Expert) Ratings — launched today! There’s a lot of detail both on the model landing page and in the accompanying methodology story. We’ll be updating the model nearly every day, and will have lots more analysis to come.

FLIPR is in the lineage of the f/k/a FiveThirtyEight election forecasts1, but we’ve also put a lot of obsessive work into refining and improving it. We greatly appreciate your patience as we’ve gotten everything ready.

Subscribe now

Some elections fill me with a sense of existential dread. Even the most carefully designed models, which dutifully consider not only how to forecast each state but also how to calibrate the degree of uncertainty in the forecast, will have trouble evaluating the impact of a once-in-a-century pandemic like in 2020 or a major party switching candidates midstream like in 2024.

The next tier down from this are elections that seem relatively normal on the surface, but where typically reliable indicators are contradictory. In 2022, President Biden’s deep unpopularity and the long history of the president’s party performing poorly in midterm elections suggested the possibility of a red wave. Polls of individual races, however, and special election results, were much more equivocal, perhaps reflecting the degree to which the Supreme Court’s decision to overturn Roe v. Wade had galvanized Democrats. In the end, the race-by-race polls were basically right.2 Democrats pared their House losses to just 9 seats and actually gained a seat in the Senate, but it wasn’t an easy election to forecast.

The 2026 election is not in either of those categories. It looks like a pretty normal midterm, to the extent anything can be normal in American politics these days. And normal midterms mean bad news for the incumbent president’s party, especially when the incumbent is as unpopular as President Trump. Polls, historic trends and leading indicators are all fairly consistent in pointing toward Democratic gains.

There is, of course, still considerable uncertainty in the outlook. We’re still 84 days out from the election, for one thing. The uncertainty is something that inherently probabilistic models like our midterm election forecast, a.k.a. FLIPR, are specifically designed to grapple with.

In the House, Republicans basically won the war in an unprecedentedly aggressive round of mid-decade redistricting, which left the map with a mild Republican bias. That gives Democrats a hurdle to clear. Not to be overlooked: extremely partisan maps drawn by both parties have greatly reduced the number of competitive districts in the House. Even if Democrats have a great year — say, they win the national popular vote for the U.S. House by 9 percentage points, which would qualify as a landslide by modern standards — they’ll have trouble matching the 41-seat gain that they had in 2018.

In the Senate, the hurdle for Democrats is much higher. The current composition of the Senate is 53 Republicans and 47 Democrats3, and Vice President JD Vance casts the tiebreaking vote. That means Democrats need to gain a net of 4 seats to take control of the chamber.

In a roughly neutral, presidential-year-type political environment, that would be a nearly impossible task, requiring gains deep into Republican territory. Moreover, it’s plausible that Democrats will have to pick up 5 Republican seats, not 4, because Abdul El-Sayed currently trails Mike Rogers in the polling average in Michigan, a state that Democrats are defending.

But every available indicator suggests that this very much isn’t going to be a neutral political environment. The Democratic Party has a whole host of problems, which we chronicle obsessively here at Silver Bulletin. Turning out their voters in midterms isn’t one of them, however.

And for midterm election purposes, you’d rather have a little internal disarray than have a president with a 38 percent approval rating on the basis of an unpopular quagmire in the Middle East and understandable voter anxiety about spiking gas prices. Not to mention voter fatigue over an 80-year-old, lame duck president at a time when the incumbency “advantage” has turned into persistent anger all around the world toward the status quo and whoever is in charge of it.

Let’s get into more specifics about these indicators, and then we’ll discuss why FLIPR is spitting out the probabilities that it does.

Polling, history and recent elections tell the same story

Our generic ballot average, which we update every day, currently shows Democrats ahead by 6.7 percentage points. Silver Bulletin readers and other election nerds can become obsessed with every little zag in the polls, but our version of the generic ballot has been fairly stable for several months now.

However, that 6.7-point margin might actually lowball Democrats. That’s because most generic ballot polls at this not-exactly-early-but-not-exactly-late point are being conducted among registered voters rather than likely voters — and therefore, don’t account for turnout. FLIPR makes a likely voter adjustment. In line with our longstanding practices, the adjustment is based mostly on polls that release both likely and registered voter versions of the same survey. Typically in such polls, the LV version has been better for Democrats than the RV version. Thus, FLIPR’s version of the generic ballot shows Democrats ahead by 8.1 points instead.

How about the longer sweep of history? Well, FLIPR has a calculation for that, too. We estimate a “prior” based on all congressional elections since 1946, calibrated based on the president’s approval rating, the result of the previous election, and whether the election was a midterm, and adjusted for ever-increasing partisan polarization. In that calculation, Democrats are favored to win the popular vote for the U.S. House by 8.7 points instead.

Trump’s approval rating? It’s now roughly -20. And contrary to the media narrative that tends to view Trump’s support as ironclad, even some of his base is wavering; only 22 percent of Americans strongly approve of Trump’s performance in office. Republicans who feel ambivalent about Trump probably won’t vote for Democrats — but they might not vote at all.

What about polls of individual races? That story is slightly more complicated. There are a few rough patches for Democrats. El-Sayed, for example, currently trails Mike Rogers by about a point-and-a-half in our polling average for Michigan — though the more advanced versions of our forecast deem him a favorite to eventually catch up. But Jon Ossoff leads by 9 points in Georgia, a state where Democrats rarely win anything by more than a point or two. Roy Cooper leads by about the same margin in North Carolina. And James Talarico is slightly ahead in Texas. Even Iowa is currently polling as a toss-up.

Don’t trust the polls? Well, in the last major set of elections in November 2025, Democrats had basically a perfect night. Polls underestimated them, and their margins in Virginia, New Jersey and other states were consistent with a blue wave.

But isn’t the DNC having a fundraising crisis? Yes, but that’s partly because Democratic donors are finding better ways to spend their money. Among Senate candidates, the top 7 in fundraising so far this cycle are all Democrats.

To repeat: Democrats certainly have plenty of problems. I’m one of those people who thinks the empirical evidence strongly points toward more moderate candidates being more electable. In addition to El-Sayed, our polling average currently has Francesca Hong in a tossup in the Wisconsin gubernatorial race — her primary is tonight — and our model is much less confident that she’ll revert to the mean because gubernatorial races are still considerably less partisan than races for Congress.

Democrats are clear favorites in the House; the Senate is roughly a toss-up

Read more

In the Silver Bulletin/FiveThirtyEight tradition, we’ve always given names to our sports models (PELE, ELWAY) — but never to our election forecasts. I suppose we’d like to think that our presidential model is renowned enough at this point that it doesn't necessarily need any further identifier. But we’ve decided we do need a name for what we informally call our “midterms model.”

That’s because “midterms model” is actually a misnomer. The model forecasts both Congressional and gubernatorial races, but it’s trained on both midterm years like 2022 and years where these races coincide with presidential elections.

“House, Senate and gubernatorial model” is a mouthful, however. So now we’re calling our model FLIPR 🐬 instead.1 In line with our tradition of silly backronyms, FLIPR stands for Forecast with Leading Indicators, Polls and (Expert) Ratings.

We like this name because it’s sort of a triple entendre:

• The principal goal of these elections is flipping individual seats, and ultimately the House and Senate.

• FLIPR is a hat tip to the inherently probabilistic nature of the forecast — we run 40,000 simulations every time the model is updated. It’s not as simple as “flipping coins”, but the simulations are necessary because of the complex relationships between different types of races.

• And the FLIPR acronym is also rather descriptive. FLIPR essentially consists of three different layers stacked on top of one another: polls, non-polling leading indicators (e.g., fundraising, incumbency) or what we sometimes call “fundamentals”, and then a final layer of race ratings from longstanding outside expert groups like the Cook Political Report.

In fact, there are three versions of FLIPR that correspond to each layer:

• Lite derives as much information as possible from polls alone;

• Classic is polls + fundamentals;

• And Deluxe is polls + fundamentals + expert ratings.

While Deluxe is our default version, almost all of the charts and tables on the FLIPR landing page allow you to, uh, flip back and forth between these versions. The way FLIPR is designed, the expert ratings layer is probably the least important — the proverbial “icing on top.” But if you want a forecast without any expert ratings, we have you covered with Lite and Classic.

Subscribe now

FLIPR is part of the FiveThirtyEight model lineage

FLIPR has a new name — but is it a new model? I’d say probably not. If you’re a former FiveThirtyEight reader — I retained the rights to all of the models I designed for them per my deal with Disney — you’ll recognize that some of what I’ve just described sounds familiar. In fact, some of the code in FLIPR just celebrated its 16th birthday, having survived since the first midterms forecast I built back in 2010 when FiveThirtyEight was stationed at the New York Times.

We think this track record is one of the things that makes FLIPR unique — competing models come and go, succeed and fail, but few have been around long enough to truly have any sort of long-term track record. And our midterms forecast performed quite well in both 2018 and 2022.

However, FLIPR and its predecessors have a lot of moving parts. And we’ve made refinements every year. While the basic foundation is similar to 2010, nearly every component of the model has been swapped out for a new version at some point. It’s a little bit like the paradox of the Ship of Theseus: if every component of a model has been upgraded at some point, is it still the same model?

The model underwent a near-complete refit in 2018 when the Lite/Classic/Deluxe concept was introduced. It also underwent another suite of changes in 2022 to reflect increasing political polarization. And even though we really liked the 2022 version of the model, we wound up making more changes and improvements this year (2026) than I was originally expecting — rather than just “turning it back on.”

These changes have a common theme: they don’t change the underlying structure or philosophy of the model so much as seek to execute on our original design with more precision, robustness, and internal consistency.

I have complicated feelings about AI coding tools — and I very much disagree with the idea that they’re a substitute for human judgment. But as a complement to human judgment, they’re pretty nice. Probably, in fact, they’re better for reviewing and refining an existing model than for building one from scratch. Basically, every single line of code from our 2022 model2 has been audited and, in many cases, improved.

Many of these improvements are not highly visible: they concern things like faster program runtime, better error detection, and better “code hygiene” that makes FLIPR easier to audit and update.

But having software that speeds up some of the most tedious aspects of coding also freed up time to test out different hypotheses and clear a long list of items off of our punch list. Some of these upgrades fell under a longstanding category of “would be nice to have, if only we had the time.” For example, FLIPR now explicitly simulates ranked-choice voting, as used in Alaska and Maine. Building that code took a couple of hours and relied heavily on our expertise; it’s not nearly as simple as telling an AI coder “ok do ranked-choice voting now.” It probably would have taken a couple of days instead of a couple of hours in the world before Claude and ChatGPT, however.

Still, ranked-choice voting is a fairly low-stakes problem since the other 48 states don’t use this process. Other questions we asked of the data are more consequential. For instance, as elections become more predictable on the basis of overall partisanship with less ticket-splitting than ever, while at the same time the polls have had some rough years3, should our model increase the relative weight on “fundamentals” as opposed to polls? The answer is basically yes4, and FLIPR now does that. Another question: does the political environment now “lock in” more quickly, such that indicators like the generic ballot are more predictive even months ahead of time than they once were? Again, the data says “yes”. In an era of intensive partisanship, races don’t necessarily shift that much between August and November, and there are fewer unexpected shifts. FLIPR now accounts for that, too.

Why forecasting midterms is hard

As we’ve said, FLIPR forecasts congressional and gubernatorial races based on a combination of polls, fundamentals and expert ratings. The model is fit primarily based on Senate races since 1990, and House and gubernatorial races since 1998.

Compared with our presidential forecast, which mostly defaults toward being “polls only” by election day, FLIPR is more of an even blend between polls and fundamentals (and in Deluxe, expert ratings, though these have less influence than the other major components). This is both because non-polling data is richer for House, Senate, and gubernatorial races than for the presidency — there are around 500 races combined in each midterm, rather than just one presidential race — and because polling data is sparser. Many important House races receive little to no nonpartisan polling, for example, so we need other methods to forecast these races.

I’ve emphasized the multi-layered nature of FLIPR: polls, fundamentals, expert ratings. But there’s also another dimension that cuts at a 90-degree angle to this. In FLIPR, there are both “micro” indicators that apply to one election (e.g., the polls in a particular Senate race, or how much money each candidate has raised in that state) and “macro” indicators like the generic ballot and polarization5 that affect all of them.

The middle zone between the micro and the macro is a rich and sometimes turbulent environment, however. You can’t just forecast each race individually and apply some sort of Monte Carlo simulation on top because race outcomes are somewhat correlated. There are years like 2016 when a party (in that case, Republicans) beats its polls across the board in a way that indicates systemic bias in the forecast rather than variance in individual races.

Still, the correlation among congressional races is less than that in presidential elections — the major difference being that you don’t have the same candidates on the ballot in every race. In fact, this is especially true in midterm years as opposed to presidential years, when the presidential race serves as more of an anchor on everything. The 2018 midterm was strong for Democrats, but also featured pockets of Republican strength, like in Florida or in Senate races in rural states that ousted Democratic incumbents. The 2022 map was patchy, too, with essentially a mini “red-wave” in New York and New Jersey that didn’t extend to most other parts of the country.

The rest of this document is more practical and less philosophical. I’ll explain how FLIPR works in seven high-level steps. Then I’ll give you a fairly exhaustive list of what’s new or different in the model in 2026.

FLIPR in 7 Steps

Step 1. Calculate polling averages for individual races and the generic ballot

The common thread between many of our politics models is our polling average algorithm. The code we use to calculate our Trump approval ratings and generic ballot averages is mostly the same6, for instance. And mostly the same process is also used in FLIPR:

• Polls are weighted based on recency, sample size, and their Silver Bulletin pollster rating;

• The polls are adjusted in various ways, such as for house effects;

• We take care to ensure that the model responds to new polls well, neither being “too bouncy” or too impervious to new evidence; we’ve found the best way to do this is to calculate what are essentially various LOWESS regression curves, and then average more aggressive and more conservative versions.

The difference between FLIPR’s polling averages and others like our presidential approval ratings is that FLIPR makes a more comprehensive series of adjustments:

• FLIPR’s house effects adjustment — house effects are persistent differences in polling firms tending to rate Republicans or Democrats more highly — is more robust because FLIPR has more data to rely upon: it can derive these both from the generic ballot and from polls of individual races. It’s also more punitive to expressly partisan or internal campaign polls when these are released to the public, essentially starting out with a prior that such polls are biased.7

• FLIPR applies a likely voter adjustment whereas our regular polling averages do not. Basically, every poll of registered voters or all adults is adjusted toward what we think the poll would say if it were conducted among likely voters instead. This is calculated by comparing polls that publish both registered voter (or all adult) and likely voter versions of the same survey. In the large majority of such surveys this year (2026), for instance, Democrats gain ground on the LV versus the RV version, probably reflecting greater enthusiasm. Thus, RV/adult polls are shifted toward whatever party tends to win these comparisons — so far this cycle, Democrats — although the magnitude of the shift is shrunken toward a prior8 until we have enough such polls that FLIPR deems the prior can be entirely phased out.

• Finally, FLIPR applies a timeline adjustment based on changes in the generic ballot. For example, if a race was last polled in March, and Republicans have gained 3 points in the generic ballot since then, the model will take a fraction of those 3 points9 and add it to the Republican candidate’s margin. Because our generic ballot average uses fairly conservative settings, this adjustment isn’t usually that pronounced. But it does mean that our polling averages in individual races can shift slightly even if there is no new polling in that contest.

So basically, FLIPR seeks to estimate what a poll of a given race would say today if it were conducted on a likely-voter basis. In this way, it hopes to avoid relying on “stale” polling.

FLIPR also applies the likely voter adjustment to its averages of the generic ballot and presidential approval, which our regular averages do not. This makes a material difference. As of early August, for example, Democrats are up by about 6.5 points in our default generic ballot average, but that advantage expands to around 8 points after likely voter adjustments, which is the version that FLIPR uses. (The likely-voter adjusted version of the generic ballot can be found on the FLIPR landing page.)

Step 2. Estimate the national environment and calculate CANTOR similarity scores to derive the “Lite” snapshot of each race

The “Lite” version of our forecast is supposed to be “polls only”. But that creates a problem for races that have little or no polling. Our solution is to impute polling-like estimates for these races based on FLIPR’s calculation of the national political environment, and polling in similar races using an internal system we call CANTOR (Congressional Algorithm using Neighboring Typologies to Optimize Regression).

FLIPR’s estimate of the national environment is based partly but not entirely on the generic ballot. It also calculates a “bottom-up” estimate of what polls of individual Senate and House races imply about the national picture.10 For example, say that the generic ballot has Democrats up by 8 points, but polling of key races for Congress is more consistent with something like a D +4 environment instead. FLIPR will basically take an average of these11 , pegging the environment at something like D +6 instead.

Implicit in the bottom-up calculation is what we call a “partisan lean score12” for each state and district, which is based on how it voted in the past two presidential elections relative to the country overall, as well as in recent state legislative elections. For congressional districts, estimates reflect new boundary lines after redistricting.

Texas has a partisan lean of around R +11, for example. So if, say, Democrat James Talarico leads Republican Ken Paxton in our Texas polling average by 2 points, that reflects a 13-point outperformance relative to the baseline. Individually, that’s a noisy estimate — and Paxton is arguably a poor candidate. But when you apply this technique to all polled Senate and House races, these differences tend to cancel out. (Though the implementation of this is more complicated in practice than in this simplified example.13) Newly this year, the bottom-up estimate also places more weight on states and districts that are similar to most others nationally, so a poll of a swing district in Pennsylvania will have more impact on the calculation than a race in Hawaii or rural Alabama.

While you could just calculate a reasonable proxy forecast based on the national environment + the district’s partisan lean score, CANTOR takes a couple of further steps. One is to adjust this prediction based on something we call elasticity: how sensitive a state or district is to changes in the national environment. Canonically, a state like New Hampshire is very “swingy” — it has a lot of cross-pressured, moderate voters — while a state like Georgia is more bifurcated between strongly Democratic (e.g., Black voters, young professionals) and strongly Republican groups (evangelical whites). This elasticity calculation, which is derived from voter-level microdata14, is employed at various points by FLIPR, although for the most part it has a relatively minor impact.

CANTOR itself is a more involved process. Basically, it calculates similarity scores between any given combination of states or districts, creating what’s essentially a set of 48515 x 484 pairwise comparisons. These similarity scores, in turn, are based on a variety of political (e.g., partisan lean score), demographic (racial composition, religious composition, college-educatedness, etc.), and geographic characteristics (e.g., region16, geographic distance, urbanization) and how similar states or districts are across the entire set of these factors.

CANTOR then evaluates whether polls of individual races tend to be stronger or weaker than the baseline in similar states and districts. For example, say that CANTOR is calculating its estimate for Iowa’s 2nd congressional district in the northeastern part of the state, one of the most competitive races this year. Democrats are mostly outperforming presidential-year partisan baselines in every race this year, but does that overperformance tend to be concentrated in states and districts that are similar to IA-217? If so, CANTOR will also project further outperformance there.

Step 3. Add fundamentals to calculate “Classic”

“Fundamentals” basically get at the exact opposite question from a “polls only” model. Suppose that there were a national moratorium on state and district polls.18 How would you predict individual races? Well, you could get pretty far by looking at things like district partisanship, incumbency, and fundraising, especially if you were also able to use national polling series like the generic ballot (even if you didn’t have any local polling data).

FLIPR actually calculates two different versions of the fundamentals, however. What we call “simple fundamentals” is just the generic ballot plus each state or district’s partisan lean score. “Fancy fundamentals” uses a whole host of factors based on a large-scale regression analysis of past races since 1990, conversely. Many of these factors fall under the category of “candidate quality,” such as experience in elected office and even the presence or absence of scandals.

Prior to 2022, the model only used “fancy fundamentals”. But the more that polarization increases — and it’s higher than ever now — the more that partisanship crowds out everything else. “Simple fundamentals” can increasingly be equally if not more predictive than “fancy fundamentals”.19

The precise mix of “simple fundamentals” and “fancy fundamentals” varies based on overall national partisanship, as well as the type of election. For Congressional races, “simple fundamentals” now get almost half the weight in many cases. Roughly speaking, this means that candidate quality factors, including incumbency, are only about half as important now as they were over the broader period since 1990. However, “simple fundamentals” tell you comparatively less in gubernatorial races, where there is still much more ticket-splitting; e.g., Republican Gov. Phil Scott will probably be re-elected again in Vermont.

New this year, FLIPR also looks at how much “simple fundamentals” and “fancy fundamentals” diverge. In Vermont, for instance, partisan lean and the national environment would imply that a Democrat would almost certainly be elected governor, while “fancy fundamentals” account for Scott’s strong popularity as judged by his large re-election margin. In cases where “simple fundamentals” and “fancy fundamentals” are not aligned, the model trusts fundamentals less in general. In these cases, it both puts more emphasis on polls and (in Deluxe) expert ratings, and increases the overall uncertainty associated with the race.

The specific factors used in the “fancy fundamentals” calculations are as follows. Some factors can only be used in races with incumbents20 or only in Congressional races (not governor), and these are designated accordingly:

• Partisan lean scores for the state or district;

• The national political environment, as derived in Step 221;

• Incumbency; in fact, we technically fit separate regressions for incumbent and open-seat races. FLIPR’s handling of incumbency is slightly complicated:

  • It uses congressional approval polls to calculate whether the overall political environment is strong or weak for incumbents;

  • It assumes the incumbency advantage is larger in smaller, more idiosyncratic22 states and districts, where constituent services tend to be more important;

  • It uses the incumbent’s margin in her last competitive race, adjusting for the political environment at the time of the last election and changes in district boundaries;

  • It looks at whether the incumbent is in their first elected term, and whether they defeated an incumbent in their previous race;23

  • And for states that have redistricted, the incumbency advantage is less when there’s little overlap between a member’s old and new districts.

• Fundraising, based on individual contributions only (so not party or PAC money) with a bonus for contributions from within a candidate’s home state;24

• Elected experience on a 4-point scale from 0 (never elected to anything) to 3 (current or former elected senator or governor);

• Party loyalty scores based on congressional voting records. Candidates who vote against their party more often actually outperform electorally, other factors held equal. This calculation is only used for congressional incumbents; there is no proxy for it for governors or for challengers (even if they’ve served in Congress before);

• Scandals25, where new scandals are treated as being more important than “old” scandals when a candidate has already survived an election with the scandal having been known to the public.

Also, Classic does use a pinch of CANTOR’s proxy estimates for polling, though at a relatively modest weight. While CANTOR is a cool system, it’s generally less accurate than either fundamentals or polls directly from the state or district.

Step 4. Add expert ratings to get “Deluxe”

Every year, we have a debate about whether to retire the Deluxe version and stick to a model that uses purely objective inputs. Even though the expert ratings reflect other people’s judgments and not ours, there are some undesirable properties from including them. For instance, if experts are looking at Silver Bulletin forecasts to help calibrate their ratings, the entire process becomes somewhat recursive. Furthermore, in recent elections, the Deluxe version of our forecasts typically only “calls” one or two additional races correctly across the ~500 races we forecast every year compared to Classic.

But we’ve decided to continue publishing Deluxe and indeed treating it as the “default” forecast you see on the page — you can always click on the Classic or Lite versions if you prefer them.

That’s because, in the original design of the model, expert ratings were intended to serve a fairly narrow purpose: basically to account for idiosyncratic factors in each race that are not necessarily well-captured by the fundamentals, basically to shift around a little bit of vote margin from state to state without changing the macro picture much. Expert ratings are not a good way to estimate the national environment. Although the expert forecasts are good at discriminating between relatively stronger and weaker races for each party, they have a weird and persistent bias. Namely, expert raters tend to implicitly assume a neutral political environment (i.e., a generic ballot of about zero) until quite late in the cycle. As of early August, for instance, the unadjusted expert ratings imply a political environment of around D +2 versus the roughly D +7 environment calculated by FLIPR Classic. In a D +2 environment, a candidate like Talarico probably wouldn’t win despite Paxton’s weaknesses, but a D +7 would require much less overperformance.

FLIPR’s predecessors have always been aware of this tendency and “de-biased” the expert ratings before applying them. But in auditing the code this year, we discovered some shortcomings in the way this de-biasing process had been applied by the model. Basically, because of some artifacts in how races are rated, there was a bit of leakage from the experts’ implicit assumptions about the national environment (i.e., wrongly assuming the generic ballot = 0) into the Deluxe forecast. And we specifically do not think the experts’ views of the national environment provide much utility at all; their value is solely in the rank-ordering of different races relative to the national environment. In working on the model this year, we developed better techniques for removing this bias completely.

So the steps in going from Classic to Deluxe are as follows:

1. Translate the expert forecaster categories (e.g., “Lean Democratic”) into an implicit forecast of the margin in the race (e.g., D +6).26 The expert raters used in this calculation are Cook Political Report, Inside Elections and Sabato’s Crystal Ball.

2. Estimate the national environment implied by the translated expert ratings.27

3. De-bias the expert ratings by shifting them to FLIPR’s estimate of the national environment. For instance, if FLIPR estimates a D +7 national environment and the experts’ implicit national environment is D +2, we’d shift the expert ratings by the equivalent of 5 points toward Democrats.

4. Blend the de-biased expert margins with Classic. The weight assigned to expert ratings is generally smaller now than in past years. It’s typically about one-sixth of the forecast for Congressional races with polling and one-quarter of the forecast for gubernatorial races with polling28, though higher for races without polling. Before, expert ratings got closer to one-third of the weight in Deluxe across the board.

We believe this better serves the model’s original intent of using expert ratings to gently nudge the forecast in individual races — giving them more say in races where other indicators are less reliable, and they have a particularly strong opinion — without letting the experts change the overall national picture.

Step 5. Translate the current snapshots into forecasts, accounting for projected changes in the national environment

This is one of the simpler steps. FLIPR calculates an estimate of the House popular vote based on macro-level factors in elections since 1946: presidential approval ratings, the House popular vote in the previous election29, polarization, and whether the election is a midterm. As of early August, for example, this calculation forecasts Democrats to win the popular vote by around 8.5 points (mostly because of Trump’s very poor approval rating).

Meanwhile, Classic and Deluxe have their own estimates of the current national environment, which is currently closer to D +7: we call this the snapshot. The snapshot is then nudged toward the prior — so this year, for example, Democrats are expected to gain slightly more ground between August and Election Day. We don’t apply this adjustment in Lite, since Lite is supposed to be a “polls only” model. Instead, in Lite, the forecast is basically the snapshot, but with greater uncertainty.

I say “nudged,” however, because the effect of this adjustment is relatively gentle, and essentially fades out by Election Day. Indeed, as the generic ballot has become more stable, the prior has somewhat less utility than it used to in forecasting elections even months in advance.

Step 6. Account for uncertainty and calculate correlations between races

A candidate with a 3-point lead in a race where the margin of error is 5 points will win her race 88 percent of the time. This might be a good estimate in a well-polled Senate or gubernatorial race under stable conditions on the eve of the election. If the margin of error doubles to 10 points, however, because the election is a long way off, polling is sparse, or the data is contradictory, she’ll only win 72 percent of the time instead. That’s much better than a toss-up, but still a considerably less confident forecast.

Because all of our models are probabilistic, we tend to care a lot about this kind of stuff. It’s not just that elections are uncertain: the degree of uncertainty is predictable, at least to some degree. Abdul El-Sayed’s near loss in the Michigan Democratic primary despite a large polling lead wasn’t that surprising given that it’s not uncommon for primary polls to miss the mark by double digits, for example.

So in FLIPR, uncertainty is higher when:

• There’s less polling in the race;

• There are a lot of undecided voters in the polls

• There’s more time to go until the election;

• The various subcomponents of the model disagree with one another, e.g., there’s a big divergence between fundamentals and polls.

However, this is only half the battle — and it’s the easier half. As I’ve said, the errors in forecasts of congressional and gubernatorial races are somewhat correlated with one another, although less so than in presidential elections. So FLIPR not only has to estimate the uncertainty associated with each of the ~500 races that it forecasts, but also how the uncertainty in one race relates to another one. If Democrats beat their forecast in the Iowa Senate race, for example, they will probably also do so in the Iowa gubernatorial race and in Iowa’s 2nd congressional district — but we wouldn’t expect a race in Hawaii to be particularly correlated with Iowa.

FLIPR approaches this by breaking uncertainty down into 3-5 interlocking subcomponents, depending on how you count them:

• National error affects all races in the forecast roughly equally. If, in one simulation, the overall map is 3 points more Democratic or more Republican than in our forecast, we basically30 just shift every race by 3 points in that direction. This, in turn, can be broken down into two further subcomponents:

  • “Drift”, our term for the change in the national environment between the current forecast and Election Day. There tends to be less drift — fewer macro-level shifts in the polls — in political climates with high polarization like the one we’re in now. And drift zeroes out as Election Day approaches.

  • Election Day error: basically the possibility of a systematic polling/forecasting error on Election Day itself.

• Local error, conversely, applies only to one race or one state. In fact, we now account for both of these conditions:

  • In each simulation, we take one random draw that applies to all races (House, Senate, governor) in a particular state. (This is a new wrinkle for 2026.)

  • And then we also draw a random error that applies only to exactly one race.

• Finally, correlated demographic error is an intermediate layer that sits between these and is the most complex layer. Basically, we use all the variables used in CANTOR to simulate the possibility of underperformance or overperformance among particular subgroups. In one simulation, for example, Democrats might do better than expected among Hispanic voters, but worse than expected in the Northeast, and worse than expected among Catholics. This can produce the sort of geographic or thematic clustering of errors that we’ve seen in recent elections like 2018 and 2022 and FLIPR tries to create some maps like these. The intuition is simply that errors are more correlated across similar states and districts.

Step 7. Simulate the election 40,000 times, including projecting turnout

Since almost all of the parameters that FLIPR needs have been generated by this point, the simulations themselves aren’t particularly complicated from an algorithmic standpoint.31 There are a few small wrinkles, however:

• FLIPR uses slightly fat-tailed distributions to account for the possibility of rare, truly-out-of-sample events, although this effect has been dialed down from previous years;

• We use variance reduction techniques that make the simulations more stable, so the effective sample size is larger than 40,000 simulations;

• And the simulations do get more complicated in the case of multi-way races:

  • In races with prominent independent candidates, FLIPR simulates the independent’s vote share using a right-tailed statistical distribution in which the independent usually doesn’t come close but has some outside chance of gaining momentum and being a real factor in the race;32

  • In Maine and Alaska, FLIPR now carries out a fairly faithful simulation of ranked-choice voting;

  • In states like Georgia where elections go to runoffs if no candidate gets 50 percent of the vote, FLIPR simulates this too;

  • And FLIPR simulates the complex “jungle primary” process in Louisiana’s House races, which are using it for the last time this year.

FLIPR also probabilistically simulates election turnout and calculates an estimate of the national House popular vote. The turnout simulations are essentially a post-processing step; they don’t actually affect our calculations of vote shares, or the probability of a given party winning a particular race or the House or the Senate. Nevertheless, the design is fairly elaborate:

• Turnout in individual states and districts is benchmarked based on the competitiveness of the race, whether there’s a Senate or governor race in the district, past turnout in the state or district, and FLIPR’s estimate of the voting eligible population (VEP), adjusted for population growth.

• Overall national turnout is forecasted based on polls that ask voters whether they’re more excited or less excited than usual to vote in the upcoming election.

• The local and national estimates are then reconciled in various ways. FLIPR requires turnout in House, Senate and gubernatorial elections in the same state to remain within designated bounds. It applies a modest correction if it infers from its estimate of the national political environment that one party’s voters are likely to turn out in greater numbers as compared to past elections. And it treats its national turnout estimate as authoritative; if, say, summing up its projections in individual districts produces an estimate that overshoots the national projection by 2 percent, it will just give every district a 2 percent haircut.

Like basically everything else in FLIPR, these turnout estimates are treated as probabilistic rather than deterministic; we estimate both the mean and the variance.

Changes since 2022

Since the rest of this document is quite long, I’ll be brief here. But you’ll basically see a few major themes.

• Thinking more carefully about how increasing polarization affects all of the parameters in our model. We changed some of this in 2022 but left others alone.

• Thinking more carefully about the relationships between House, Senate and governor races within particular states.

• Accounting for some predictable differences in behavior between midterms and HSG behavior in presidential years. For instance, races are more correlated in presidential years.

Here is a fairly exhaustive list of changes:

The national environment and forecast uncertainty

• We’ve now more carefully calibrated the amount of “drift” in the forecast from the current day to Election Day. Previously, the model blended drift and Election Day uncertainty together. The new specification is more consistent with our presidential forecast. As part of this process, we discovered that the national picture now “locks in” earlier than it used to.

• FLIPR’s estimate of the national political environment now blends the generic ballot with a “bottom-up” estimate derived from state and district polling, rather than relying on the generic ballot alone. This was already being done for CANTOR but had not been applied to the rest of the model.

• In calculating its prior for the national popular vote, the model now makes a bigger distinction between presidential years and midterms. Midterms tend to produce more lopsided outcomes than are typically seen in congressional elections in presidential years.

• The model now assumes that races are slightly more correlated with one another in presidential years than they are in midterms.

• We’ve now introduced a statewide variance layer in addition to local and national variance.

• Fat tails are less severe than in past years, as we now have more cycles without huge outlier outcomes.

• We’ve cleaned up some inconsistencies in the projection of the national political environment between Lite, Classic and Deluxe.

• Variance-reduction techniques are employed to keep the simulations more consistent.

Fundamentals, CANTOR and expert ratings

• The overall weight given to fundamentals is now differentiated between congressional and gubernatorial races. Fundamentals tend to be more predictive in the former than the latter because congressional races more closely follow the partisan lean of a state or district and the data is generally higher-quality.33 As compared to the 2022 build of the model, fundamentals are considerably upweighted in our congressional forecasts but slightly downweighted for governors.

• The ratio of “simple fundamentals” to “fancy fundamentals” in Classic is now tied to our polarization coefficient rather than using a fixed ratio.

• “Fancy fundamentals” now use FLIPR’s overall estimate of the national political environment, including imputations from state and district polls, rather than just the generic ballot. “Simple fundamentals” are still based on the generic ballot alone.

• When “fancy fundamentals” and “simple fundamentals” disagree, the model now puts less weight on fundamentals and tends to trust the polls more. When they agree, it increases the weight on fundamentals.

• CANTOR and the “bottom-up” estimate of the national political environment have been improved in various ways, including by weighting races that are more representative of the national political environment more heavily and more explicitly accounting for selection bias in which races get polled.

• A wider array of demographics are now accounted for by CANTOR, particularly including religious groups (e.g. Jewish voters and Muslim voters) that were lumped in with other groups before.

• The weight given to expert forecasts versus the other components is generally now lower than it previously was in Deluxe, and now varies inversely with how reliable FLIPR thinks the fundamentals are in each race. Essentially, the expert ratings are used as more of a fallback in cases where other indicators may be unreliable.

• In Deluxe, the model is now much truer to its original intent of not allowing expert forecasts to affect the national environment — only the forecasts of individual races.

• Gubernatorial races have always been treated as being less partisan than congressional races by the model, but this distinction is now more complete. For example, gubernatorial forecasts rely less on the “fundamentals” because partisan lean is a less reliable signal for them.

Polling and turnout

• As turnout patterns change, the model has discarded its prior that the shift from registered voters to likely voters typically favors Republicans. It now uses a more neutral prior, but more quickly overrides it as more polls that directly compare likely-voter and registered-voter results become available.

• The model now calculates its own native version of presidential approval ratings that makes a likely-voter adjustment.

• Turnout estimates within a state are now reconciled between House, Senate and gubernatorial races to ensure they match one another within certain bounds.

• Turnout estimates are also now partly adjusted to better match FLIPR’s estimate of the national political environment. This adjustment is capped so that it cannot overwhelm other information in the turnout model, and the resulting estimates are still allowed to diverge because of Senate and gubernatorial races, ballot structure and other local factors.

Miscellaneous

• In 2022, we introduced a bonus in our fundraising data to account for contributions from within the candidate’s state, which tend to be more predictive of the eventual outcome. But this was implemented clumsily and excluded small-donor contributions where the donor’s state isn’t aggregated by the FEC. We now look at the ratio of in-state to out-of-state contributions for donors where this information is available, and then extrapolate it to all types of individual contributions, including small-donor contributions.

• In certain rare circumstances, we now allow the fundraising estimates to be overridden and calculated by proxy instead. Namely, these circumstances are when a candidate (say, Troy Jackson in Maine or Darline Graham in South Carolina) is a late replacement and hasn’t yet had time to file a fundraising report, or when a candidate (say, JB Pritzker in Illinois) eschews or discourages individual contributions to their campaign. Because FLIPR does not account for self-funding, these candidates would previously have been treated as raising zero money, but treating this as missing data is more appropriate.

• The model now explicitly simulates ranked-choice voting.

• It also now explicitly accounts for Alaska’s top-four primary system, in which more than one candidate from the same party can appear on the general-election ballot.

It’s just past 4 a.m. here on the East Coast, and as I write this, Abdul El-Sayed leads Haley Stevens by only 1.4 points in the Democratic primary for U.S. Senate with 95+ percent of the vote counted in Michigan. The margin is narrow, but when there aren’t a lot of votes left, will be hard for Stevens to make up. El-Sayed has been declared the winner by NBC News, and is considered the almost certain victor by prediction markets, but other news outlets haven’t called the race yet. I’ve never been much of an election night modeling guy — my whole thing is pre-election forecasts — so I’m not going to offer an opinion about whether NBC and the prediction markets got ahead of themselves. Most of you will probably read this in the morning, and the results may be clearer by then, but the points I’ll make in this newsletter aren’t particularly contingent on the exact final result.

Subscribe now

However, this is one of those cases where the headline doesn’t tell the whole story. A 1- or 2-point win for El-Sayed would be much narrower than the results predicted by polls; the last seven surveys of the race as tracked by the New York Times had El-Sayed up by margins ranging between 10 and 19 points. I will note that an earlier Glengariff Group poll had Stevens ahead by 7, which is actually likely to end up closer to the end result than most of the other surveys, though it was pretty far off, too.

I’ve been putting the finishing touches on our midterms model, and that’s been going well; it should launch (early) next week. But I wrote some notes to myself while following Michigan in the background over the course of the evening. Considering the hour, I’m mostly just going to reproduce those with some uncharacteristic brevity.

11 quick thoughts on the Michigan result

1. We should start by acknowledging that an El-Sayed win by any margin is a significant accomplishment. It’s not exactly surprising, given the success of more progressive candidates elsewhere and that he’s been ahead in the polls for a couple of months now. But this is a guy whose only previous attempt at elected office resulted in a 22-point loss to Gretchen Whitmer in the 2018 Michigan gubernatorial primary. Michigan, where I’m from, isn’t New York City or some super liberal district in Colorado. This is a big W for progressives in a purple state.

2. With that said, I think AES and some of his supporters are guilty of managing expectations very poorly. Some of the polls I mentioned were from firms affiliated with El-Sayed’s campaign. I don’t understand the utility of publishing internal polls showing yourself ahead by double digits — personally, if I ran a campaign, I’d make it a strict policy to never release internal polls, which are often highly misleading. And if you’re going to do that, you probably ought to come closer than within ~10 points of the actual result. El-Sayed has also predicted a big win for himself against Mike Rogers, who won the Republican nomination uncontested tonight, but who is a relatively moderate Republican for the Trump Era. As someone who can be a little trollish himself, and is sometimes accused of being cocky, I’ve kind of vibed with AES’s confidence on some level, but I’m also not running for office and would be a terrible candidate if I did.1

3. In presidential primaries, results relative to “expectations” matter a great deal. Indeed, you can literally even model this, as we did in 2020. If AES had been going against Stevens in the New Hampshire primary, and had been favored by 13 points in polls but won by 1 or 2 instead, his chances of eventually becoming the Democratic nominee would probably have declined tonight, because the 1 extra delegate he won or whatever would be outweighed by the perceived loss of momentum and the message that donors, the media, etc. took from that. Primaries for Congress don’t work that way: there’s no runoff in this race, no second chance for Stevens next week. Still, this race has big implications for how Democrats will navigate the political landscape going forward, including specifically in 2028. I’ve generally been bullish about the left’s chances in 2028. But the way this went down won’t provide the morale boost that a bigger win might have.

4. Instead, a very narrow AES win is in the sweet spot of the “second-guessing zone” for the Democratic establishment. Instead of accepting that they have a lot of problems, they’re going to ask what they might have done differently. Mallory McMorrow actually got 4 percent of the vote despite dropping out (so El-Sayed isn’t likely to hit 50 percent). What if she had endorsed Stevens? Or what if she had been the nominee? With some very notable exceptions such as Michigan Gov. Gretchen Whitmer, a lot of the Democratic establishment, including Michigan incumbent Sen. Elissa Slotkin, stayed on the sidelines in the race.2 I don’t necessarily know that their endorsements would have helped Stevens, especially since a lot of El-Sayed’s rhetoric was about how the Democratic establishment, AIPAC and wealthy donors had been putting a heavy finger on the scale.3 But the establishment might regard this as a game in which they took the starters out at halftime, and almost won anyway, rather than one they were destined to lose the whole time.

5. It’s clear that progressives are rising in the Democratic Party overall. But it’s also clear that there’s a fair amount of race-by-race variance. Just tonight, in Missouri’s 1st congressional district, former Squad member Cori Bush lost to a moderate Democrat by 22 points in a comeback bid, while one of the most moderate Democrats in the House, Marie Gluesenkamp Perez, easily advanced to the general election in Washington’s 3rd district. Roy Cooper is the Democratic nominee in North Carolina, and that hasn’t been particularly contentious. The media likes simple narratives, and “progressives rising” is by no means wrong, but nobody should be bending the knee at this point.

6. With that said, there’s a fair amount of cope in establishment circles too. Stevens loses? Well, a lot of establishment types will just assert that she was a bad candidate. Andrew Cuomo? Bad candidate. Janet Mills? Bad candidate. These things are rarely so obvious in advance, though. Stevens won election to the House four times in a suburban Detroit district with perfectly normal margins, for instance.

7. Prediction markets have an increasing tendency toward overconfidence. In my chat with Eli on Sunday, I said explicitly that giving Stevens only a ~2 or 3 percent chance was too low, given that polls in Congressional primaries are often wildly off. Historically, in gambling-adjacent activities, there’s a bias toward longshots. But I think that may have flipped in the other direction given all the attention given to prediction markets and that people tend to treat them as oracular when they shouldn’t, perhaps producing some self-reinforcing feedback loops.

8. Just to put a finer point on this, the average error in polls in presidential primaries since 2000 is around 9 points of vote margin (!), more than twice President Trump’s (in)famous polling outperformance in 2016. And polling error tends to increase the further you go down the ballot. So if it’s 9 points in a presidential primary, it might be who-knows-what in a congressional primary, but we’re probably looking at something in the double digits.4 Many people, including traders at prediction markets, tend to be much too crude in the heuristics they apply toward polling error, failing to recognize that polling is relatively accurate in some types of races but routinely quite inaccurate in others and that much of this is predictable.

9. Further, the whole trajectory of polling in this race was weird. At one point, the McMorrow campaign successfully pressured a small news outlet into quashing a poll that showed them way behind. At another point, El-Sayed falsely accused a different pollster that had Stevens narrowly ahead of having been funded by AIPAC. Many of the pollsters in this race have mediocre pollster ratings, aren’t rated at all because they’re new, or aren’t used by Silver Bulletin because they don’t meet our standards for professional polling firms. There can sometimes be cases of the blind leading the blind in polling, producing a lot of herding toward a sometimes very wrong consensus. The constant, frankly often very amateurish “discourse” about polling on Twitter and elsewhere on social media doesn’t help matters either. There are fewer and fewer pollsters I trust to do their own thing, instead of being influenced by “the narrative”.

10. With that said, Congressional primaries are hard to poll, and Michigan is a particularly hard state to poll for various reasons. Among them: a) Stevens’s best groups were probably rural white voters and working-class Black voters, two groups that are notoriously hard to reach in surveys; b) AES was counting on big turnout from younger voters, but they often don’t turn out in big numbers in primaries; c) I think it’s pretty clear that AES had the more enthusiastic supporters, but this is a mixed indicator because of non-response bias: more enthusiastic voters might be more likely to vote, but may also be proportionately even more likely to participate in surveys when what you might call “Shy Stevens” voters don’t; d) Michigan has a reasonably large Arab-American population, but not large enough to yield a reliable crosstab in (say) a 600-person survey, so instead Arab-Americans routinely fall between the cracks in opinion research.

11. Tonight is probably bearish for Democrats in the general election, but I wouldn’t overstate that. I’ve always thought people are too eager to derive implications about general election strength from primary results. Part of the reason that Democratic voters are moving to the left is that moderate voters are leaving the party, and AES will need to win them in the general election. If you’re a Silver Bulletin reader who happens to be a Democrat, one positive would be if El-Sayed takes tonight’s results as a signal that he needs to be more conciliatory between now and November.

Thanks for reading, and look for the model next week.

Next week, we’ll launch our midterms model, with complete forecasts of the 435 U.S. House races, 35 Senate races, and 36 gubernatorial races on the ballot this November, as well as the overall likelihood of each party controlling each chamber and all kinds of other details.1

The model performed quite well in 2018 after the code was basically rewritten from scratch, and again in 2022 after some further tweaks mostly to reflect increasing political polarization. So we’re not looking to reinvent the wheel. However, AI tools are helpful for refining existing code, so we’ve cleaned up some minor inconsistencies, made the code faster, and tested a few hypotheses that should help around the margin.

We feel pretty safe about the timing because we got the model basically fully “working” over the weekend. Still, we need a day or two to make some further coding refinements, a day to write up the documentation, a day or two to test out some further empirical improvements on our punch list, a day or so to finalize the graphics, and some time to double-check everything and kick the tires on anything that looks weird. We’ll also surely have some “take” for you about Michigan after Abdul El-Sayed and Haley Stevens have their primary on Tuesday night. That’s a finite amount of work, but it probably lands us at early next week, not this week, for model launch. We greatly appreciate your patience and are looking forward to having a model-heavy autumn with both the midterms and ELWAY.

Yesterday, I chatted with Eli about the model, the midterms and Michigan — and Trump’s approval rating, which once again recently hit a record low in our tracking. You can find the video below, though it’s choppy in places2 so there’s also a lightly edited transcript below.

Nate Silver: Hey Eli, how are you doing?

Eli McKown-Dawson: Doing pretty well. How about you?

Nate: Good. I’m going to have some news for you and for Silver Bulletin in a minute that explains why I’m sadly kind of bleary-eyed.

The model is working, which is good. That doesn’t mean it’s done, but it means that it takes these files that we — mostly you — spent months gathering. I think there are 68 different input files that we use in the program.3 It goes through 14,000 lines of code, and it spits out coherent projections for the state and the national picture.

There are a lot of things we still have to vet. By the way, you almost gave me a heart attack last night. One of the variables that we use is a candidate experience variable, which runs on a scale of 0 for never elected to anything to 3 for a senator or governor. In the [data] we were using, a candidate with no experience got coded as a 9. Nine is sometimes used for missing data. The model interpreted a 9 as you’re fucking Gandhi crossed with Reagan crossed with Obama crossed with the Pope. You’re the most popular incumbent ever. So it had all these [inexperienced challengers] crushing [incumbents]4, and I’m like, “wow, we’re really out of consensus. That’s interesting.” And then we fixed that, and it became more normal.

I will say, we are going to spend a little bit more time debugging, vetting, building better graphics, and looking at new features. My guess, in true honesty, readers, is that we will launch the model a week from Monday, so the 10th. It’s not official — we’ll talk about it in our planning meeting. If life depended on it, we could probably launch this week. We have a big primary this week in Michigan, which we’re going to talk about a lot. But basically, we are on the final lap, I would say, Eli. On the checkered flag lap.

Eli: I do just want to say, I saw that message yesterday about the nine versus the zero coding. I’m fairly certain your original message about that did say to code those people as nines. So maybe it’s not just me making the nine mistake. But that’s okay.

Nate: It’s a one-line change in the code. I recoded all nines to zeros and all blanks to zeros. But it scared the hell out of me.

We [published] the World Cup model a couple of months ago, and in that model, you have the bare bones of a simplified version that actually kind of works. And then you refine it, and there are more and more steps that you add. Maybe you add secondary data sources or better ways to algorithmically process the data. So you get to see the results refining in real time. Of course, you don’t want to be too results-oriented.

[But] with the midterm model, 80 percent of the work is in gathering these 68 fucking files that we have, and making sure that the 550-odd races5 that we’re forecasting — governor, House, Senate — making sure all that information is accurate, merging candidates’ FEC IDs with their pollster IDs and that kind of stuff. So by the time you actually see what the model says for the first time, you’re 90 percent of the way through the work.

What’s the national environment looking like?

Nate: Do you have any questions about the model? I don’t want to commit to what it says because we haven’t debugged it yet and we haven’t vetted any of the outputs.

Eli: About what it’s saying right now, you mean?

Nate: Yeah. I’ll be vague as needed.

Eli: This is a tough one, but it’s one of the races I’m more curious about. What is Maine looking like? Maine Senate. What does the model think about that?

Nate: It has (Troy) Jackson favored. [But] I’ll put it like this: the notion that Maine and Michigan could be more challenging states for Democrats than you’d expect … [that] doesn’t ruin Democrats’ chances of taking the Senate as much as you’d think. And now I’m really spoiling content that we should be putting behind a paywall.

I think people forget the longer tail of races like Iowa, for example. If you look at Iowa, it is polling as a toss-up, basically. With the fundamentals, my guess is the model will have it as “lean R” — I don’t remember offhand — but Iowa is a big deal. Even a state like Nebraska, potentially, where Dan Osborn is running as kind of a quasi-populist independent and presumably would caucus with Democrats if it came down to that. There’s this longer tail of races that might get a little bit neglected, and it’s a longer and larger playing field than I think people might assume. We’re not going to be too far out of the consensus. There are other models, some of them are good, and they’re kind of mostly pointing in the same direction.

The other thing that I thought was interesting: a lot of [this] will change after Labor Day — but a lot of the generic ballot data, and sometimes the horse race polls, are denominated in registered voter currency. The way our model adjusts for likely voters, it used to be that Republicans reliably had a turnout advantage in midterm elections. Now that’s changing, as Democrats have the higher-turnout, college-educated coalition. We at this point are just starting out with a prior that you have zero conversion from RV to LV. However, we look at polls that survey both, and in those polls, Democrats gain a net of about two-and-a-half points among likely voters versus registered voters. So when you see the generic ballot number, which is D +6.2 or something…

Eli: It’s D +6.5 right now.

Nate: 6.5, right. Some of those are likely voter polls, some of them are registered voter polls. If all of them were likely voter polls, then that probably pushes it up to a +7 or a +8. As we’ve discussed before, if you’re in a +7, +8, +9 environment, that’s when some of these red states that are PVI +11 come into play. They become almost toss-ups. Democrats have also had very good fundraising numbers, for example, even though the DNC is having a great deal of trouble, and even though Elon Musk has pledged to donate X number of millions of dollars to the campaign. But in individual contributions, Democrats use their enthusiasm [edge]. They use tools like ActBlue to really penetrate pretty far down the map. And some of these populist candidates — maybe the model won’t love them, they don’t tend to have a lot of elected experience, and that is a variable that empirically matters — did raise a lot of money, too. So those things are all explaining why the playing field is a little bit broader for Democrats than you might think.

We’re [also] going to look at, for example, whether the generic ballot is more stable now. It used to be that no one was paying attention to Congress until after Labor Day. In recent years, you’ve been locked into the environment a little bit quicker.

One [other] thing we always look at or reevaluate is how correlated different races are. If you think about a presidential map where you have Kamala Harris and Donald Trump on the ballot in every state, it would be impossible to imagine Kamala Harris losing Maine but winning Ohio. In the Senate, you have 35 separate races that all have different candidates. States will have very non-uniform turnout, depending on how much money is spent there and how many races are competitive. Some will have almost a presidential-year turnout. Some will be very light.

And in 2018 and 2022, the midterm results were more regionalized than people expected. I think in both years the first substantive, important state to report its results was Florida. Eli, you’re from Florida. Florida has turned quite red, and so in both those years, the first results for Democrats looked disastrous. It turned out that actually Democrats had a very good midterm in 2018, but they lost the Senate seat in Florida. And they had a fine midterm year relative to expectations in 2022, when an unpopular Biden was president.

You also had some regionalization here in the New York-New Jersey area in 2022. We had a little micro Republican red wave that didn’t show up anywhere else in the country. We kind of made our bread and butter on models that understand properly the correlation between different races. That, in some ways, was the big Silver Bulletin slash FiveThirtyEight innovation. But people overadjust. It’s not as correlated in midterm years as it is for the presidency itself. And furthermore, the correlation between presidential races and midterms is a little bit lower when you don’t have an actual president on the ballot, and instead you just have an expectation based on previous voting patterns.

We’re also going to make the model faster. These AI tools are both incredible and at times incredibly frustrating, but usually they can find improvements that just make the code run faster and more efficiently. Eli, what do you want to talk about? Do you want to go down to Michigan?

Eli: I think Michigan. We kind of hit on that a little bit already, talking about the Senate and the model and what it’s thinking. One thing to add before we do, though, just on the registered voter/likely voter split: I looked at this last week. Right now, in the recent generic ballot polls, still something like 60 percent of them are among registered voters. So over the next couple of months we probably will start to see more of that shift, because right now most people really are still polling registered voters. We’ve not hit the point where that shift is happening yet.

Nate: Which is reasonable, right? People don’t necessarily know if they’re going to vote right now, and to some extent we want surveys of public opinion based on the broader population. But yes, that’s about a two, two-and-a-half point gap. It can sometimes be even more in individual states [than] on the generic ballot.

The generic ballot is a very weird, abstract question that behaves differently from other polling series. You wrote about this6, and we’ll repurpose that piece as a Model Talk at some point. But other people’s generic ballots are too bouncy. For whatever reason, empirically, there’s not as much herding on the generic ballot as on other polls. Meaning, in this Michigan race, every single pollster is El-Sayed +12 in the primary.7 That’s a little herding-ish. On the generic ballot you don’t have that as much. There are very strong house effects, as we call them. Different pollsters have different views. They don’t seem to be anchored to anything quite as much. What that means is that you have to be more patient and wait for more data to have a reliable average. And people who haven’t looked at that empirically wind up with polling averages that bounce around too much.

There are mathematical ways to test this. You don’t want it to be autocorrelated. This is getting more technical than I thought for a Sunday morning quote-unquote brunch chat. But if [your average] goes from D +5 to D +6, the predicted reading for tomorrow should not be D +5 or D +7. If you’re efficiently accounting for [all] information, then you set things such that it moves in the way that best predicts next week’s value on the generic ballot.8 And you need to use fairly sluggish settings on the generic ballot as compared with presidential approval, where you have a higher density of polls and maybe more consistency in how pollsters ask those questions and how people interpret them.

Eli: We saw this a little bit over the past month or so. In our average, the generic ballot peaked at around D +7, and then it reverted about a point to D +6. Other averages that weren’t doing quite as much smoothing as us had maybe a three-ish point drop, which was way too much. Because even with the usual variation — just last week there were a lot of incredible polls for Democrats coming out, D +7, D +10 — you still do have people showing it at D +4. Some pollsters are just very, very consistent, showing the generic ballot not moving at all. And when you get these periods where that’s all that’s coming out, you don’t want your average to over-index on those polls and revert too much to the D +4 area, which is definitely too low for what the environment actually is.

Nate: And look, we also have some precedent based on history. If you are looking at what presidential approval is … and you account for the fact that we do have greater polarization now … you’d probably expect Democrats to be +7 or +8. That’s [also] a familiar number if you’ve seen recent, non-2022 midterms.9 Why 2022 was only a relative ripple for Republicans is an interesting question. It may reflect the Democratic turnout enthusiasm advantage. It may reflect the shadow of Trump being unpopular. It reflects candidate recruitment in some cases.

But people don’t change their minds about this abstract question of which party you would vote for for Congress as often as they do about Trump. Let’s say you’re a cross-pressured voter, so there are some things you like about Trump. Maybe you like him on the economy, and maybe you think it’s cool how he got our player unsuspended from a red card in the World Cup. But then the war with Iran starts up again, and you hear [about] that, and you see some YouTube video, and you’re like, “oh yeah, that kind of sucks.” You change your mind more [often]. Or if you’re [a voter] in the Stevens-El-Sayed race. That race has now stabilized, but you’re picking up lots of advertising, lots of information about the candidates. But party [affiliation] is much more glacial. It gets locked in pretty early, and I think we’re seeing the effect tangibly.

Trump’s approval, if anything, has tanked even further [though].

How low can Trump’s approval rating go?

Eli: Today, his net approval is -20.6, which is an all-time low, both first and second term. At least in our retrospective averages, that’s the new low.

Nate: And this has been kind of the pattern. I think we were early, Eli, about telling people that the floor on second terms is a lot lower. Bush became profoundly unpopular in his second term, for example. There’s no need to rehabilitate the candidate Weekend at Bernie’s style for re-election. Then people are aspiring for how they’re going to position themselves for 2028, potentially. People are like, “well, at some point you’re going to run out of repercussions here if I defy you.” And [other Republican candidates] are concerned about their own races in D +8, D +10 type environments.

Eli updates the approval ratings every day, sometimes twice a day, three times a day. Has anything surprised you about this? Or what patterns have you learned from studying this obsessively for a couple of years?

Eli: I do think it’s been interesting to look at these numbers in relation to the Iran War, because there really has been some movement based on what’s going on with that, and especially how it’s related to things like gas prices. We had this period a little while ago where there was a stretch of ceasefire-ish type stuff as negotiations were going on. Gas prices went down, and that was a period for a few weeks where Trump’s approval rating was actually ticking up — both in our average, and even if you looked at the same pollsters who poll every week or every month, you saw his numbers improving a little bit across the board. Now things have heated back up, gas prices are back up, and you’re really seeing now just a string of awful polls for him.

That and tariffs. People really do not like tariffs. Liberation Day was the first big drop we had in the average, very early on, and tariffs are also now popping back up. It’s been interesting to see how those two things — people don’t like them, and when they are salient, and when they start affecting gas prices or other prices, it does seem to really impact the approval ratings.

Nate: Gas prices are canonically an extremely visible artifact. Usually, you go into a foreign country in the Middle East, and you start a war, and things deteriorate gradually. Unless people have families in the military, or are defense contractors, [they] might not be following the conflict from day to day that much. Here, there’s a fucking buzzer, where anytime it looks like this ceasefire — or the Schrödinger ceasefire — is breaking down, then it’s reflected immediately in how much oil is going for, and within a few days you have producers trying to anticipate the effect on future gas prices and changing their prices in that way. So it’s a very, very quick feedback loop. Sometimes the stock market is a little bit of a feedback mechanism too. There is concern about the “is AI a bubble” story, along with Iran as this perpetual concern, and then [interest] rates.

I kind of annoyed some resistance-y Bluesky types a couple of weeks ago by saying that [some of the recent stories] about Trump are kind of boring and stupid. I think I was triggered by the second iteration of the White House Correspondents’ Dinner … he was insulting to some journalists in ways that were inadvisable. People [were] like, “oh my God, this is the worst thing to ever happen.” It’s like, it’s the worst thing to happen since two days ago or something. But the public actually does not follow politics in the way that people like us do — who are blogging about it every day, or doing Substack Lives every day.

Eli: No.

Nate: The public cares about big, substantive stuff. They care about economic anxiety. They care about Iran. Even the Epstein stuff — maybe it hurt Trump a little bit, but there’s not a clear line that you can draw. So tariffs and Iran are the kind of public triggers. Maybe ICE a little bit. He seemed to suffer a little bit when we had two people killed by ICE back in the winter. It’s actually fairly predictable.

Eli: That’s interesting too, because Trump’s immigration approval rating is also right around his second-term low again, at about net -13, which is where it was earlier with the two ICE shootings. So that’s also been on a bit of a downturn, even though it hasn’t been in the news as much.

Nate: But not as much as the tariffs number and the affordability-slash-inflation number. Those ones bounce around more. I think if you asked the White House, they’d say, “Well, sure, people aren’t happy about the economy, it’s not fantastic, but we can kind of win them back with cultural grievance type issues.” And it’s kind of almost the reverse, if anything. It’s like, “you know what? You’re full of shit about the ceasefire, it keeps breaking down. You’re full of shit about tariffs lowering prices. So maybe you’re full of shit about ICE too.” That’s kind of more how it works. It’s a negative halo effect, as opposed to being a distraction.

Eli, how low do you think Trump could get? What’s your 20th percentile estimate of approval as of, let’s say, February 2027 — half a year from now?

Eli: Right now, we’re pretty close to breaking 60 percent on the share of people who disapprove of Trump. I think I’d have to look a little bit at some of the other historical numbers. I think it’s possible that it gets a fair bit lower. It depends on what happens with Iran and that sort of thing. It feels like it could hit -30 at some point. That would be a very low [number]. But I think if things like gas prices do not improve, if the war in Iran becomes a more widespread conflict, I think that could start to cause bigger problems. And maybe even post-midterms, when Trump is even more of a lame duck, he could start to lose even more sway with the party, because then things really are going to kind of switch over more to 2028. What is your low estimate here? Do you have one?

Nate: By the way, to satisfy the many readers who understood that Biden’s age was a big problem: Trump turned 80, was it a month or so ago10? He’s also losing his fastball, if he ever had a fastball, a little bit more. But … if you look at actuarial curves, there’s an inflection point around 80 — he has the hardest job in the world, [so] give him that much credit — [but] that could lead to more poor decision-making.

What’s the chance of Trump on some day in the next six months achieving a -30 net [approval rating]? I think the chances might be 25 percent or something. It’s not too crazy.

[And] if you look at his strong approval numbers, only 20.8 percent of people have a strongly approving view of him. People fuck that up. They think, well, he might only have — what’s his overall approval number?

Eli: It’s 36 or 35 percent.

Nate: People say “he might only have 35 percent, but they’re the most dedicated 35 percent.” And that’s not true. If anything, the disapprovals are stronger: 47.6 percent of people have a strongly disapproving view of Trump. So if you look at that spread, it’s already a -27, and so far that has been a leading indicator for how other people feel. There is some floor. [For] the net approval [to go] below -30 — or the gross approval below 30 percent — that seems harder. But who knows?

Eli: You do have, in those 20-ish percent of people who strongly approve, some of them are going to stick with him no matter what happens. I’m not sure how much lower that strong approval is going to go. But maybe we’ll break 50 percent strongly disapprove at some point. That seems more possible.

Nate: That would be noteworthy. There can be a permission structure too. These [soft Trump disapprovers] are not people who are going to vote for Pete Buttigieg or AOC in 2028. A lot of them are people who would probably even vote for Trump again if they absolutely had to. But — and to get [back to] the midterm model — you’re seeing this differential in fundraising, you’re seeing this differential in voter enthusiasm as measured by likely voter versus registered voter polls. And people might just be like, “you know what, I’m just not going to turn out for this midterm.”

Turnout was very high in 2018, and pretty high in 2022. One irony: there are polls where you ask how enthusiastic voters are — our model actually calculates a turnout number; we’re gathering that data. Democrats might not mind one-sided turnout in 2026. Trump has never been a particularly helpful president as far as helping [other Republican] candidates. And in a midterm, (a) he might endorse candidates who are not particularly winners, and (b) he’s not the most enthusiastic campaigner, and so forth. Are they having that stupid fucking convention?

Eli: Yes. It’s in Texas. It’s going to be in Dallas.

Nate: I thought it was somewhere fun. Nothing wrong with Dallas. Maybe we go or something. Dallas in September, see the Cowboys, maybe. [Thinks.] There’s zero chance we’re going to go to that, or even spend more than two seconds watching a YouTube clip. I don’t like conventions. I don’t like the State of the Union. I like debates, man. Debates are fucking awesome.

Eli: I do like debates.

Nate: But I don’t like the constructed stuff in politics so much. It’s stagecraft.

Eli: Often it’s just a show, and there are formalities that go on, but it’s nothing particularly newsworthy — unless you’re one of the people who goes through everything Trump says line by line and clips it for X.

Republicans need at least 2 big things to go right to hold the House

Nate: So, prediction markets — I’m not going to reveal our [model’s take] yet — give Democrats around an 80 percent chance of winning the House and a 50 percent chance of winning the Senate. Maybe it’s up to 82 or 83 for the House, I forget.

Eli, describe to me the world where that 20 percent [chance of a Republican House] happens.

Eli: Oh man, okay. I think it’s either a world where there’s a huge reversion in the generic ballot national environment, or the polls are just wrong in some big way and we’re heading for another year where we’re really underestimating Republicans. I think it would have to be a world where Trump gets a little less unpopular over the next few months. Gas prices go down. Maybe the war in Iran ends and it goes very well, and gas prices go way down. And he’s like, “oh, you know what, tariffs are canceled, we don’t like tariffs anymore.”

I think there would have to be a lot of things that go either right for Republicans or wrong for Democrats for that to happen. Right now, if the polls are at all accurate, the national environment is just too blue for Democrats to lose the House realistically, because there just aren’t that many competitive districts. You’d really have to see a big change in what the national environment looks like — maybe people get really unhappy with Democrats for some reason. Because really, if we’re in not a D +6 or 7 environment but a D +3 environment, Democrats can lose the House, because [the map] sort of favors Republicans. They narrowly won the redistricting wars, at least for this cycle. And so if we’re off in that direction, it’s possible. But based on other indicators like gas prices, like Trump’s approval rating, you’d have to have a lot of things that are all pointing in one direction be off. It’s a 20 percent chance, but I don’t know.

Nate: I don’t know. Sometimes there are, from a meta-forecasting question, the 20 percent chances that you [can] see [happening], and then there are some that you don’t. You know what I mean? We don’t really use our gut per se in the model, that’s why you have a model. Sometimes I’ll have a gut instinct: “well, if I were to bet on this, I’d bet one side or the other.” I don’t think I’d bet the 20 percent side, really.

The [other] thing we are looking at, that I talked about before, is: has the political environment become more stable from the summer through November, as a result of greater polarization locking things in earlier?

Everyone has a way to convert the generic ballot plus other inputs into individual race forecasts, and they seem to show that somewhere around a D +3 is the break-even point, where if Democrats win the popular vote for the House by [only] three, they could lose the House. By the way, there are lots of complications. Democrats have nominated candidates, I think, in all but three House races. This is usually a good sign. They’re not going to win in these deep red states in Alabama, but it’s usually a good sign of a healthy party. Republicans have more blank districts.

But anyway, I think you need maybe two or three ingredients. One of them is a big systematic polling error of the sort that we haven’t had in a midterm recently, but we have in some of the recent presidential years.

Eli: Yeah.

Nate: Two is bullish, consumer-facing economic signs in the … third quarter. That certainly requires some de-escalation in Iran. It requires that you don’t have an AI bubble popping, or other kinds of trauma to the economy in different ways. But if gas prices were declining, we’ve seen tangible impacts already on [Trump’s] polling going from bad to catastrophic [when gas prices increase].

And the third theme, that I’m sure we’ll be discussing a lot for the rest of the year, is: are Democrats leaning too far left? There’s the micro and the macro. In individual races, are Democrats nominating candidates who — if you believe in the median voter theorem — are suboptimally far away from the median voter? And can Trump turn that into a national storyline, that the Democrats are crazy and woke is back? Can they nationalize that story beyond individual races? Because Democrats are nominating [moderate] Roy Cooper in North Carolina, for example. Talarico has a pretty progressive record, frankly, but has been kind of quietly running to the center in that race in Texas. But the DSA-type candidates tend to be more unapologetically proud of being on the left, and don’t necessarily seem to be team players in quite the same way.

So can you have two out of the three: the economy magically recovers, a polling error, or Dems in disarray slash the median voter theorem is still powerful, and [Republicans] win the meta-argument over “are [Democrats] too far to the left?” Two of those three would make the race competitive. And three of them probably means a Republican hold.

Eli: And I think it’s also difficult because — it’s a small sample size, obviously — but usually the out party will gain a few points on the generic ballot between now and the midterms, historically, if you look back at our averages. Obviously, there have been some exceptions, like after 9/11, that was an exception where the out party didn’t gain. But otherwise, it’s another sign in Democrats’ favor.

Nate: Trump’s approval is at -21, rounding off. The generic ballot is at -6 for Republicans, or -7. There is a universe of people who are open to “orange man bad” arguments, and a lot of them will vote for their local Republican member of Congress. But you know what’s going to happen, Eli? A lot of them aren’t going to vote. They’re like, “I’m a Republican who is sick and tired of tariffs and Iran, and if you ask me at this point, I’d say I disapprove of Trump. Maybe I’m excited about 2028 or something. I’m just not going to vote. I’m going to go to the gym or something instead of taking the time to vote, or watch TV.”

The Michigan Senate primary is far from a toss-up

Nate: Eli, we have not talked about Michigan. We should do Michigan before we wrap up.

Eli: Yeah.

Nate: Who’s going to win the primary on Tuesday?

Eli: When it went from a three-person race to a two-person race, you had two polls come out that were conducted right after Mallory McMorrow dropped out, where Stevens was leading by a couple of points. Since that point, though, every poll has shown El-Sayed +10 to +15. It’s been really consistent. Now, a lot of those have been polls from relatively favorable sources — they were conducted by some pro El-Sayed PACs — but you also have nonpartisan polls showing basically the same result.

You have potential for large polling errors in primary polls, and that is a thing that could happen. But that’s also a fairly robust lead. And I think even if you were looking at the older polls before it became a two-person race, that’s not too surprising. I wrote about this a few weeks ago: Stevens would have had to win 80-plus percent of the former McMorrow voters to actually pull even with El-Sayed, who was already leading in that three-person race. So I think El-Sayed is in a very good position to win this primary, unless again there’s some massive polling error such that all the pollsters — even the usually good-track-record Michigan-specific pollsters — are very wrong about what this electorate is going to look like. Plus the DSA left — I know El-Sayed doesn’t call himself a democratic socialist, but he’s very much backed by those same people — they have been doing incredibly well in Democratic primaries this whole cycle. So that’s another point in his favor.

Nate: And they’re organized. They turn people out, including in places where turnout tends to be low.

The sequence, as you mentioned, is a little weird, in that a month ago El-Sayed’s leading in all the polls by 10 or 12 points on average. And then a bunch of stuff happens. There’s this Graham Platner fiasco. McMorrow drops out. Stevens starts to get more media attention. I’m from Michigan, and I might say Haley Stevens possesses one of the more exaggerated versions of Michigan mannerisms. It’s not how the median Michigander talks. It’s not even how the 90th percentile talks.

But anyway, you have these polls. It was El-Sayed +10 or whatever. Then stuff happens. The next two polls actually show Stevens winning by some margin. You can look at our article in real time. We do not say, “oh, now she’s the favorite.” We say, “oh, it’s probably tightening.” Then there are some El-Sayed internal polls that show, no, they’re like, fuck you, we’re still winning by 13 points. And then nonpartisan polls show the same thing. Everyone kind of has the same number: El-Sayed +10, 12, 13, 14, 15, or whatever else.

By the way, when there’s consensus, I don’t regard that as a good sign in polling. I think Michigan polling can be a little bit of the blind leading the blind. It’s one of those upper Midwestern states where there have been problems in recent years. And this is a race where you have big splits between Black voters and white voters. What happens to Arab, Palestinian, Middle Eastern voters in these polls depends on the pollster. They can get clustered into some cross-tab, when you probably want to survey them individually. The voting patterns can be very different.

But yeah, look, I’ll put it like this: if you’ve been kind enough to listen to 45 minutes of this Eli-Nate live stream, here’s some gambling-adjacent advice. Stevens was, when I checked last night, at 3 percent on Polymarket and Kalshi. You’re going to lose this bet almost always. But 3 percent probably has to be too low.11 If you look at — I think people aren’t adjusting for how error-prone polling is in primaries. If this was a general election, 3 percent would be the correct number, or maybe even a little too high.

Eli: Yes.

Nate: All the disastrous polling that we had in 2016 — the biggest errors were off by a net of four, four and a half points. Well, El-Sayed leads by 11 points in the polling average, so it’d be two and a half times that. [But] there are just much wider margins of error in congressional primaries. El-Sayed is counting on getting younger people out to vote who may not vote in primaries ordinarily — though Michigan has a fairly high-turnout state. And you have had a relatively large amount of news.

The thing I totally omitted: at some point after those polls came out — the first couple showing Stevens actually leading, and then the El-Sayed internals — Gretchen Whitmer, who is still quite popular, endorsed Stevens. If we had done this [chat] a week ago, we’d say, “okay, well, look, all of the polling evidence probably favors El-Sayed, but it’s uncertain, and you have this endorsement.” I would have been surprised that that endorsement didn’t apparently move the numbers more. Are you surprised by that? Our timing is good — we [would have looked] like idiots for saying, “oh now, it’s a toss-up,” which it probably [wasn’t]. But I thought that [endorsement would have] moved things a little bit more.

Eli: I kind of did too. [Stevens has] done well on the Michigan endorsement front, because of course the current holder of that seat [Gary Peters] also endorsed her — that happened a few weeks earlier — but that plus Whitmer, I thought that might have moved the needle a little bit. But so far, at least based on the polling, it hasn’t much. I don’t necessarily know what Gretchen Whitmer’s strategy is here, because it’s going to look very bad for her if she’s done this last-minute endorsement and then it doesn’t work. I’m not sure what the calculus there is. But I think we do look better doing this now [instead of a week ago], because we’re not saying, “oh, it’s a toss-up because of the endorsement.”

And yes, I think three to four percent is too low. Again, if this was a general election, that would make more sense. But people do forget how error-prone primary polling actually is. But again, that still doesn’t change the fact that El-Sayed is absolutely the favorite.

Oh, okay. Well, it looks like Nate has left the stream. I know he’s been having some internet issues, so I guess we’re going to wait and see if Nate hops back on.

Nate: Now I’m on my phone. For some reason, the technology stack here at Silver Bulletin headquarters is not as robust as one might hope.

Will El-Sayed win the general?

Eli: I was just talking about Michigan more. We got a question about what this is going to look like in November if El-Sayed does win.

Nate: Let me put it like this: it’s a race that would have to be monitored carefully [for November], as opposed to one that you would assume would be a Democratic pickup with a more conventional candidate in a D +8 environment.

Eli: Yeah.

Nate: Now, with that said, there have been a couple of polls for El-Sayed lately against Mike Rogers that have been quite bad.

Eli: Yeah, he’s been behind in those, while Stevens, who’s also been tested, has been leading. And you know, that’s not a great place to be in Michigan when you have other candidates in redder states who are winning their polls. Democrats like Talarico have been doing well, and Peltola in Alaska, obviously.

Nate: Yeah, if Talarico were to win and El-Sayed were to lose… But I do want to say one thing. At this point in time, when you are at the end of a long, contentious primary, your general election polls tend to suffer. Because maybe Stevens voters, or vice versa, are like, “El-Sayed? Fuck that. They’ve been beating up my candidate, and I’ll vote for Mike Rogers, just watch me.” [But] whoever wins on Tuesday, Democrats will have three months to persuade voters to come back into the fold and not do that. And that’s a pretty important mechanism that usually works for parties.

But still, it would not be ridiculous to see Democrats lose that race with El-Sayed. And the way I look at it, it’s data. We have had this abstract debate for years about how much, if at all, does moderation help. The answer is it actually helps quite a bit, if you’re trying to actually look at empirical historical data. But now we won’t need to have hypothetical data, or data inferred from House races people don’t pay attention to, because you’re going to have cases like the Senate race in Michigan, cases like the gubernatorial race most likely in Wisconsin, where you have explicitly progressive candidates.

Although we will say, El-Sayed has advocated for Medicare for All, but he’s not a member of the DSA. He calls himself a capitalist. He says, I’m a capitalist, and I understand the flaws of capitalism as a capitalist. So he has interesting messaging that’s not quite as boilerplate. It’s sometimes just more interesting than Graham Platner. Graham Platner was a kind of Reddit DSA candidate. And El-Sayed is an interesting candidate, a former podcaster running for office.

Eli, there’s hope for all of us. If the podcaster can become a U.S. senator, then it can happen for anyone.

Eli: I think the other point goes back to what we were talking about a little while ago with the general environment for Democrats. These candidates could end up underperforming by a large amount and still win. Right now, on our generic ballot tracker, we have state-level projections based on the current generic ballot. For Michigan, right now, it’s D +6.5. So El-Sayed could win the primary, underperform by five points, and still potentially win the Senate race. And you could end up seeing that in other places too, like Wisconsin. If you have a very, very good environment for Democrats in November, bad candidates can still win these elections. I’m not sure that’s going to do wonders for the factional fights in the Democratic Party.

Nate: I do want to distinguish governor races from Senate races, because if you are voting for a national body, then people are more partisan, and rationally so. You might say, well — and this was kind of the argument the Platner bros were making toward the end of the race — this is really awful, but the alternative is Donald Trump controls the Senate still.

That logic isn’t as persuasive for a gubernatorial race [which has fewer national implications]. It’s like, “well, fuck you, you nominated that candidate. Nominate a non-DSA candidate next time.” And so I would differentiate that, and our model will too. It will weigh partisanship less in governor races than in Senate races.

But one thing that helps the narrative for progressives — I know it sounds dumb, what about the substance, who cares about the narrative? Well, the narrative matters for how Democrats behave in the future. If El-Sayed wins by three in a D +8 environment, all the smart-alecky people will be like, “well, according to WAR, he was fine, but he was below replacement.” And that won’t persuade people. People just talk about W’s and L’s. They aren’t looking at margin of victory very much. It’s ironically the same type of thing that you wind up looking at in sports betting a lot. It’s very hard, if you lose a football game or a soccer game or a basketball game, for people to say, “oh, it was relatively good relative to expectations.” It’s a very nerdy Eli-and-Nate way of thinking that won’t carry the day in a Democratic primary.

But to be fair, what if El-Sayed wins by 10 points? You’d have to give credit to that, and you have to update your views based on the evidence. There are some reasons to think that everything apart from partisanship is becoming less important, so there might be less of a penalty for nominating a quote-unquote far-left or far-right candidate.

Things are also overdetermined. We do not, I believe, currently have any Arab Americans in the U.S. Senate. Michigan has the largest Arab American population — proportionally, I think.

Eli: Yeah, per capita.

Nate: But that kind of thing can matter. If you look at Dearborn, Michigan, the most Arab American city in the country — what did Harris get, 42 percent of the vote12? And Jill Stein got a bunch of votes.

Eli: Jill Stein got a lot of the vote in those areas in 2024.

Nate: And midterms are kind of low turnout. Maybe El-Sayed sparks more turnout among young voters. Again, that’s not the way I’d bet — I’d bet on him underperforming — but we have to always be open to evidence. Politics is changing in certain ways, and progressives are benefiting from that.

Do you think Stevens and El-Sayed are quote-unquote bad candidates or do you think they’re quote-unquote good candidates?

Eli: If you look at Stevens historically, she’s a meh candidate. She’s fine. Her Split Ticket WAR is around R +1 to D +1 across her various cycles. So she’s just, at least in her House races, been very much a neutral-ish type of candidate, which I guess is what I’d expect if she won and was in the general this year. Which again, in a year that’s going to be good for Democrats, neutral-ish I think is great. I think you’d win with neutral-ish.

El-Sayed, it’s hard to tell. If I had to bet, I think he would underperform a little bit, at least based on these initial polls. Maybe he won’t. Maybe the sort of energy that’s going through all these Democratic primaries right now, where these kinds of candidates are doing very well, maybe that really will carry over into the general, and all these young people are really going to be energized and turn out, and he’s going to overperform. But if I had to put money on it, I would imagine he would underperform at least a little bit, because Mike Rogers will be a decent candidate.

Nate: [Rogers] is as close to a normie, old-era Republican as you can get. Absolutely.

Eli: This isn’t a Texas situation, where the Republican candidate is a mess and you get the benefit of a bad opponent.

Nate: That’s right. I could go on a long thing about why I think Talarico isn’t as fantastic a candidate as the amount of donations he’s getting would seem to suggest. But he’s running against a kind of objectively terrible and scandal-plagued candidate, and that helps a lot. And by the way, his polling has been pretty good.

Eli: It has been pretty good. I see so many Talarico donation ads pop up on YouTube, always asking me for money.

Nate: I never want to be too prescriptive with our readers, and I don’t like giving any type of partisan recommendation on where to spend money. But I will say this, because I’ve said it before: you probably don’t want to donate to the high-profile race that you’re hearing the most about. Those candidates have more fucking money than they’re ever going to need for anything useful. So whether you’re a Democrat, Republican, independent, whatever else, think about your ROI in terms of, does this candidate actually need more marginal resources, or are you getting really heavily diminishing marginal returns?

Eli: And Talarico has a lot of money. As someone who has been spending a lot of time recently digging through FEC filings for the model — he has a lot of money.

Nate: Anything surprising in the fundraising data, Eli? Or is it pretty consistently better for Democrats than Republicans?

Eli: Yeah, Democrats do really well with this. For context, for the listeners, our model is using individual contributions here, as opposed to money from PACs or self-funding. And I think the Democrats are doing really, really well on that front. Honestly, as long as I’ve been paying attention to things, it’s been very consistent, and that’s definitely continuing. Whereas on the Republican side, lots of self-funders like Ramaswamy.

Nate: And we don’t count that, because we aren’t using fundraising as a sign of resources so much as a measure of voter enthusiasm and organizational strength.

In fact, we actually weight contributions from within your state five times more heavily than contributions from out of state, because that indicates local support. We actually refined this process [this year] — it was originally implemented in 2022, but let’s be honest, there were a couple of things that were not as smooth as they could have been about that implementation.13

And it is true, you have this Jaime Harrison race in South Carolina a couple of years ago for Democrats, where he was probably not going to beat Lindsey Graham, and wound up losing by 10. And there was a lot of money that went into this probably pipe dream of a race. Of course, you don’t know what’s a pipe dream ahead of time. It could be that Iowa or Alaska is kind of a white whale race for Democrats this year, potentially.

But Democrats, they like donating. They like smashing that button and donating money to their candidates, and that’s shown up more and more in the data. And Trump, by the way, because he’s kind of self-funded slash funded in mildly corrupt — sometimes more than mildly — quid-pro-quo ways, he’s probably not terrific for the more traditional small- to medium-sized donor Republican fundraising channel. Whereas Democrats like donating to campaigns.

Eli: Yes, absolutely.

Nate: Eli, what else have we got? What else is on your mind? What else are you looking forward to debating, or not debating, about the midterms?

Eli: It’s been so much candidate quality stuff lately with all the Democratic primaries. Genuinely, for my own curiosity, I kind of do hope El-Sayed wins, because I want to see what that kind of general election campaign looks like. Because again, the Democratic candidates of this DSA-adjacent wing have been doing very, very well, but a lot of that wing’s early wins have been in very blue districts or blue areas, like New York City with Mamdani, for example.

I’m very interested to see what a general election campaign from that style of candidate looks like in a competitive state, because so often you end up having these candidates win Democratic primaries in places where the Democratic primary basically is the race, and the general is more or less a formality because the Democrat is almost certainly going to win. So I think that’s why you often see that wing very focused on beating Democrats in primaries. And so I’m interested to see how a general election in a purple state in a neutral year would actually look.

Nate: And [El-Sayed] is not scandal-plagued in the way that Graham Platner is.

Eli: No, so it’s a cleaner test. He was a Rhodes scholar. He’s a doctor. So if he does poorly, it’s not going to be because of scandals.

Nate: Personality is a different category, and he was literally a former lax bro. He’s got a more aggressive personality that people might find either fun and charming or off-putting. I’ll be interested to see the gender splits in this race a little bit, whether he’s getting younger men as much. But it’s a pretty fucking clean test of how much ideology matters in a race like this.

And I will say, for all the WAR-type data, when nerds gather data on underperformance or overperformance in Congress, all of that is from fairly low-stakes races, where they’re not competitive. It’s like, “oh, you’re supposed to win your blue district by 40 points instead of 32. There’s support for moderation.” But this is a clean test. I’ll put it like this: if he were to lose the race — especially if Democrats have a good night overall, or if they don’t, I think either way — if [El-Sayed] loses, then you’ll hear about that race until the end of time from the moderates. If he wins by anything more than a nail-biter, then probably the opposite.

Eli: You’d hear a lot about him from progressives. And maybe he’ll run for president.

Nate: Unlike Zohran, he’s from the United States. That’s not meant to insult Zohran, but you have to be born in the United States to be eligible to be president. And he’s young — I guess he’s 35, right?

Eli: Yeah, he must be at least 35.14

Nate: It is interesting. It’s kind of a selection effect thing. A lot of those progressive candidates are quite charismatic, at least the ones that win their primaries. You don’t have to like Abdul, but it’s a brand of charisma. A certain kind of … charisma, in the way that Haley Stevens has both charisma and anti-charisma in a very strange kind of way.

But there is something to the fact that the archetypal left-wing candidate everyone knows is Bernie Sanders. I think you could call Bernie frumpy. I don’t think Bernie would even mind being called frumpy. Most of these candidates — I mean, Zohran is wearing the suit. El-Sayed is clean-cut. AOC. All these people are pretty good-looking, well-presented people. That’s not a total coincidence. Politics is part symbolism, and it’s part saying, “okay, I’m going to play the part of the more normie role in society.” But that’s going to be an interesting contrast, I’d say.

And it is important too — I know I’m repeating myself — that you have a relatively normal Republican. There are other states where they would nominate Mike the fucking pillow guy Lindell, or whatever … but this is a state where you have one of the more traditional, OG Republicans.

Eli: And that does matter, because otherwise it’s not a good test at all, because you can’t tell what’s going on, whether it’s a bad Democratic candidate or a bad Republican candidate causing problems. And I do think the results of these sorts of races are going to matter for Democrats going forward, in terms of when they’re thinking about 2028, who to nominate, that sort of thing. If progressives do very well in the midterms, that wing of the party is going to use it as evidence that their brand is going to be a winning way going forward — maybe nominating AOC.

Nate: Yeah, look, the thing we haven’t mentioned — we haven’t mentioned Israel and Gaza and AIPAC very much. The second biggest theme after Medicare for All has been AIPAC, AIPAC, AIPAC. Maybe the biggest theme, frankly, coming from El-Sayed.

Eli: I saw some interesting data, I think from The Argument, a couple of weeks ago. They were looking at how often candidates are mentioning Israel or AIPAC in their emails, and I think El-Sayed was maybe the second or third most frequent mentioner of those topics in his fundraising emails.

Nate: And of course, AIPAC is complicated, because it can be kind of this motte-and-bailey thing, where on the one hand it can be a question about Israel, obviously, and a proxy for Israel policy. On the other hand, it can be this question about corporate, national donors. And look, I don’t think it would be the worst thing if there were more of a stigma for candidates who raise money from big PACs and corporations. Maybe it’s time to have that conversation.

Obviously, Israel is very divisive. Maybe it won’t be so divisive [for Democrats in the future] though. Maybe people will just think, well, if you’re a Democrat, you have to be kind of explicitly pro-Palestine now, and that might be the message that people take away. Keep in mind that Democrats in general, especially establishment Democrats, are not the most risk-taking people. They tend to be followers. They don’t skate where the puck is going, they skate where the puck was five seconds ago, behind the play. But that does make them predictable in a lot of ways.

I think people will extrapolate very heavily from this race, and properly so. You could make excuses for the moderates in the New York City mayor’s race — “okay, well, New York is pretty progressive.” Michigan is Michigan. Michigan is exactly what it looks like. We were looking last week at historical primary results. Michigan is now an early voting state, maybe the most important early voting state, because it comes right before Virginia and Super Tuesday. But Michigan voted for Bernie in what was it, 2020?15 It voted for John McCain in 2000, when that was one of the few states where he beat Bush. It voted for George Wallace in 1972, back in the day. It voted for Jesse Jackson. So Michigan has a little bit of a rebellious streak that maybe Ohio doesn’t.

Here’s my proposal, Eli: why don’t we take one of the Ohio senators and give it to Michigan?

Eli: Then they’d get three?

Nate: Michigan is quite a bit of a better state than Ohio, I would argue.16

Eli: I don’t have a dog in either fight, but I guess if I had to pick, I’d pick Michigan. I’m not a super partisan on either the Michigan or Ohio side. But if I had to pick, I’d go with Michigan.

Nate: No, I mean, Ohio doesn’t bother me. If you had to get rid of states, if you had to go to 48, obviously you would fold Delaware into Pennsylvania, and merge the Dakotas.

Eli: The Dakotas is the first, yeah.

Nate: And then what’s next on the chopping block? Rhode Island?

Eli: It’s got to be, right? You’ve got to do the small ones. I think after the Dakotas, you have to go by size.

Nate: Yeah, I think Rhode Island probably gets it next. Rhode Island [merged] into Connecticut.

So, readers, listeners: our plan for the week is that we’re going to probably have a “think piece” — well, it depends if Michigan is actually close — it might be more of a diagnosis of what’s going on. Probably Wednesday morning or late Tuesday night, we’ll have some type of what-it-all-means piece, and then we’re mostly focused on the model. Hopefully we’ll get you a sports article or something at some point. But full disclosure, we’re probably looking at a week from Monday, give or take a day, for the model. If it winds up being this week, it wouldn’t be the thing that surprised me the most in the world. But usually it’s pretty rare that toward the deadline things resolve faster than you’re expecting.

Eli: And we’ve also usually found that unless you’re trying to launch before some sort of deadline — for example, before the NFL season starts17 — it’s good to take some extra time with the landing pages to make them look nice and work well before we launch them.

Nate: The honest truth of when will the model be ready? It’s probably Friday or Saturday. You don’t want to launch things on a Friday or Saturday, especially in the middle of August. That’s some standard media advice that we’ve found holds up in the Substack world.

[Though] here we are, Eli, on a Sunday, having a very, very, very wonky conversation. Maybe more wonky than I expected. But maybe we’ll end on that note. So everyone, lots coming soon. There’s lots of backlogged content that I’m looking forward to writing when these back-to-back projects with the World Cup and the midterms are ready to go. So a little slow for a week here, but then I think we’re going to have a great August.

Thank you so much for listening and for subscribing.

Eli: Yeah, have a good week, everybody.

The presidential nomination process is a strange “game.” Early states like New Hampshire and South Carolina don’t actually have very many delegates, and Democrats use a highly proportional system, so it’s very unlikely that they’ll prove to be the tipping-point states in determining who wins the majority at the 2028 convention. They matter mostly to the extent that the media and “party elites” (e.g., donors, elected officials, partisan influencers) believe they matter and treat them as reliable signals about the viability of the candidates.

And sometimes those expectations can be gamed. In 1992, Iowa Sen. Tom Harkin received 76 percent of the vote in the Iowa caucuses, with “uncommitted” in second. But this produced little in the way of momentum for Harkin because Bill Clinton and the other candidates had taken their ball and gone home. The media went along with this small conspiracy to deny Harkin his momentum1, camping out in New Hampshire instead. Clinton dubbed himself “The Comeback Kid” despite finishing second there to Paul Tsongas — and the media went along with that framing too.

So I don’t want to treat the order of states too mechanically. Although “momentum” can be measured statistically though indicators like national polls, candidates will seek to manipulate expectations and exploit various biases in media coverage. Moreover, the early states aren’t typically that predictive of the nominee. Many of them have a lifetime success rate below 50 percent in picking the eventual winner.

But having built a model of the primaries in 2020 — honestly, maybe the most complicated forecast we’ve ever designed — I have quite a few thoughts about the new calendar. Don’t worry: we’ll have a lot more coverage of the midterms soon. (Yes, we’re finally making excellent progress on the model.) But this is the first newsletter that I’m going to label as a Model Talk column related to the 2028 election.

That’s because last week, Democrats proposed their 2028 primary calendar:2

• South Carolina: Saturday, January 22, 2028

• Nevada: Tuesday, February 1

• New Hampshire: Tuesday, February 8

• New Mexico: Tuesday, February 15

• Michigan: Tuesday, February 22

• Virginia: Tuesday, February 29

• Super Tuesday: March 7

Did Democrats forget about Iowa? Nope, after a colossal screw-up in reporting results in 2020, Democrats relegated it outside of the early-voting window. Meanwhile, New Hampshire has been demoted to holding the third primary.

New Hampshirites will complain that the state has a law requiring it to have the first primary.3 I don’t see why the Democratic Party or anyone in the other 49 states should really care about that4 — though along with Iowa, it might try to hold an earlier, unsanctioned contest. If these states seek to defy the calendar, the question is whether candidates and the media play along.

In 2008, Michigan and Florida jumped the queue in the primaries, and were punished with having their delegates stripped.5 Hillary Clinton and Barack Obama pledged not to campaign there — though Clinton sort of cheated on this promise in Florida, while Obama wasn’t even on the ballot in Michigan. Clinton’s wins in these states were treated, at least in my recollection, with an extreme asterisk by the media, providing only some fraction of the momentum that they usually would.

Putting minority-heavy states first could help moderate candidates

Nobody has a right to complain too much about the new calendar, though some progressives are kvetching about it. It’s reasonably well balanced between different parts of the country and the different racial groups that make up the Democratic Party. And it probably diminishes the importance of the first four states, regardless of their order, as compared to #5 and #6: Michigan and Virginia, which will offer opportunities to build up momentum before Super Tuesday, when far more delegates will be at stake.

Still, overall, I would say the new calendar puts a slight thumb on the scale in favor of more moderate candidates.

In most cases, you might associate concerns about racial equity with the more left-wing factions of the Democratic Party. There is, however, an ironic twist when it comes to the perpetual debates about the primary calendar: it’s usually the moderates who express more need for racial balance.

The reason is a phenomenon that might seem counterintuitive, but is well-known to any veteran of the process. White voters overall are more conservative than members of racial minority groups. But among Democrats, the ideological ordering is reversed. White Democrats are generally really liberal, while Hispanic and Black Democrats are more moderate. You’re seeing that play out in Michigan right now, where Haley Stevens is hoping for a strong performance among Black voters to come back against Abdul El-Sayed’s coalition of white liberals and Arab-American voters in the state’s primary next week.

In data I gathered from the Cooperative Election Study from 2016 through 2025, 65 percent of white Democrats identify as “liberal” or “very liberal”. But only 44 percent of Hispanic Democrats and 41 percent of Black Democrats do so.

All of these figures probably lowball the liberal share we’ll see in 2028, both because the Democratic Party is becoming more liberal and because more progressive voters are usually more enthusiastic than moderates and turn out at greater rates in the primaries. But I don’t see any particular reason to expect a reordering of the relative ratio of progressives to moderates among the different racial groups within the party.

Let’s take the rest of this in three stages. Since this is a Model Talk column, hopefully you’re not averse to a little bit of obsessive detail. Because in the primaries, small details tend to matter.

• First, I’ll show you some data on the ideological makeup of the new First 6 states. (And the other 44 states also.)

• Next, I’ll show you the historical results from the early-voting states, plus now demoted Iowa.

• And then, I’ll describe some findings from our experience in actually modeling the primaries about how much the order of the states matters in the first place, and even speculate a bit about how this might affect some of the leading candidates.

Read more

“Agency” is one of those terms — see also “priors,” etc. — that I associate with “Silicon Valley rationalist types” and more broadly with the community that I call the River in my book. It’s perhaps slightly more intimidating than it needs to be. But it also captures something that there isn’t exactly any good synonym for. Here’s how I define “agency” in On the Edge1:

Agency [is] … being empowered to make robust, well-informed decisions, knowing which factors are inside one’s control. [...] The term refers not merely to having options but good options where the costs and benefits are transparent, don’t require overcoming an undue amount of friction, and don’t risk entrapping you in an addictive spiral.

Cate Hall, the author of the excellent Substack Useful Fictions, has what is undoubtedly a better definition:

Agency [is] … the entirely learnable capacity to both see and act on all of the degrees of freedom that life really offers.

I met Cate while playing a low-stakes tournament during the 2021 World Series of Poker, and it turns out we have a lot in common. Like me, Hall quit a corporate-track job to play poker — indeed, with far more success than I had, at one point becoming the #1 ranked female player in the world. She’s since had a number of other adventures, including being the CEO of the Astera Institute and a co-founder of Alvea, as well as being the author of an outstanding new book, You Can Just Do Things: How High-Agency People Get What They Want Out of Life, which I spoke with her about on Thursday.

Both the poker life and the writer’s life offer a tremendous amount of ostensible freedom. You don’t have a boss, and you don’t necessarily have a lot of structure unless you impose it on yourself.2 There very much are consequences to a lack of productive effort — in some ways, in fact, they’re more direct consequences than when you’re working for a corporate paycheck3. But you mostly bear those consequences yourself. These are careers for highly agentic, self-motivated people, in other words.

I introduced “agency” in On the Edge in the context of an extended discussion about the proliferation of gambling. As a liberal with libertarian tendencies, you might say I have a prior against imposing limits on people’s choices. For instance, although I think ubiquitous sports betting is probably net-harmful for society, that doesn’t mean I think it should be banned. (Slot machines? That’s another question.) At the same time, I don’t think it’s as simple as saying “nobody has to gamble unless they want to, so all you’re doing is giving people more options.” Both brick-and-mortar and especially online casinos use all sorts of manipulative techniques to get you to gamble more, and it can require a lot of self-discipline to avoid these traps.

Social media algorithms, and algorithms in general, literally use some of the same techniques, establishing powerful defaults that can be hard to escape from. These are likely to become even more coercive in the age of AI, mining data on each user’s revealed preferences — essentially, pushing exactly the right buttons so they keep pushing buttons and increasing revenue and “time on site.” I think there’s far too much of a tendency in Silicon Valley to entrust decisions to AI — or to the billionaires who run the labs — on behalf of the rest of us.

What Hall’s definition captures that mine misses, however, is that agency is very much a learnable skill: few of us come out of the womb as fully realized, high-agency people. She has discovered this the hard way, including a multi-year drug addiction that she said “utterly wrecked [her] life.”

You Can Just Do Things is a brave book, both because of Hall’s honesty and because it breaks through certain taboos. Why do so many of us (raises hand) allow ourselves to become entrapped in dissatisfying professional relationships, for example? I’m not necessarily a believer in “radical transparency4,” but there are lots of times when more honest communication would allow us to get more of what we want out of life, while possibly making the other party better off as well, and we avoid it out of risk-aversion or fear of failure. This can also extend to friendships and even romantic relationships.

A lightly edited transcript of our conversation follows below.5

Subscribe now

Nate Silver: I’m here with Cate Hall, who I met playing poker a few years ago. I think Cate and I have a fair bit in common — we probably would have run into each other one way or another. Cate has a great new book out called You Can Just Do Things: How High-Agency People Get What They Want Out of Life, co-authored with Sasha Chapin.

Cate, I want to ask you how you came to write this book, but first I want to define the term “agency.” It’s a term I like. It’s a term people in the broader — I don’t know if I want to use the word “rationalist” — community use a lot. I know you define it in your newsletter, Useful Fictions, which is a great Substack, by the way. What does it mean, exactly?

Cate Hall: Good question. It’s kind of like a Rorschach test, because everybody has a slightly different definition. The one I use most often is the capacity to both see and act on degrees of freedom that other people overlook. So it’s a combination of observational capacity and then courage in acting on those degrees of freedom.

Nate: I like the courage part. This is a very courageous book. Would it be insulting to call it a self-help book?

Cate: No.

Nate: It’s kind of a self-help book for high-agency people, for a certain type of person who maybe has a lot of degrees of freedom in their life. I really liked it. But it’s a brave book, and a vulnerable one. How did you come to write it?

Cate: I’ve thought quite a bit, from a personal-journey perspective, about human agency over the last five years. Five and a half years ago I was really deep in drug addiction, and my life had totally fallen apart. I was broke. I felt like I was unemployable. Relationships had withered away. A lot of the content in the book, though it’s definitely not a book about addiction, comes from the experience of recovering and rebuilding my life. I put in a lot of cycles thinking about how to become a more effective person starting from a place of being really stuck.

The book actually came about as the result of the very first Substack article I ever posted, which I wrote at my husband Sasha Chapin’s suggestion. I dashed off an article about how to be more agentic, with a few tips, and the reaction was really strong. I got a ton of Substack followers. I had two book editors reach out and ask if I was interested in writing a book about it. And I was like, “Oh, okay. I guess this is a topic people really want to hear about. I guess I’ll write a book about it.” Working with Sasha on it was incredibly fun, and also incredibly necessary from the perspective of actually getting the book done.

Nate: I’ve never co-authored a book. Well, I guess I have in the old Baseball Prospectus days, for OG fans. My partner works on the Substack, and that creates interesting dynamics. One thing you talk about in the book that maybe I’m not the best practitioner of is hyper-honest communication. Does that go into the relationship, too?

Cate: Yeah, absolutely. Writing the book together was way more fun than I thought it was going to be. I kind of expected bad things to happen, because a lot of people report that when they try to do creative projects with their significant other. But I think we’re each well suited for very different parts of the writing and editing process, so it was actually pretty seamless, and we managed to not drive each other too insane along the way. We have a very honest relationship. We’re good at surfacing disagreements before they become lingering resentments.

Quitting law for poker

Nate: A book is a super high-agency thing to do. You start out literally with a blank page — or in your case, a blog post. The deadline is too far off. You talk in the book and in the newsletter about the importance of rapid feedback. In poker you get feedback after every hand. It’s very noisy, but you get it. With a book, you [didn’t] get that feedback until this week, when it’s in the hands of readers and reviewers.

I skipped over some of your biography […] you went to Yale Law School, right? And you were in Washington — am I making that up?

Cate: No, that’s right. I practiced in Washington for a while.

Nate: On a track to be one of these people who go [argue] before the Supreme Court one day. And then you quit to play poker — which has parallels. I quit my first job to play poker.

Cate: I was about 30 years old and a lawyer in D.C. at the time. I was in the process of discovering that I really didn’t want to be a lawyer around the same time that gambling became legal in the D.C. area. I picked up poker as a fun hobby, and then fell in love with it, and it slowly started eating more and more of my time. For some months before I officially told my law firm I was leaving, I was sneaking away to play poker a little too much. I remember my mother was very concerned that I had a gambling addiction, and I was like, “No, I’m just unhappy as a lawyer and I have a hobby that I like.”

But at some point it seemed like the law firm thing was going super well, in the sense that my career was advancing. I had a bit of a come-to-Jesus moment where I was like, “Oh, I need to get out of here, or I’m going to be stuck forever.” So I told my firm I was leaving shortly after bonuses were announced one year, which they always love. And everybody was like, “Okay, well, you’re obviously having a midlife crisis. Why don’t you take a year off and then come back?” I considered doing that, but I was completely in love with poker from the moment I started playing it, so I never seriously considered it.

Nate: And you had a lot of success. You were very high in the GPI — Global Poker Index — rankings at some point. I think you were the number one ranked female player in the world. Was poker one of those things that was as exciting as you expected it to be, or did it become a burden once it was your career?

Cate: I only played professionally for about three years, which is probably a good amount of time to play professionally without getting totally jaded. It definitely becomes less fun when you’re playing for a living than when you’re playing recreationally. Unless you’ve been a stockbroker, it’s hard to convey how demoralizing it is to show up at work and lose money day after day. You go on these streaks where it’s just brutal, especially if you’re a tournament poker player. And you start to feel like there’s no escape from the poker community, and the dynamics at the poker table.

Nate: I don’t think people realize, in the typical tournament, about 10 to 15 percent of players cash. If you’re good, you might cash more. But the real skill is being able to run very deep, and not just get these “min-cashes,” as we call them. There aren’t a ton of other fields where you fail 88 percent of the time. Even if you enter 200 tournaments a year and you’re a good player, you’re still going to have a losing year almost half the time.6 That’s just the math of the distribution, with the prize money concentrated at the top. What positive and negative habits does that lifestyle teach people?

Cate: It is such a head trip. Part of it is the fact that you can lose money a lot of the time. But there’s an additional psychological element, which is that if you play tournament poker — which is primarily what I did — something disappointing is always happening unless you win the tournament. You always have this very sad, numbing, terrible feeling at the end of the day or the tournament. I don’t feel like human psychology is well suited for that kind of environment. If you go super deep in a tournament and make back three or four times your buy-in, you’ve been sitting there playing for often multiple days, there’s a huge pile of money at the top, and you just feel like shit most of the time. You’re not thinking, “Oh, I made four times my money.”

I don’t know if there’s a useful thing to learn from that, besides maybe a bunch of different dissociation strategies. I do think there are habits of thinking you pick up in poker that are useful for other parts of life. I got a lot in terms of learning how to think probabilistically, to think in terms of expected value, learning how to be comfortable with failure, more experimentation and more doing stuff publicly. But I wouldn’t recommend adopting the psychological disposition of a poker player, because it tends to be sort of warped and mangled by the specificities of the profession.

Is addiction the opposite of agency?

Nate: There is a lot of compulsive behavior in the community — sometimes gambling, sometimes substance abuse, sometimes personality quirks. But the feeling of running deep in a tournament is really fun. It’s one of the best things I’ve ever experienced, and it makes up for all the constant pain and struggle a little bit. So how did you fall out of poker? Is this the point at which you developed addiction issues?

Cate: Yes. I had long been a psychonaut experimenter with drugs, and at some point I had this really intense spiritual experience — basically the first spiritual experience of my life — on drugs, and it became my complete obsession overnight. Replicating that, chasing that feeling. As obsessive as I had been about poker, I became about seeking that thing. At the time I stopped playing, I was still the top-ranked female poker player in the world, and I basically just left and almost never played again.

Nate: I think I played against you at the World Series in 2021 — it was you and me in some $250 H.O.R.S.E. tournament. Is it as simple as saying that addiction is the opposite of agency, or is it not quite that simple?

Cate: I think that’s a really interesting way to look at it, and I use that frame sometimes. If you think about agency as the capacity to unlock additional degrees of freedom, addiction is really the opposite of that. One of the things I talk about in a TED Talk I did is that when I would go out every morning to buy drugs and come home, I’d see people walking around doing totally normal things, and I felt incredible resentment toward those people because of the degrees of freedom they had — to just be out walking on the street, going to lunch with a friend. It was so far from what I felt I had the capacity to do at the time. Addiction does seem like stripping away everything else in service of mashing this one button over and over again, and that feels like the opposite of agency to me.

Nate: How did you get out of that spiral?

Cate: The real answer is that rehab was really important, and I was so resistant to rehab. For the first year and a half of my addiction, I wasn’t trying to recover. For the last two years of my addiction, I was trying to recover. I was doing all of the 12-step programs, and a bunch of stuff other than going to rehab, but I kept getting a month sober and then relapsing, and having a terrible relapse for months at a time. Having the really sustained timeout — the 60-day timeout from life that rehab required — was necessary for me to recover. Until that point, I had this idea that if I went to rehab, things would fall apart in some way that they hadn’t already, which is very classic addict thinking. But eventually I did it, and I really recommend it to anybody who’s having trouble recovering.

Nate: This is a kind of double-edged sword. People are very good, especially smart people, at rationalizing, and that can be rationalizing negative behavioral pathways.

Cate: Absolutely.

Does everyone actually want agency?

Nate: What if you’re just, quote-unquote, genetically super high agency? That’s one critique I’d have of the book: for a lot of people, it’s great to have every door be open. One of the important things you talk about is increasing your surface area for luck. It seems cheesy, but you go out and don’t just sit in your apartment, and you have conversations and meet people and explore the world. When I think about the friendships I’ve developed, a lot of them, it wasn’t like I planned, “I’m going to go make a new friend.” Usually it was like, I didn’t want to go to this stupid fucking event, and you go anyway, and you meet somebody. It’s more spontaneous than planned. But do all people want choice and agency?

Cate: I don’t know if everybody wants that. So the question is, to what extent do people differ in their capacity for this, maybe in addition to the desirability of it. Is that fair to say?

Nate: Yeah. I know people who are kind of paralyzed by choice, is the cliché. Same with the donkey who [is stuck between] two [pools] of water and they wind up staying in the middle. I think some people find the amount of choice kind of overwhelming, potentially.

Cate: I don’t know exactly what accounts for this, but experientially, the feeling of having more agency doesn’t feel like having an overwhelming amount of choice. You can think in terms of knowing what you want, and then knowing what to do about that, and I think about agency as more the latter than the first one.

There is an element of agency that’s about seeing choices you have that you might not otherwise recognize. But I think you can be really high agency in many senses and still not have a great idea of where to direct that agency, because actually knowing what you want in life is sort of a pre-agentic behavior. I know people who are really good at picking the actions that will get them the real-world result they want, but not at thinking about what they care about. I’m not actually sure that agency is super helpful when it comes to discriminating, at a basic level, between the different things you could choose to do with your life, or different options. So you may need an additional book for that.

Nate: It’s interesting in the debates about large language models. They have a more explicit utility function or loss function than people do. Sometimes when I wake up, Cate, I don’t know what the fuck to do. You wind up doing things that are firm and tangible. Having a Substack requires a lot of agency, or comfort with that. I can’t get fired, really, unless I’m more stupid than usual on Twitter. But you have a very direct linear relationship between the amount of work you do and the number of subscribers you get. In some ways, it’s a little bit crippling. Back when I worked for ABC News, I would take as long as possible at the World Series of Poker, because I had no incentive-based compensation, and I thought they were going to kill FiveThirtyEight, which they did. So I was like, “Fuck you.” Now it’s money out of my pocket every time I’m not writing a newsletter or doing something to promote the newsletter.

Cate: This is a very poker player way of thinking, too. One of the traps poker players fall into is that for every moment you’re spending on leisure activities, it’s really easy to figure out the exact hourly rate you’re sacrificing by not being at the table, and so people get into weird psychological loops about this.

Nate: When I was playing online poker — back then the games were really easy to beat, at least in my opinion — it was kind of like having an ATM. A high-variance ATM, where on average you withdraw $200 an hour. Sometimes, oops, you lost $10,000. Sometimes you gain $15,000.

This is one of the tensions you’re exploring in the book: how explicit [to be] or not to make [behavioral choices]. One thing people would be better off being explicit about is thinking about things that save you time or save you money. I remember I was traveling to Europe with my partner, and we had forgotten to bring the European plug adapters, and I said, I know it’s overpriced, but I’m going to buy chargers at the airport. I’m sure they’re marked up by 15 euros, but otherwise you have to find some little electronics store in the center of the city, and your credit card doesn’t work. But just taking the time to like give yourself the luxury of convenience. On the other hand, if you’re totally maximizing for everything, it feels like you wind up being quite miserable. And some things feel icky to be more explicit about.

You give a lot of concrete advice in the book about situations in the workplace, and to some extent situations involving friendships and relationships, where people are not very good at communicating things explicitly — especially men. I don’t think I’ve gotten in a real fight with a friend in a long time, but maybe I should. Maybe friendships drift apart, and maybe I should be getting into more arguments with my friends. What’s the barrier to communicating more explicitly and effectively? There are lots of examples in the book. I remember one about having a friend who always brings along a third friend, and you’re like, that’s a little bit weird. How do you broach those conversations?

Cate: This is one of the pieces of advice in the book that I’ve found most personally transformative: instead of having a feeling internally, just sharing it in a lot more circumstances. It feels extremely not allowed at first to just make a bunch of subterranean dynamics explicit, but once you start getting in the habit of it, it feels so liberating. Things I do at work all the time are just checking in with people: “I feel like you might be feeling a bit frustrated. I can imagine these reasons for it. Is that the case?” In other social circumstances, sometimes it’s, “I feel like maybe we have this awkward dynamic behind the scenes. Do you want to talk about that?”

It’s really interesting to think about why people never do this. The way I frame it in the book is “movie logic,” because in movies you see characters doing things where you’re just like, why don’t you just have a fucking conversation about what is happening here, instead of taking self-harming actions? And I wonder if it is a little bit downstream of movies, where in movies you need that for dramatic effect, but people download their ideas about the types of conversations that it’s okay to have from movies or entertainment. So I encourage people not to try to build dramatic tension in their own lives, but to just simplify things by making dynamics as explicit as possible.

Nate: I feel like there could be movies that end after 15 minutes. A couple rents an Airbnb and it looks like it’s haunted, and they’re just like, “I got a weird vibe about this place,” and they get a different Airbnb across town, and the movie ends.

Cate: One of the examples we use in the book is La La Land, where the whole blow-up of this great relationship could have been avoided if they’d just had one conversation where she said, “I want to see more of you,” and he said, “Great, I want to be on tour so that I can provide for us. How should we balance those competing considerations?” But then it’s not a very good movie, and it doesn’t win an Oscar.

Acting your way into right thinking

Nate: Is it fear of failure? That sounds very basic on the one hand, but one thing about being an entrepreneur, about being a poker player, about quitting your Supreme Court–track job to play poker, is that clearly you’re not afraid of failure, Cate. But how can other people learn that skill?

Cate: I used to absolutely suck at this. Part of the reason I became a lawyer without actually caring about being a lawyer was that it never required me to feel rejection or failure. It was super easy to be good at it. It was super easy to get into a good law school because of the type of brain that I have. And I more or less implicitly never wanted to be in a position of failure. I remember having this feeling up through my 20s that if I wasn’t literally the best at something, not only was it not worth doing, but life was measurably worse and maybe not worth living. I had a really extreme view of failure. For myself, addiction was pretty helpful for getting me over this, because I started failing a lot, and I sort of reset my expectations for what a successful life looks like.

There are some intellectual tricks in the book for thinking about failure. One that I’ve found persuasive is this concept that if you’re not putting yourself in a position to fail, you’re necessarily capping your total returns in life, because you’re not doing the things that are lower probability but higher consequence. But I think this is one of many areas where you have two ways of looking at an issue. One is, I will be able to think my way into acting better — I’m going to have the right set of thoughts about failure, and then failure won’t hurt anymore. The other is acting your way into right thinking, and a lot of the book is about acting your way into right thinking. It’s going to be unpleasant, but it’s a lot less unpleasant than you think, and the more you do it, the less unpleasant it feels. I’ve gone from a person who actually kind of wanted to die when I failed at things to just, “That’s fine, next thing,” over the course of my life. And it feels pretty learnable.

Nate: One thing you learn from poker — actually, one of the worst habits I probably have as a poker player — is that I will take actions that remove decisions later on. So classically, you might go all in, because then the hand’s over. When, in principle, if you’re a skilled poker player, you should want decisions that complicate things, where you have sub-decisions further down the line, where you have more options and more opportunity to press your skill. But people don’t seem to like that.

On your blog, you define agency as the entirely learnable capacity to see and act on degrees of freedom. But the learnable part — people might not realize that. People think, “Oh, I’m introverted,” or “I have impostor syndrome.” How do you persuade someone that they can get better at this stuff?

Cate: Hopefully part of the answer is the book. Identity is one of the things that keeps us kind of stuck in life. Having an identity is useful for a lot of things that you might want to do. But people also develop limiting identities: “Well, I’m not a charismatic person,” or “I’m not a confident person, I’m an introvert.” And they start to think about those things as immutable characteristics, as opposed to things primarily composed of learnable sub-behaviors.

Charisma is an example that I use in the book. People think about themselves or other people as being naturally charismatic or not, and it is true that there is natural variation in terms of how charismatic you are by default. But a lot of the things that signal charisma, that get interpreted as charisma, are actually learnable sub-behaviors: the way that you interact with somebody physically, the types of things that you think to talk to them about […] I think agency is an example of this, where when you break it down into more learnable sub-behaviors and small mental shifts, it becomes quite transmissible.

I happened to osmose a number of agentic habits from my environment over the course of the last decade. I think there are other life circumstances — like the desperation caused by addiction — that sort of prompt you to see new degrees of freedom. But that’s kind of the reason that I wrote the book. So please read the book.

Spiky agency in the Bay Area

Nate: You live in Berkeley, or Oakland, something like that?

Cate: The Bay.

Nate: How do you think about the typical California AI-rationalist-adjacent person? Are they high agency? You go to something like the Manifest conference and people are certainly on their own wavelength — kind of eccentric, kind of [a] hippie vibe, and there’s a lot of talk about agency. Are these people agentic? Because there’s also a lot of groupthink.

Cate: I think that’s true in a lot of subcommunities, right? Are you familiar with David Chapman’s “geeks, MOPs, and sociopaths” idea, about the different stages of development of a subcommunity?

It’s basically that the people who start a new subcommunity are generally free thinkers. They have their own ideas that differ from those of society. But the longer that subcommunity persists, the more people who are doing following behavior and not thinking about things from first principles get into it. It’s just kind of part of the life cycle of any community: it starts with the people who are wild and less constrained by what has been done before, and then it moves in the direction of conformity.

On the whole, rationalists and EAs are very high agency. I have found the same thing to be true of them that is true of most people generally, which is that people tend to be selectively agentic. People can be super high agency in certain areas where their self-identity allows it. I was professionally agentic, and agentic in the sense of noticing problems in the world and wanting to do something about them, long before I became agentic in the sense of figuring out who I wanted to be in an interpersonal sense, and how I wanted to relate to my romantic partners and especially my husband. And that’s a very common profile in the Bay Area: super high agency in certain domains and then lower agency when it comes to addressing some of the things that feel more like identity. “I am an introvert.” “I am” — you know, whatever, you pick your adjectives. Spiky.

Nate: I met for dinner with a startup founder recently. Really, really young guy, very smart guy. I’m sure he’ll be very accomplished one day. And he’s like, “I realized that I’m really smart, but I was very socially awkward, and I realized that was my big barrier to success.” And for a nerdy startup founder who now has, I think, probably above-average social skills, it doesn’t seem fake or contrived in the way it can a little bit sometimes from Bay Area people.

But something you said earlier kind of resonated with me, which is not letting yourself get pinned down to any one identity. You have a wider range of different types of people you’re interacting with, you get better at code-switching, and you’re kind of hedging your bets a little bit more. Agency and optionality, having lots of good choices, are very closely related. That’s hard for people, though. You’re a CEO and a co-founder and a poker player and an attorney and an author and a Substacker. You saw colleagues who were going to have some house in Maryland or Virginia, and they win some award one day, and then they’re 65 and retiring and thinking, “Man, my life sucked.” Why do people feel the need to stay typecast, or why can’t they break out of it?

Cate: I think it is one of the key stages of personal development, and it is one that probably many or most people never get to. There’s another way of thinking about personal development that my husband is a super big fan of, which is [the] Kegan stages. A lot of self-help and coaching is aimed at one particular transition in the Kegan typology, which is going from stage three, where your sense of self and of what is good is environmentally determined, to one where you are sort of choosing that for yourself. And I think it’s a minority of adults that are able to attain that.

It is still a little bit of a mystery to me why it happens for some people and not others. Certainly, the more constraints you take into your life, in the sense of sets of expectations you have adopted, the harder it gets. The classic thing is that once you are married and have kids, it becomes a lot harder for people to consider changing their whole identity around. There are good reasons why they don’t want to do that, so I think that’s a constraint. If you can do it when you’re young, that’s helpful. But it’s strange and tragic to me that it does not happen for more people, and I really hope that part of what the book does is prompt a few more people to have that kind of self-discovery — that maybe the way that they have arranged their life already is not the way that they want to have it forever.

Nate: One other thing that you talked about that was a profound insight is that sometimes when you’re at the beginning of a new relationship — maybe more often a work relationship — people become afraid of confrontations that define the scope of the relationship. Whereas for me, if you’re unable to resolve a small conflict early on, inevitably you’re not going to be able to resolve the bigger conflict later, when there’s a lot more wear on the tires in two years, three years. So not being conflict-avoidant actually saves you trouble down the road.7

Cate: Massively. Yeah, I think that’s massively true, and I’ve certainly found it to be the case for me. I know that my default tendency is not to raise stuff as early as possible, and instead to allow resentment to accumulate over some period of time until it becomes a much bigger deal. And a lot of maturity for me has been learning to have slightly uncomfortable conversations at a much earlier point in time, so you don’t end up putting yourself and the other person in a much worse situation further down the line.

Nate: Do you find that a lot of people would also want this? With friendships, there’s this thing where you silently reciprocate. Let’s say both you and your friend [secretly don’t feel] like hanging out. Sometimes you’re like, we’d probably both be better off if we just spent the night in. Or the reverse: I’m really bored, we hung out last night, but I’d really like to catch up, so maybe we’ll go hang out again today. Why do people find it hard to communicate these things? Or do most people think you’re weird if you’re that explicit about things that maybe [most people wouldn’t] talk about directly?

Cate: Having written a book about agency and addressed some of the topics that I have, I feel like I should have a better theory of why the human default is the way that it is, and I don’t. I do know that when I am honest in situations like that, by and large the response that I get from people is relief at the honesty. And I know that there is some selection effect — I end up selecting for close relationships with people who want that and value that — so perhaps my vantage point is no longer representative, and actually most people will hate it if you’re honest with them. But I kind of would be surprised if that’s the case. I live in the Bay Area. I don’t know what normal people like.

Nate: Selection effects are very powerful. The choices you make create path dependencies, and those are very important, and I think people are sometimes reluctant to open other doors. But some constraints are helpful, right? Like, the first thing that I’m going to do every morning is go work out or go run, and if I don’t do that, then my whole day is ruined. It seems like sometimes people need these absolutes, where it’s [harder for them] to say, “Okay, eight out of 10 days I’ll run in the morning, but sometimes it’s not practical.” Sometimes the routines are helpful. Or do you see that as a substitute for more sophisticated forms of agency?

Cate: Yes, it is sometimes good to have standards and routines for yourself. Having forcing functions, where you constrain your freedom in limited and sort of pre-committed ways for the greater good of some goal that you care about, is in general really good. And the question is just about how you scope those commitments so that you are adhering to them and they’re having the desired effect. We use the metaphor of tying yourself to the mast in the book, which is very timely because The Odyssey came out this week. I think that is good for routines, and I think it can be good for relationships. My relationship with my husband is phenomenal, and probably also would not exist at all if we hadn’t gotten married pretty soon after we got together. The first year and a half of marriage that we had was pretty difficult, and without the forcing function of having already made a commitment to each other, that probably would have driven us apart.

AI and the disappearance of friction

Nate: Can we talk a little bit about agency and AI?

Cate: Yeah, totally.

Nate: In my book, I kind of hypothesize that AI might be helpful for hyper-agentic people, but deprive a large majority of the world’s population of agency. When you hear Sam or Dario or people like that talk, do you think they put enough value on people’s agency, or do you think that’s a blind spot, potentially?

Cate: I think it’s a really hard question. I agree with your overall take that AI will, on the scale of individual humans, be either quite bad or quite good for agency, and that won’t be uniform across humans generally. I have found that AI gives me the capacity to do a ton of things that I would not be able to do otherwise, and it has freed me up to experiment with a lot of things professionally that I otherwise would not have.

But if you take anything from the example of recommender algorithms, it should probably be that technologies that are designed to eliminate friction in service of giving you a particular experience can be really agency-destroying. And I am worried about moving into a way of relating to technology where we are shaving as much friction off of the human experience as possible, and what that will do to people’s sense of an internal locus of control, or actual ability to affect the world. I hope that it ends up being good for human agency, but I am skeptical, for I think the same reasons that you are.

Nate: Yeah, friction can be underrated. Despite recent events related to Taco Bell, I kind of actually like Taco Bell. If there were a Taco Bell in my apartment, like there is in some of the Taco Bell commercials, I’d probably gain 30 pounds overnight, just because having some barrier matters. You see this in a more serious way with gambling addiction and sports betting. The physical act of going to a place and standing in line at a sportsbook and placing a wager and getting maybe the slightly judgmental look from a ticket writer — there’s no hard barrier there, but in practice, that’s a lot harder than downloading an app and pressing a button. You’d say people just have more choices now, but it’s a little bit complicated with any type of addictive behavior.

Cate: I think the gambling analogy captures some aspect of it. With sports betting, it’s probably not the feeling of having earned the sports bet that makes the friction the good thing. But there is an aspect of that in a lot of aspects of life, where actually the feeling of overcoming barriers is where a ton of the value in the human experience lives. I think you will hear this all the time from people who have recovered from addiction, but you also hear it from other people. I’m deeply skeptical of the ability of humans to have really good, meaningful lives if you eliminate all opportunities to feel like you are overcoming some barrier in the world. And obviously that has limitations — I don’t think that all suffering is inherently good because you have an opportunity to recover from it — but having a perfectly smooth experience also does not seem to produce human flourishing.

Nate: That’s the problem with all these utopias, whether it’s a socialist utopia or a post-AGI singularity utopia. I want fucking conflict. I want to have to fight for things. I don’t want some Holodeck-type thing where I can have whatever I want. That seems like it would be almost depressing in a way. I think utopias profoundly misunderstand human nature.

Cate: It’s really hard to write utopias for that reason.

Nate: How do you feel about the notion of revealed preference, which is another slightly nerdy, rationalist-adjacent term? Let’s say I’m the kind of person who’s going to start working out. If you go weeks and weeks and weeks and you’re paying the gym membership and not working out, on some level, is that your revealed preference? How should we weigh the fact that what people think they should do, and what people say they’re going to do, and the way they actually behave are different?

Cate: I get a lot of shit from my husband for my views about revealed preference. I am not a fan of revealed preference as a concept, simply because I am intimately familiar with the ways that you can really, really want one thing and do something else instead. And chalking everything up to revealed preference obscures a lot of useful information about what is going on underneath the surface. There really is a difference between somebody who is trying to recover from addiction and is wanting that, and somebody who’s just excited to keep using. I do get pushback from my husband on this, and I think I’ve made more of a bugbear of this in the past than I am ready to stand by now.

My current way of thinking about things is that all of the difficulty goes away if you stop expecting yourself to be a single unitary entity that has one set of desires that is carried through you at every level. And instead, what we are is some sort of parliament of sub-personalities that have different holds on us at different times. It’s kind of like a negotiation between the desires of different parts of us, but not a unitary self.

Nate: Yeah, it can be kind of lazy in a way. A lot of people become aware of the concept of revealed preference, and [they might be like] yeah, my revealed preference is to stay out playing poker [all night] at the Bellagio the night before I play the Main Event. Just because you do that every year doesn’t mean it’s a good habit.

Cate: Yeah, and it doesn’t mean that you really want it in all of the relevant senses.

Nate: That’s what’s weird about it, too. Typically when I go to Vegas, I’ll usually have some night early on where I stay up too late playing cash games, and it kind of fucks up my schedule for a couple of days. And the funny thing is, I kind of know I’m going to do it. In poker, the game usually gets better later at night, so it’s harder to quit. If you’re down money you want to keep playing, and if you’re up money you’re having fun and don’t want to quit either. It’s sort of weird how, without commitment devices, people [track] into their future less rational self … maybe revealed preference is overrated.

Where should you start?

Nate: Cate, this is like an Oprah question, but if someone wants to become more agentic, what are two or three things they can start doing today?

Cate: My number one piece of advice is one that nobody ever wants to hear: I really recommend having an anonymous feedback box, where you tell everybody you know that you’re doing it, you put it on social media. I think that a lot of what keeps us stuck in life is failing to get accurate information about ourselves from the world, and having an anonymous feedback box has been totally transformative for me.

People don’t want to do it because they don’t want to hear mean things about themselves. But even for somebody who’s a little bit publicly visible and has people who don’t like them, the proportion of stuff that I get that is intentionally mean is very, very small. There is a lot of stuff that is like, “I don’t know if you realize that you do this, I don’t think you intend it, but —” useful pieces of feedback on my management style. And I feel like I’m in a much better place today than I would be without it. I recently updated my sense of personal style and developed a wardrobe because of a piece of anonymous feedback that I got a few months ago. So that’s my number one thing.

Nate: That sounds terrifying to me. Maybe it’s because I do get a lot of unrequested feedback.

Cate: People are so mean to you just to your face. How much worse could it be?

Nate: They’re not, really. I’ve only had maybe once in my life, a negative interaction in person. Most people are cowards … behind keyboards. One time at the 2016 Democratic convention, some Bernie guy got up in my face.

Can you tell who the person is who’s giving you anonymous feedback, or does that not really matter?

Cate: I try not to fixate on it, and people are generally pretty careful to frame things in ways where it’s hard to know.

Nate: As the world becomes more comfortable, does that make people less agentic?

Cate: As it becomes more comfortable?

Nate: Yeah, because you can kind of just go with the flow, and now things are better, usually. You have better medicine, you live longer, the food’s better. I don’t know if the housing stock is better. But you can kind of just be more of a bot and not think as much. If you look at the history, people were miserable for most of human existence. There’s this false nostalgia for, oh, I wish I lived in the time of Homer. People were pretty fucking miserable for the most part.8 So is it easy just to tune out and not be agentic?

Cate: I think this is kind of the concern with AI, which is that things just become really easy in ways that are not ultimately productive for self-development. I do think there’s a difference between being active and being agentic. The bull case for human agency in an era of advancing technology is basically that less of your time and attention is consumed getting your basic needs met. Basic can just be shelter and food, but it can also be whatever standard of living you want to have. Maybe that requires less and less effort from individual people. And what happens when you free up that time, what you do with it, is kind of up to you. You can think about it as an invitation to greater agency, because the default is just boredom and stasis, but it also creates an opportunity for action in a bunch of directions that you might not otherwise have.

Nate: I do think it’s worth stating that being healthy and able-bodied, not being actively discriminated against in whichever society that you’re part of — I’m not, like, super woke, but there is some element of privilege in having a lot of agency, right?

Cate: I think that’s true, and I’m hesitant to speak for other people. I do think one of the nice things about agency is that for most people, in most circumstances, there is an opportunity to develop more of it, and it can be useful no matter what your situation is. It’s kind of just a generally useful skill set. It doesn’t require you having a super-genius IQ or a lot of connections. It doesn’t mean that there aren’t things that can constrain it, there absolutely are, but I think there is more room for malleability in different sets of circumstances than people maybe naturally think there is.

Nate: Are there things that you think a woman reader might get out of the book, or a [male] reader, that the other gender wouldn’t? Because, to stereotype a little bit, agency feels very male-coded.

Cate: Yeah.

Nate: What would you say to women who are reading this book in particular?

Cate: So I thought about this, because the PR firm that I hired was really excited about pitching me for a bunch of women’s shows about how women can be more agentic. And I thought about, is there an actual angle here for women in particular? Is there a different message for them that I would want to give? And I think my answer is just, not really. I think it’s true that agency is more male-coded. It’s also more dark-triad-coded than baseline-human-coded. And the core desire that I have in writing the book is exposing these skills of agency to a greater range of people who can benefit from them. I don’t know if the average woman will take more than the average man from it as a result of that, but I think it’s a pretty thoroughly all-purpose book.

Nate: Is there anything else you want me to ask, or want to ask me, or any other plugs for the book? Which I really do recommend — I read it twice, in galleys and again yesterday prepping for this. It really is a book that isn’t easily pigeonholed.

Cate: The only other plug that I would give for it is that we tried to be absolutely ruthless when it came to prioritizing the reader’s time. We started with an 80,000-word manuscript. We worked it down to a 50,000-word manuscript. It’s something that can be read by most people in a long afternoon, and I’ve been getting reviews from people who read it in one sitting. So it is short. There are no wasted words. It’s written for a Substack audience in particular. I personally do not buy a lot of self-help books, for multiple reasons, including that they are time-wasters. But I would encourage you, even if you don’t think about yourself as a self-help person, to consider buying the book and giving it a shot.

Nate: This is to say, “self-help” isn’t a bad word. But this book, I think, will appeal to people who would never be caught dead buying a self-help book.

Cate: Personal development is what we call it.

Nate: There we go. That’s the euphemism. Thanks so much, Cate.

Now that the World Cup is over, we’re fully focused on getting the midterm forecast up and running as soon as possible. I’m going to be honest that between publishing a ton of stories over the past two months, highly competitive primaries, redistricting, and various other distractions, this has taken a little longer than we were hoping for. So we’re going to err on the side of publishing less for the next couple of weeks until the model is ready to go.

However, one benefit of working on the model is that there are usually some fun spin-offs from the various data that we collect for it. For instance, in calculating the state-by-state correlations used by the model, we rely on voter-level microdata from the Cooperative Election Study, a large-sample online survey conducted every year or two. We’ve aggregated responses to the CES in each year it was conducted during the “Trump Era” — from 2016 through last year — into a dataset of more than 400,000 voters.

That gives us high-fidelity data for capturing voting trends among smaller subgroups that usually get lost in the shuffle. Today I’m going to focus on one such group: voters with high educational attainment but low to middle incomes. Because income and education are positively correlated, there aren’t all that many of these voters. I’d argue, though, that they’re vital to understanding emerging trends within Democratic politics, and particularly the rise of the Democratic Socialists of America (DSA).

Subscribe now

The DSA occupies the high-education, lower-income corner of the electorate

As a result of contentious Democratic primaries in Michigan, Maine, and New York, I’ve been reading a lot of conversation about the role that class plays in voting patterns in the United States. In these primaries, candidates like Graham Platner and Abdul El-Sayed who are associated with democratic socialism often claim to have the mandate of the working class. But in practice, their strengths tend to be with college-educated voters. Even before his scandals, Platner wasn’t particularly doing better among voters without college degrees than previous (losing) Democratic nominees in Maine. El-Sayed has relatively weak support from Black voters in polls, meanwhile.

But does that make them hypocrites? For what it’s worth, they certainly both have interesting backstories. Platner didn’t graduate from college, but came from a fairly privileged background. I don’t think it would be unfair to classify him as “downwardly mobile.” El-Sayed is in some ways the polar opposite: a first-generation American who went to a good public school in suburban Detroit and is obviously a high-achieving person.

The larger point, though, is that I think we ought to be careful about conflating class with educational attainment. Apologies for the cliché, but class in America reflects a tapestry of race, educational attainment, and income/economic privilege — and even that doesn’t capture everything.

I’m reminded here of Peter Turchin’s elite overproduction hypothesis. It basically argues that America is producing “too many” college graduates relative to the number of high-prestige, well-paying jobs available to them. People who wind up in this situation often carry heavy student loan burdens or other sorts of debt. Even if their material situation is objectively “just fine,” it’s rarely great. They’re gloomy about their own financial prospects, and feel aggrieved about the state of the country. Their class status is ambiguous: privileged in some senses, but they’re part of the “have-nots” in other meaningful ways.

What would be ideal here, to get closer to the notion of class, is if we had data on voters’ family backgrounds. Did they grow up rich or poor? Were their parents white-collar or blue-collar? Did they go to a private high school? Unfortunately, you’ll rarely, if ever, see this sort of question in political surveys, even ones as comprehensive as the CES.

Still, we can get a fair amount of texture by running through various permutations of income, educational attainment and race, all of which are available in the CES. I’ll look at who these voters preferred1 in the 2016 through 2024 presidential elections.2 Putting race aside just for a second, I’ve classified voters into four groups based on educational attainment, cross-hatched with five groups based on their self-reported household incomes.3

Share

Based on this grid, what is the most Democratic voting group in America? It’s voters with postgraduate degrees but lower-to-middle household incomes of $30K to $60K per year.

This is very much also the sweet spot for the DSA. In the group’s most recent member survey in 2021, 80 percent of members aged 25 or older had bachelor’s degrees, but 45 percent had household incomes below $60,000 per year.

This is unusual, because education and income are usually substantially positively correlated. In the composite CES data, only 19 percent of Americans with bachelor’s degrees or higher had household incomes of $60K or below. The DSA rate is around twice that.4 These voters aren’t necessarily “blue collar” in the sense of working with their hands. But that doesn’t mean they’re privileged. Common occupations among DSA members include teachers, non-profit workers, social workers, and public sector employees.

By contrast, who are the most Republican voters in the country? It’s not necessarily voters with a lot of “economic anxiety”. Rather, it’s the precise opposite corner from the DSA: high-earners who didn’t attend college. About 60 percent of voters with a high school education or less, but whose households earned incomes of $60,000 or above, preferred Trump to the various Democratic presidential candidates. (You can probably think of various stereotypical professions here too: small business owners, independent contractors, real estate developers, used-car dealers, etc.)

Could this just be some weird composite artifact of race? In the DSA’s 2021 survey, 85 percent of members reported their race as white5, which certainly isn’t representative of the country or the Democratic Party overall. But in the broader CES data, we find that the patterns mostly hold once you disaggregate by race. Let’s speed-run through the various major groups. Here is the CES data on voting by income and educational attainment among white voters alone.

Among whites, there’s particularly strong stratification in political preferences by educational attainment, with incomes making less difference. Still, there is some very mild “U-shapedness” — i.e., both the poorest and the richest whites are more blue than middle-income ones. The most Democratic group among all white voters is the same one as in the aggregate data: voters with graduate degrees making between $30K and $60K per year.

Next, the same numbers among Black voters:

While it’s true that Black voters are not a monolith, their preferences are notably more uniform than for whites — and mostly very blue. But what’s the most Democratic Black voting group? You guessed it — Black voters with graduate degrees earning $30K to $60K per year. In fact, with 95 percent of voters who meet this description voting for Democrats, it’s probably the most D subgroup in the entire voter universe.

How about among Hispanics?

This chart hits a bit differently. Among Hispanic voters, voting is considerably less correlated with educational attainment than among whites. However, Hispanic voters earning higher incomes are considerably more Republican than lower-earning ones. Hispanic-Americans (and Asian-Americans) have high rates of small business formation. As I’ve argued in the past, Democrats may be doing a poor job of appealing to entrepreneurially-minded Americans — and DSA-type messages may be particularly unappealing to them.

Still, it’s basically the same pattern again. The top-right corner of Hispanic voters (high education, low-to-middle income) is blue. The bottom-left one (high income, less education) is much more purple.

High-education, lower-income voters are cultural libs, too

Another complication with candidates like Platner and El-Sayed is that much of their messaging isn’t focused on bread-and-butter economic issues. Instead, like other Democratic factions, democratic socialists can believe in the “omnicause”. Apart from Medicare for All, the signature issue of El-Sayed’s campaign is Gaza, with a huge amount of his campaign communications focused on the extensive spending by AIPAC and affiliated groups in support of his opponent, Haley Stevens. Meanwhile, in his message announcing he was formally withdrawing from the campaign, Platner signed off with a series of cultural signifiers — “F*ck ICE.6 Free Palestine. Up the Hearts” — the latter a reference to Portland, Maine’s professional soccer team.

That’s not to say the DSA wouldn’t be better off in the long run if it were more agnostic on cultural issues. Even Bernie Sanders once had relatively moderate stances on immigration and guns. Even Zohran Mamdani has backtracked from his more left-wing positions on issues like policing.

But politicians who truly have some long-term vision are rare. Most of them are trying to play the ball where it lies. The fairway consists of college-educated, mostly younger, mostly white progressives who are disillusioned by their economic opportunities and/or with the direction of the Democratic Party. We could get into the sociology behind this, but this is a group that typically uses “post-material” cultural issues as their rallying points.

In addition to asking voters about their presidential preference, the CES also asks voters to rate their ideology on a scale from 1 (“very liberal”) to 5 (“very conservative”). “Very liberal” wasn’t a very common response: only around 10 percent of the electorate. But which voters were most likely to check the “very liberal” box?

You’re probably getting the hang of this by now: it’s voters with graduate degrees making between $30K and $60K a year. Some 21 percent of those voters say they’re “very liberal”, about twice the population average.

As I discussed last week, voters who describe themselves as very or extremely liberal tend to place more emphasis on cultural issues — particularly gay rights, trans rights, the environment, race relations and Israel/Palestine — as compared to other voters. They’re only about average or even a little below-average in how much they prioritize things like health care and jobs, by contrast.

So what Platner and El-Sayed are doing might be smart politics, particularly in El-Sayed’s case given Michigan’s relatively high number of Arab-American voters.

Noah Smith recently argued that the “socialist left” and the “woke left” should be regarded as separate political movements. But I’m not so sure that’s true. This data helps to explain the relationship. Messages like “Fuck ICE” and “Free Palestine” are catnip for college-educated liberal Democrats of all kinds. Culturally left-wing but economically comfortable Democrats are probably more likely to join other Democratic factions like the #Resistance. Culturally left-wing but more “economically anxious” ones are potential targets for more anti-establishment factions like the DSA, by contrast.

DSA members aren’t necessarily struggling to pay the bills, nor are most of them traditionally “blue-collar”. But they’re likely to be genuinely upset about their station in life as well as how things are going with the country. Without wanting to psychoanalyze too much, it can be appealing to anxious-feeling voters of all political persuasions to feel as though they’re part of some greater, more altruistic struggle against corrupt elites or injustice. The DSA is betting they’ll stick around for the socialism — but the culture wars are often what get them in the door.

PELE, the new international soccer model that Silver Bulletin launched in conjunction with the 2026 World Cup, uses a mix of match results, player market values, and historical economic and geographic data to rate all 211 FIFA teams. PELE covers all competitive international soccer matches since 1872 and is updated after new matches are played.

PELE was quite accurate in the World Cup. The system’s top 2 teams in our ratings as of the start of the tournament (Spain and Argentina) met in the final, and our #3 and #4 teams (England and France) also finished with those exact rankings after being eliminated in the semifinals. It also identified some more surprising contenders like Norway.

We’ll continue to update PELE with other international match results and will consider other special extensions in the future, such as for the 2028 European Championship. We’re also evaluating whether to extend PELE to club soccer.

You can get a more formal introduction to PELE on the methodology page. But the best way to demonstrate the breadth of the system is by example. So let’s start with our overall PELE ratings. One tip: we’d encourage you to view PELE on the web rather than in email or the app.

Share

PELE ratings

Each team gets a continuously updated pair of ratings: PELE and Tilt. PELE is the headline number that predicts overall team quality. It’s basically a fancy variation of an Elo rating system, though it’s affected by changes in player market values and player ages as well as match results.

The rest of this page contains:

• A comparison of PELE and FIFA rankings;

• Tilt ratings;

• Offense, defense and round-robin results;

• Future match projections;

• The Atlas of PELE, our 12 footballing regions and home-field advantage;

• Market values and projected team trajectories;

• And, detailed historical PELE data since 1872, including the best teams of all-time.

The last few “advanced” sections are a bonus for paying subscribers. We greatly appreciate your support and we’ll consider extending PELE to club soccer depending on the amount of interest in the model.

Subscribe now

Ready for some ranking controversies? You can already get a few hints for how PELE differs from other systems. Because PELE is designed to be predictive, it can be more sympathetic to recent underperformers and more skeptical of teams on a possibly unsustainable hot streak. Market value data can help teams with star talent like Norway in PELE. Here’s how PELE compares with the current FIFA rankings.

Tilt ratings

Offense and defense are more fluid in soccer than in sports like American football, but there are some predictable patterns. Tilt ratings indicate a team’s tendency toward attacking or defensive play. They are based on both long-term tactical preferences and the personnel on the field — the presence of Erling Haaland makes Norway more attack-minded than they otherwise would be. Tilt measures the impact on the total number of goals scored combined between both teams.1 For instance, if Germany’s Tilt rating is +.6 (“very attacking”) that means games involving Germany will tend to feature 3+ goals rather than the ~2.5 that is more common in modern international matches.

Offense, defense and round-robin results

Now that we have both PELE and Tilt ratings, we can do some neat things. Most importantly, we can forecast matches! Our round-robin results project every pairwise combination of match results between the 211 current FIFA members (weighted slightly toward more challenging matchups2). Also, this process allows us to derive offensive and defensive ratings for each team, as measured by the projected number of goals scored and allowed against the same competition. Teams with negative Tilt ratings like Colombia and Senegal will produce lower-scoring matches with more draws.

Future match projections

But better yet, let’s forecast some actual games. We’ve uploaded the schedule for all future international matches that we’re aware of. For games eligible for extra time, the projections reflect the results at the end of regulation play, consistent with how the odds are usually listed at betting sites.

Read more

⚽ The latest World Cup predictions: Updated July 19, 2026

Spain wins its 2nd World Cup as Argentina is overwhelmed.

Thank you so much for following our World Cup coverage! We were very pleased with the reaction from subscribers — and honestly also with the results, as the top 4 teams in PELE at the start of the tournament (Spain, Argentina, England, France) also finished in that exact order in the official standings.

I don’t know if you’d call the final an anticlimax, but Spain was obviously the better side. Even if they benefited from a second yellow card that sent off Enzo Fernandez, putting them a man up, Argentina generated literally zero chances until very late into extra time. Spain is very, very good — already one of PELE’s top-rated teams of all time — relatively young, and will co-host the World Cup in 2030, with the chance to build a potential dynasty.

We’ll continue to update PELE with international match results and consider future expansions of PELE, including 2028 Euro and 2030 World Cup forecasts, as well as possibly extending the system to club football.

We’ve updated the team rankings and historical data sections with the result of the final, but left other sections unchanged from before Spain’s win. This is the last and final update to our World Cup page! For our methodology, see here.

See also: PELE International Football Rankings.

Subscribe now

Jump to a section

• Part I — Team ratings — PELE ratings, adjusted for the World Cup

• Part II — Knockout projections 🔒 — odds at each stage, likely opponents, the bracket

• Part III — Match projections 🔒 — win probabilities, scorelines, and the knockout score matrix

• Part IV — Track record & history 🔒 — historical performance, upsets, milestones

So, who’s gonna win the World Cup? While our model doesn’t deviate too much from the conventional wisdom, we don’t have teams in quite the same order. Each time we run the model, we play out 100,000 simulated World Cups, accounting for the difficulty of each team’s draw, meticulously calculating tiebreakers, and even the possibility of a dark-horse contender getting hot and unexpectedly being in top form.1 Let’s look at the PELE ratings and some specific adjustments that we’ve made to them for the World Cup.

Part I. Team and group ratings ⚽

Our baseline PELE ratings are adjusted in several ways for the tournament. First, we account for differences between PELE’s algorithmically calculated rosters and the actual, announced World Cup rosters. Thus, we correct for the impact of injuries or otherwise unexpected player absences.2

Second, we update each team’s rating based on the quality of their performance relative to PELE’s expectations for each match. And third, we adjust for home-field advantage on a match-by-match basis.3 The factor is customized for each team: as it happens, the U.S., Canada and especially Mexico4 have above-average home-field advantages.

Ratings and forecasts will be updated at the end of each match day5. Paid subscribers will also get access to our full suite of PELE ratings and our forthcoming midterm election forecasts.

Subscribe now

Part II. Knockout stage projections ⚽

Here is each team’s probability of winning the World Cup, and advancing to each stage of the knockout round, using an old-school, FiveThirtyEight heatmap format.

Read more

We’re probably not the only ones feeling sad about the end of the World Cup. Through 102 matches, the quality has been elite, with stars from Messi to Mbappé to Haaland having turned in A+ performances even relative to unreasonable expectations. With certain notable exceptions — England after taking the lead against Argentina on Wednesday — the play has been creative, daring and offensive-minded, and the 2.91 goals per match scored so far in the tournament is the highest since 1970. The finale is at 3 p.m. on Sunday in New Jersey, when Spain and Argentina will play one another in the World Cup for the first time since 1978.

Argentina and Spain were our co-favorites at the start of the tournament — a slight deviation from the conventional wisdom, which had France and Spain as co-favorites instead and Argentina a tier behind. So we’ll admit to feeling pleased with how the results have gone from a PELE/predictions standpoint.

But there was nothing inevitable about this outcome. In the knockout round, Argentina’s game-clinching goals have come, respectively, in the 111th, 93rd, 112th and 92nd minutes. Argentina and Spain might even have met in the R32 and not in the final if one of them had finished second in their group while the other won theirs.

We can’t promise you a great final. Historical World Cup finals have included some duds, in fact — a fair number of choppy, defensive stalemates, as well as a handful of blowouts. What we can guarantee, however, is that whoever wins will be able to stake a claim for being among the greatest World Cup champions ever. A Spain win by any means other than on penalties would leave them with the highest PELE rating of any World Cup champion immediately after the World Cup. Argentina, meanwhile, would be the first repeat champion since Brazil in 1958 and 1962, a less competitive era when only three knockout wins were required instead of five to claim the championship.

Their stories are different, however. Argentina has been remarkably persistent. Their case is about the cumulative success of Messi in his late 30s, winning not only back-to-back World Cups but also the Copa America in between. The better comparison for Argentina might not be to the 1958/1962 back-to-back Brazilian champions, but rather the more veteran 1970 Brazil squad. Still, there are veterans — Pelé, who might have won the Golden Ball if the award had existed then, was 29 in 1970 — and then there’s Messi, who is 39. Apologies to our non-American audience, but the best comparison we can think of is to that other type of footballer: Tom Brady.

Spain, meanwhile, is a team on the rise. You could say they’re “peaking at the right moment”, but as one of the youngest teams in the tournament, and with the World Cup set to be co-hosted in Spain in 2030, they could be set up for a dynasty. If Spain wins1 on Sunday, it would set an all-time record with 38 consecutive international matches without a defeat.2

Spain crashed out after winning their first World Cup title in 2010 and their second straight European Championship in 2012. They were eliminated in the group stage of the 2014 World Cup in Brazil, and the R16 in both 2018 and 2022, in both cases on penalties. Their combined record in those years: 3 wins, 3 losses and 5 draws3, more suitable for Los Estados Unidos than a team that had just broken through to become only the eighth nation to win the World Cup.

But Spain has been building momentum pretty much ever since, with a perfect record in World Cup qualifying, seven wins in seven matches at the 2024 European Championship, and a run to the 2025 UEFA Nations League final, which they lost only on penalties.

You can find our official prediction on the World Cup landing page — spoiler alert, it’s close. I hope you’ll pardon the cliché, but this really is about as close as you’ll get to the irresistible force against the immovable object. So let’s talk more about the matchup, and how we got here.

Argentina is still king of the hill

PELE, like other variations on Elo ratings, pretty much takes the view that the best team is the best team until proven otherwise. Which isn’t exactly a controversial premise, we realize. But Argentina, which had the 2nd-highest PELE rating entering the 2022 tournament in Qatar, finished with the #1 ranking after defeating France on penalties in the final.4 You’d assume they’d at least be considered one of the favorites unless there was an obvious decline in performance.

But even as perhaps one of the most popular teams in the world, Argentina wasn’t given particularly much credit by betting markets, starting the tournament with only about 10 percent odds5 of winning: about half the chance that PELE calculated for them.

There was basically nothing in Argentina’s record since winning in 2022 that warranted such a bearish view. Their record isn’t as pristine as Spain’s. But it’s very good: in 46 matches since Qatar, they have 39 wins, 4 losses and 3 draws.6 They won Copa America and lapped the field in South American qualifying, where clean sweeps are basically impossible because of the large number of qualifying matches (18), many of which are in challenging road environments like Brazil and Ecuador. Their worst result since the last World Cup was a 2-0 loss to Uruguay at home in qualifying, but that was almost three years ago.

Technically speaking, match results aren’t the only factor in PELE. There’s also a slow process of mean-reversion applied in between matches, based primarily on the Transfermarkt value of the players on the roster and secondarily on team age, region (South America = presumptively good), the size of the country’s economy and its soccer legacy. Argentina does just fine by those measures. While they don’t have the player depth of France or England or Spain, they did rank 7th in the world in aggregate market value heading into the tournament in the way that we calculate it7 (ironically, without giving much credit to Messi8). And Argentina is a big country in what has historically been the most overperforming football region in the world.

Our diagnosis after our extensive work on PELE, making lots of comparisons between team records and player valuations in forecasting future games, is basically that prediction markets place too much emphasis on raw talent and not enough on team results. It’s a bit counterintuitive because international football teams don’t play all that many meaningful matches. We expected to discover something like a 50/50 blend of team results and priors was optimal for predicting future performance, but the “right” ratio is more along the lines of 75/25, with the bulk of the weight going to match results and the priors having a more gentle, long-term effect.

This may be because the idea that a team is just the sum of its player components is dubious. World Cup teams don’t have the same opportunities to optimize their talent that club teams do. While France or England might be better than Spain or Argentina given a few dozen games to refine their tactics, they were outstrategized in the semifinals.

But haven’t Argentina’s results in this World Cup been a little bit sketchy, with Messi constantly pulling rabbits out of his hat?

Read more

Last week, we got some big news about a certain Democratic Senate candidate. No, not that guy. I’m talking about State Senator Mallory McMorrow suspending her campaign in Michigan. McMorrow was a rising star in Democratic politics and was the clear favorite of bettors on Polymarket to win the Michigan primary through late April. To be fair, she did perform well in some early polls of the race. But McMorrow was running with Rep. Haley Stevens to her right and Abdul El-Sayed to her left. Eventually, she was squeezed out of her position in the middle.

Now we’re down to just two active Democratic candidates in Michigan, and unlike some other high-profile Senate primaries, it’s a real moderate vs. progressive contest. Stevens is backed by Chuck Schumer and AIPAC, and was recently endorsed by Gary Peters — the seat’s retiring occupant. Outside groups are also running ads implying an endorsement from former President Obama, but it’s actually a clip from 2018.

El-Sayed, on the other hand, has picked up endorsements from Alexandria Ocasio-Cortez and Bernie Sanders. He’s also highly critical of Israel.1 “For too long, our foreign policy has been handed to us by the likes of the state of Israel and AIPAC, who has made sure that Democrats and Republicans are doing their bidding,” he said in a debate with Stevens on July 7. For his part, El-Sayed has even criticized Obama’s tendency to compromise.

If this dynamic sounds familiar, that’s because we’ve seen it play out across the country in Democratic primaries this cycle. And so far it’s mostly gone the same way. Everywhere from New York City to Colorado to, yes, Maine, progressive outsiders have been trouncing safer establishment picks.2 Is it a Democratic Tea Party? I’ll leave that for another column. But all the signs of this pattern continuing in Michigan are there. McMorrow started tanking in the polls shortly after she criticized El-Sayed for campaigning with left-wing influencer Hasan Piker, whom she called “a provocateur, to put it lightly, who says things that are misogynistic and antisemitic, and said that the United States deserved 9/11.”

Even Stevens took steps to distance herself from Israel’s government in her latest debate with El-Sayed. After Israeli Prime Minister Benjamin Netanyahu criticized her on CNN, she said he has “not made us safer, has not brought us closer to peace. And he’s endangered Jews here in America and around the world.” But is that too little too late to stop populists from racking up another victory? And if El-Sayed does win the primary, could that sink Democrats’ chances of retaining the Michigan seat, and, by extension, of reclaiming the Senate?

Subscribe now

Are progressives cruising to another victory in Michigan?

The basic El-Sayed skeptic’s position is this: Michigan isn’t New York City. You still need to win the general, and El-Sayed’s progressive issue positions are too far left for a state Donald Trump has carried twice. Stevens has made that same case. Aside from run-of-the-mill progressive stances like Medicare for All, El-Sayed supported defunding the police back in 2020. He also voted “Uncommitted” during the 2024 presidential primary as a form of protest of the Biden-Harris campaign.3

One could also cite misgivings that voters might have about “outsider,” lefty candidates after the Graham Platner disaster, but the comparison isn’t fair to El-Sayed. His credentials are impressive, from being a Rhodes Scholar and a star student-athlete to the Health Director of Detroit. He’s also run for office before, having finished a (distant) second in the 2018 gubernatorial primary against Gretchen Whitmer. His biggest scandal is calling himself a physician while never having completed a residency or being licensed to practice medicine. He can be extremely combative, however, and confident to the point of overconfidence — perhaps fittingly, since he hosts his own podcast.

But despite what could easily be considered baggage in a purple state like Michigan, El-Sayed was the clear favorite to win the primary before McMorrow dropped out. Prediction markets saw him as the third-place candidate early in the race, but he’s been on an upward trend since April. Prior to McMorrow’s exit, Polymarket gave him an 85 percent chance to clinch the Democratic nomination.

This was a case where the polls and betting markets aligned. The three most recent Michigan Democratic primary polls in the New York Times database before McMorrow dropped out were El-Sayed +5, El-Sayed +19, and El-Sayed +7. In fact, the last poll prior to McMorrow’s exit where El-Sayed wasn’t leading came back in May. And polling averages of the race only showed El-Sayed’s lead increasing in June.

That’s not entirely surprising. The war in Gaza has become an important issue in Democratic primaries, and Michigan has the largest population of Arab Americans of any state in the U.S., per capita. That group swung away from Democrats in 2024, at least in part due to criticism of the Biden-Harris administration’s approach to Israel. In Dearborn, which has the largest Arab American population share in the country, Trump won 42 percent of the vote in 2024, compared to 37 percent for Harris, and 18 percent for Jill Stein. El-Sayed could bring some of these voters back into the Democratic fold.

But perhaps McMorrow’s departure has given moderates a second chance. What if 80 percent of McMorrow voters flocked to Stevens? The race would get competitive. In fact, that would be enough to put her ahead if you take an average of all pre-dropout polls. But that approach gives a lot of consideration to older polls, conducted before El-Sayed really took off. If you downweight older polls, El-Sayed improves to a 42-40 lead even with just 20 percent of the McMorrow vote.

But post-dropout polls are more promising for Stevens. The first, conducted by Tavern Research, found El-Sayed up by 3 points when all three candidates were included on the ballot, but Stevens ahead by 1 point in the two-candidate matchup.4 Nearly every former McMorrow supporter in that poll moved to Stevens. And this week, a Glengariff/Detroit News poll showed Stevens leading with 48 percent of the vote to El-Sayed’s 41 percent.5

The new polls seem to indicate that the race has tightened. Part of that could come down to Stevens drawing support from more former McMorrow supporters than early polls suggested. Or perhaps the Platner fiasco is boosting moderates’ morale just a bit. Those Obama ads might even matter; Stevens held a large lead among Black voters in the Glengariff poll.

The Peters endorsement, which came after the Glengariff poll, could help, too. Endorsements aren’t usually decisive, but receiving the nod from the two-term incumbent of the seat you’re trying to win is a real “get.”

However, a Data for Progress poll showed almost the complete inverse result: 54 percent for El-Sayed compared to 41 percent for Stevens. But that poll was commissioned by American Priorities, a pro-El-Sayed super PAC. In general elections, candidate-sponsored polls typically exaggerate their candidate’s standing by a margin of about 4 points. We don’t have a comprehensive database of internal polls in primary elections, so we can’t say what the average bias is there, but the Data for Progress poll is showing a divergence from public polls large enough that the firm is putting its credibility on the line.

Would El-Sayed make Michigan competitive in November?

Michigan is a must-win for Democrats if they want to retake the Senate. It’s also an open seat in a year where the national environment should be good for Democrats. True, Trump has won the state twice, but they were narrow victories. This isn’t a scenario where you want or need a high-variance candidate.

Instead, you’d want someone with minimal potential downside who can coast to an easy win on anti-Trump vibes and solid campaigning. Stevens would seem to fit that bill. She’s not an electoral superstar (her Split Ticket WAR scores range from R +0.9 to D +2.8), but she’s not an underperformer either. Being a generic normie Democrat might make things harder in a primary, but it’s a boon in a general where even a neutral performance would likely net you the win. Our forthcoming forecast would also give Stevens some extra credit for held elected office when El-Sayed hasn’t. That’s not to say El-Sayed can’t win, but he’s certainly the riskier choice.

However, Democratic primary voters aren’t just optimizing for electability. While 25 percent of likely Democratic voters in the latest Glengariff/Detroit News poll said the top quality they’re looking for in a nominee is someone who can win in November, 38 percent said it was the nominee being a fighter who would stand up to Trump. If you’re looking for an anti-establishment fighter, El-Sayed is the obvious choice. The bet is that perhaps El-Sayed’s more left-wing views cost him some support relative to Stevens, but his odds of winning are still solid in this Michigan environment, and if he does, Democrats will get a Senator who better aligns with the party’s changing priorities. El-Sayed would also be the only Arab American in the U.S. Senate.

But that’s all very vibes-forward. The data tells a somewhat different story. If you take a simple average of all Michigan general election polls conducted this year, there’s essentially no difference: Stevens’s average margin against presumptive Republican nominee Mike Rogers is D +0.7 compared to El-Sayed’s D +0.2. However, El-Sayed outperformed Stevens in three of the five most recent polls, and they performed equally well against Rogers in a fourth.

On the other hand, one could argue that there is a pollster quality issue here. Michigan pollsters have a mixed track record. The only pollsters in this group with an A/B rating or higher are TIPP Insights, Glengariff Group, and Emerson College. Polls from those firms show Stevens leading Rogers by an average of 2.8 points, but El-Sayed trailing him by 2.5 points.

So … not to rehash the Moderation Wars for the umpteenth time, but this one comes down to your priors. If you think more moderate candidates generally outperform left-wing candidates, you can cite evidence from the higher-quality polls demonstrating that Stevens is the safer choice. If you don’t, you can emphasize recency instead, since the latest polls generally show El-Sayed doing at least as well in November.

There’s also a chance that the electoral environment will be blue enough by November that it won’t matter. Michigan is a truly purple state: Trump carried it by about 1.4 points in 2024 while Biden won it by 2.8 points four years earlier. But as of this post, our generic congressional ballot average is sitting at D +6.1. If that’s what the national environment looks like in November, you’d also expect the Michigan environment to be around D +6.1 according to our state-level benchmarks.

So even if El-Sayed underperforms, he’d have to underperform in a big way to lose Michigan in this environment. That’s not impossible, but just 23 percent of races in the Split Ticket WAR database (which runs from 2014 through 2024) feature a candidate under- or overperforming by more than 6 points.6 And many of those races featured consistently strong incumbents — such as Susan Collins, Jon Tester, and Joe Manchin — or consistent underperformers like Ilhan Omar. An open-seat race doesn’t fit that bill, especially because large over- and underperformances are simply rarer than they used to be.

If El-Sayed wins the primary, the general election is likely to be seen as a referendum on the electoral viability of populist, left-wing, Israel-skeptical candidates. An El-Sayed win in November will be taken by many as a sign that the progressive populist wing of the party is fully in control. But if El-Sayed loses, that same movement will potentially take the blame for costing Democrats the Senate.

Here at Silver Bulletin, we’ve frequently critiqued President Trump, the Democratic “establishment”, “Resistance Libs”, the left, the “tech right,” and the Republican Party. What’s missing from that list? We’ve had less to say about the political tribe that I at least vaguely belong to myself: the center-left.

How does one define the center-left? Well, that isn’t entirely straightforward. As Matt Glassman writes, “moderation” is associated with a cluster of attributes, including but not limited to: centrism, pragmatism, independence, willingness to compromise, obsession with “electability”, comparative optimism about the status quo, liberalism, and a preference for more complex, technocratic approaches toward governance. These aren’t entirely synonymous with one another and some elements can be in tension.

But put a pin in that for now; we’ll come back to it later. Because there’s a more pressing question: is the center-left at risk of losing its influence — even dying out, becoming the Blockbuster Video of politics1? For a cautionary tale, you need to look no further than the center-right, which has been almost completely hollowed out, with most of the public intellectuals who once occupied that space either having become Resistance Libs or joining the Trump Train.

Subscribe now

The death of the center-right didn’t come all at once. In fact, Republicans reacted to the failures of George W. Bush by nominating John McCain in 2008, probably the closest thing to a truly independent-minded presidential nominee in the modern era.2 But having the misfortune of inheriting Bush’s mess amid the worst financial crisis in 75 years, McCain lost to Barack Obama by more than 7 points. Mitt Romney then fended off the Tea Party in 2012, but also lost to Obama.

The repeated failures undoubtedly contributed to Trump’s rise to power. Trump zagged toward the populist, anti-immigrant right when the RNC “autopsy” after 2012 had recommended zigging toward the center — although he also pledged to protect Social Security and other entitlement programs, avoiding some of the most unpopular parts of the Bush-Romney agenda.3 Trump openly criticized all three of Bush, Romney and, of course, McCain. (“He’s not a war hero. He’s a war hero because he was captured. I like people that weren’t captured.”)

The center-left could find itself in a similar predicament. According to statistical measures of their Senate voting records, Joe Biden wasn’t really a centrist in the same way that McCain was. (Biden has always been much more of a loyal partisan, for one thing.) But Biden did win the 2020 nomination, helped by a cluster of endorsements and perhaps a “flight to safety” among voters as COVID hit American shores. Biden’s nomination came against literally more than two dozen alternatives, many of whom mistakenly assumed that Bernie Sanders’s strong performance in 2016 against Hillary Clinton heralded the rise of the left wing.

But maybe those candidates were just premature, not wrong. Although Graham Platner’s implosion was a setback for the left, they’ve been flexing their muscles in races from New York City to Colorado. And according to Gallup polling, the share of voters who describe themselves as “moderate” has fallen from 43 percent in 1992 to 33 percent today. Within the Democratic Party, meanwhile, self-described liberals now outnumber moderates almost 2:1.

These trends come despite the fact that, in the same Gallup polling, far more voters describe themselves as “independent” than ever before, with both major parties being deeply unpopular.

This puts the center-left in an awkward predicament. Voters might want independence from the major parties. But they don’t necessarily want more moderation; in fact, they are more willing to experiment with more radical approaches. However, the parties themselves are as partisan as ever, and social media tends to reinforce those orthodoxies. (So does Substack: Silver Bulletin is the only newsletter in the U.S. politics Top 10 classified as non-partisan by independent researchers.) Excessive partisanship may drive away moderates — but if they exit the primary voter pool, that only makes the remaining voters in the Democratic primary electorate even more progressive and partisan. Outnumbered in the Democratic Party, but having little in common with the MAGA-fied version of the GOP, the center-left is at risk of becoming politically homeless.

Share

The center-left is losing the argument about why Democrats lost in 2024

What’s more, the center-left risks being associated with the failures of the Democratic Party establishment — when, from a center-left standpoint, the problem was really the opposite: that the establishment moved too far to the left. More specifically, that Biden, despite his moderate reputation, was far too deferential to the progressive staffer class and “the groups”, particularly on immigration and on stimulus spending. Meanwhile, Kamala Harris and the rest of the Democratic Party couldn’t escape the shadow of Peak Woke. Remember this?

“Hopium” and motivated reasoning about public opinion don’t help either, as Democrats and the cosmopolitan, coastal news media are prone to mistaking the views of their most vocal opinion leaders and the most enthusiastic primary voters for those of the rest of the electorate. In a series of New York Times polls of key Senate races earlier this month, one that projected Democrats to come up slightly short of the four seats they need to flip the Senate in November, 53 percent of voters said that the Democratic Party was too far to the left, as compared to just 8 percent who said they were too far to the right.4

But Democratic primary voters, like Republican ones, increasingly occupy a separate universe. The Yale Youth Poll recently asked voters to select the more important issue for their vote from a randomly selected pair of 30 issues. Among all voters5, Israel/Palestine only tied for 25th out of 30 issues, winning just 28 percent of head-to-head “matchups”.

But among self-described “extremely liberal” voters, Israel/Palestine won 45 percent of the time. Gay and trans rights also got huge boosts. These differences are likely to be even more profound among the most politically active voters, to the point where views toward Israel have increasingly been a litmus test in certain Democratic primaries.6

Even if the media is overstating the importance of flashpoints like Israel and trans rights, enthusiasm matters a lot, and the center-left’s pet issues — affordable housing, I guess? — don’t tend to appeal to people’s passions in quite the same way. Turnout is overrated in general elections, but very important in primaries. In the primary in New York’s 10th Congressional District, where Brad Lander ousted incumbent Dan Goldman in a race with a heavy emphasis on Israel, just under 84,000 people voted out of a population of 750,000. That’s reasonably high by New York standards, but still only 11 percent of the population.

5 ways to define the center-left

Maybe the center-left can attribute some of this to bad luck. Still, making excuses is never a winning strategy in politics. And part of the problem is that the center-left may face an identity crisis. There are several different frameworks that you might use to categorize it.

First, you can describe the American political landscape in terms of partisanship. This chart of the voting record of Diana DeGette — a mainline progressive Democrat who was recently defeated by a DSA-backed challenger in her Colorado primary — categorizes all members of Congress from liberal to conservative based on their roll-call votes. In this conception, there isn’t really a center-left or anything in the center at all. Instead, the center-left is absorbed within the Democratic Party mainstream.

Actually, I lied. You can find one truly centrist Democrat: Jared Golden of Maine’s 2nd Congressional District:

But guess what? Golden, who has massively overperformed partisan baselines in his district, announced his retirement in November when faced with a primary challenge from Matt Dunlap, the Maine state auditor who is significantly to his left.7 That the Democratic Party is moving to the left doesn’t mean that the overall electorate is. In fact, that’s what you’d expect if centrists exit the coalition and so the coalition shrinks.

If you survey voters rather than looking at voting records in Congress, there is more of a middle — people who are “cross-pressured” between different views. Swing voters aren’t a myth. But their numbers are in steady decline amid ever-increasing partisanship. There’s no particular reason for a voter’s views on say, abortion, Gaza and taxation to be correlated with one another, but in a system with two extremely entrenched parties, they increasingly are.

A third alternative is to view the center-left as one of three major factions within the broader Democratic Party coalition, as I did in my story on Heather Cox Richardson that distinguished three groups:

• The Left

• Resistance Libs

• Abundance Libs

In this conception, the Democratic establishment is largely associated with the Resistance Libs and not the more centrist-leaning, pro-growth Abundance Libs. Even as polls show the Resistance Libs becoming increasingly discontented with their party, they tend to be highly defensive of Harris and Biden and convinced that they didn’t beat Trump because Democrats don’t “fight” hard enough.

But this framing isn’t common in mainstream media coverage of the Democratic Party. Instead, most news outlets treat “liberal” and “left” as synonymous, and many Democrats dubiously characterize candidates like Harris and Gavin Newsom as “centrist”.

A fourth approach is to set the parties aside and look for underlying ideological clusters, as I tried to do in my 2023 story on liberalism. That post defined “liberal” in a more European sense in opposition to the progressive left — not the way the term is usually applied in the United States. In other countries, liberalism is defined by its advocacy for individual rights, the rule of law, and a market-based approach to economics. (Though many liberals prefer a mixed economy and varying degrees of redistribution later on.) In this conception, liberalism is a separate third thing and a distinct ideological tradition that doesn’t fit neatly into the left-right spectrum.

My 2023 post now reads as slightly dated. The term “Social Justice Leftism” was my clumsy euphemism for the woke left — which I argued had some decidedly illiberal characteristics such as only lukewarm support for free speech. Still, there are things I like about this ideological triangle. I like that it recognizes liberalism as a separate and, in fact, philosophically well-grounded view rather than an ad hoc set of compromises and edge cases.

The triangle also helps to explain why both the Democratic establishment and what remains of the non-Trump Republican establishment find themselves in awkward positions. The Democratic establishment is trying to straddle the left and liberal traditions, but is increasingly struggling to hold the coalition together. The center-right, as we’ve said, is in even worse shape.

Finally, you could conceive of the U.S. as a parliamentary system. In Europe and other Western democracies, you can basically identify five major ideological clusters:

• 🔴 Labor/Social Democrats

• 🔵 Conservative/Tory/Christian Democrats

• 🟡 Liberal/Center-Left

• 🟣 Populist/Right-Wing/Nationalist

• 🟢 Green/Socialist/Hard Left

Yes, in the European tradition, the party colors are typically reversed from the convention in the United States, with left-leaning parties being red and right-leaning parties blue. And you can debate the semantics of exactly how this would map onto the United States.8

The important thing is that in this conception, liberals have their own party, rather than being part of Labo(u)r. However, this comes at a cost: they usually aren’t all that many voters’ first choice, especially in countries with strong polarization. In the United Kingdom, for instance, the Liberal Democrats have about 11 percent of the seats in Parliament, and that’s actually a recent high-water mark that resulted in part from the profound unpopularity of Tories and Labour. The real energy these days is often more in the extremes, like Marine Le Pen’s National Rally or Britain’s Reform UK. The liberal parties, however, do tend to play nice with others and often play an important role in helping one of the major parties form a coalition government.

Questions for the center-left

However it’s defined, we’re overdue to ask some questions of the center-left. I have a lot of questions, in fact, because I’m not sure how well things are going. The headline teases it as 12 questions, but it’s actually 52 (!) questions organized into 12 clusters.

In an earlier draft of this article, I paused after each cluster to give my “answers.” But that feels too didactic. You’re welcome to infer my opinions from the way the questions are posed and from what I’ve written previously in the newsletter. (I’ve also included a few editorial comments in footnotes.) But the questions are not intended to be rhetorical: I think these are generally hard issues that the center-left perhaps does not reflect upon enough.9

The questions are roughly sorted by importance, with the first four clusters being the ones I consider to be more existential.

The “Core 4” questions

1. Should the center-left explicitly avow liberalism as opposed to centrism, moderation or compromise? Is it a separate third thing with distinctive characteristics, or a sort of a 70/30 blend of progressivism and conservatism? If the latter, how does the center-left address the critique that it doesn’t really stand for anything?10

2. What is the center-left’s relationship with the Democratic Party “establishment”? Is it part of the establishment — even synonymous with the establishment? Or does it want to keep the Democratic Party at arm’s length? If the center-left seeks to channel its power through the Democratic Party, what risks does that pose as the establishment becomes increasingly unpopular? Either way, how does the center-left convey to voters that it should not be held responsible for the failures of said establishment, particularly its recent choice of presidential candidates?

3. What is the center-left’s strategy for dealing with an increasingly angry electorate? How does the center-left win Democratic primaries when Democrats want candidates who “fight”? Would an “angry center” attract more converts in an era where controversy and grievance often motivate voters? Or should the center-left bet that appealing to people’s more rational, dispassionate sensibilities will come back into fashion?11 How much should it emphasize non-ideological attributes like civility12 and candidate integrity or character? If the center-left is generally more optimistic than other political factions, should it lean into this or de-emphasize this? How does the center-left persuade voters that America has made significant progress when this view is increasingly unpopular and Americans have become less patriotic?

4. What is the center-left’s view on how capitalism is going as we pass the quarter-pole of the 21st Century? How much does the center-left prioritize economic growth versus distributional issues? Does it think government spending should be smaller or larger as a share of GDP? Is an “abundance” agenda neo-liberalism in sheep’s clothing? (And if so, is that a bad thing?) What is the center-left’s approach toward the tech sector and particularly toward AI?

Other important tactical and strategic questions

5. How organized should the center-left be — in terms of raising money, endorsing candidates, and seeking to increase voter turnout for its preferred candidates? Put another way, how much does the center-left want to engage in the “dirty work” of politics? Should there be more or fewer formal structures such as PACs, think tanks and congressional caucuses? Would building such infrastructure necessarily require the center-left to be part of the institutional Democratic Party?

6. What is the center-left’s approach toward poll-driven selection of policies and candidates? Does it too often default toward making arguments on “electability” grounds when it really prefers a given policy on the merits?13 How should the center-left feel about left-wing (or even right-wing) positions that are comparatively popular, or center-left/liberal positions that are unpopular? Even if it’s true that more moderate candidates overperform, how does the center-left win the intraparty argument over electability when many influential Democrats are disinclined to believe this?

7. Should the center-left join the Resistance Libs in regarding Trump as a 5-alarm, existential threat to American liberal democracy — or as something that will pass? Should it advocate for “vote blue no matter who”, because of Trump, as part of some broader “pro-democracy” coalition?14 If the center-left is part of the coalition, is it holding up its side of the bargain when leftist Democrats win primaries?

8. Should the center-left regard the “DSA left” as a threat to liberal values? If it does view the DSA as illiberal, how does it persuade “normie Democrats” about this at a time when the left appears to have a lot of momentum and much of the party is risk-averse and trend-following? Alternatively, are there issues where the center-left can find common ground with the capital-L Left against the mainline Democratic establishment?15

9. What is the center-left’s attitude toward the center-right and other non-MAGA parts of the Republican Party? Would the center-left have more leverage if it more often aligned with or even voted for Republicans? Would it be better off if the center-right were rehabilitated, potentially enabling a broader center coalition? In the abstract, would the center-left be satisfied with a parliamentary-type system in which it’s almost never the largest party but often plays kingmaker?

10. What is the center-left’s strategy for avoiding both the perception and the reality of elitism, especially given that it tends to attract high-earning, successful people? How much should it care about diversity of various kinds? How does it avoid Davos-type groupthink? Does it think the “expert class” is performing well or poorly? Does it think there’s too much or too little democracy? Does it share a Progressive Era vision for technocratic paternalism, or does it tend to defer to popular sentiment and self-determination?16

11. Can the center-left credibly regard itself as more pragmatic and evidence-driven than other political factions? Are they the “reasonable voices in the room”, or is this too self-flattering? Aside from effective governance, how does it demonstrate this to voters in a politics dominated by motivated reasoning and “vibes”? Should the center-left be low-key — or explicitly17 —utilitarian?

12. How does the center-left navigate Israel/Gaza? Can it accommodate people with widely divergent views on Israel and carve out a middle ground at a time when an increasing number of highly politically active Democrats view Israel as a litmus test? How can it avoid the “smalltentism” that defines other Democratic factions, especially when the center-left’s numbers are not all that large?

I suppose I’m hoping to get a debate started about these questions. And yes, that potentially includes here in the comments section. But keep in mind that Silver Bulletin has an extremely politically heterodox readership as compared to most Substacks. So please be kind and give other subscribers the benefit of the doubt.

This iteration of the World Cup has lived up to the hype so far. Say what you want about the 48-team format — which is probably here to stay unless FIFA expands to 64 teams instead. But having an extra knockout round has only raised the suspense for the teams that have survived.

Part of what’s made this World Cup intriguing is that no team has looked utterly dominant thus far. After a pristine group stage, Argentina has looked very beatable in the first two KO rounds — Cape Verde and Egypt were heavy underdogs, but the former nearly stole the game, and the latter gave it away. Meanwhile, betting markets lost faith in Spain after their 0-0 draw in the opener against the same Cape Verde, and Lamine Yamal, recovering from injury, has had a notably quiet tournament.

How about France? They began the World Cup as our 3rd-most likely winner but are now our favorite — though only by a hair over Argentina and Spain. Betting markets see them as a clearer #1. That’s reasonable using market valuations, which puts them highest in the world by our method. They’re the most fun team to watch and the first one you’d pick in a video game.

But in our system, these market valuations serve more as indicators of upside potential or team trajectory than of the quality on the pitch today. France has had some speed bumps since being the runner-up in the 2022 World Cup, like a draw in qualifying against Iceland, a home loss to Italy in the European Nations League, and a friendly loss to Cote d’Ivoire just before the start of this tournament. So far in the World Cup, they’ve played every bit up to that potential, however. Even then, they needed a penalty to beat Paraguay, and will face another very defensive-minded team (though a far more talented one) in Morocco today.

Before we preview the four quarterfinals, here is one more way to look at the odds: literally how the pie (chart) of probabilities for the eight remaining matches is divided up. There’s far more data available on the World Cup landing page, which we’ll continue to update after each game.

Read more