In 2024, I predicted Trump would win the Electoral College and finish with more than 300 electoral votes in my final election prediction on this website. My presidential forecast also called all 50 states correctly, and a few days before the election, I wrote a Trump-bullish blurb on my blog for family and friends that you can read here. To be clear, thinking Trump was favored in 2024, and then Trump winning, is a sample size of one. By itself, it does not prove much of anything about my broader forecasting or prediction ability. However, I do think it is worth pointing out that I am not someone who simply views every election through blue-tinted glasses2.

You can quibble with any individual race rating, but most of the forecast ultimately comes back to one central assumption: we are projecting a D+7.3 national environment on Election Day, which is quite blue3. I want to zoom out and talk about why we are predicting that in a macro sense.

What Democrats Have Going for Them

1. There is a lot of historical precedent working in Democrats’ favor. If Democrats win the U.S. House in 2026, it would continue a streak dating back to 2006 in which the president’s party has failed to win control of the U.S. House in a midterm election. The last exception was 2002, when Republicans retained the House in the post-9/11 political environment, with President Bush still holding elevated approval ratings after the attack. Before that, the last time the president’s party held the House through a midterm was 1978, under Jimmy Carter. People should not be surprised that the forecast expects Democrats to do well this November. There is a long historical pattern on their side, and that pattern is built into the model. Republicans keeping control would not be impossible, but it would be a genuinely notable historical event.

2. Trump’s approval rating is really poor. He is currently at -17 in VoteHub’s tracker, which is roughly six points worse than where both Biden and Trump were at this same point in 2022 and 2018. Of course, if he claws his way back to something like -12, the midterm picture would look meaningfully different, and there is obviously some chance that happens. But as of now, Republicans are heading into the cycle with an incumbent president who is not just unpopular, but unusually unpopular. It is also worth remembering that some of the same economic issues that hurt Biden during the COVID recovery are now hurting Trump. In Silver Bulletin’s issue approval tracker, Trump is at -42 on inflation, which is brutal, and probably a major reason he is struggling so much with working-class voters in the polling crosstabs. The sitting president’s party usually does poorly in midterms, and Republicans are facing that baseline problem with an incumbent president whose approval rating is historically weak.

3. Democrats are generally well-suited as a party to perform well in midterm elections. Midterm electorates are a subset of presidential electorates, and the voters who drop off are likely disproportionately helpful to Republicans right now. I’ll lean on two useful points from David Shor and Blue Rose Research. First, Democrats now appear to have the higher-propensity coalition, since voters who care more about politics tend to be more liberal in polling. Second, if only 2022 midterm voters had participated in the 2024 presidential election, Harris would have performed about two points better. Many of the polls we are getting are still “RV” polls, meaning they are weighted to registered voters rather than a likely midterm electorate. We have not yet seen this newer high-propensity Democratic coalition tested in a midterm with a Republican president, but it is not hard to see how the combination of Democrats’ generally higher-propensity electorate and the usual turnout and enthusiasm boost for the out party could create some very rough turnout dynamics for Republicans this fall.

So, to summarize, this is not a great situation for Republicans. Midterms are historically difficult for the party that controls the White House, Trump is extremely unpopular with the American public, and he is probably even more unpopular with the roughly 32% of Americans who will actually vote in the midterms this November than he is with the country as a whole. But what mitigating factors could help Republicans avoid a blue wave?

What Republicans Have Going for Them

1. The Senate map is really, really tough for Democrats, and the fact that we are even talking seriously about Senate control is a sign of how bad the broader environment is for Republicans. Even if you assume Democrats win North Carolina, Georgia, Michigan, and Maine, which is far from easy, the path is still brutal. They would need to hold North Carolina, a Trump-won state; hold Georgia, a narrowly Trump-won state where Democrats have major candidate-quality advantages; win Michigan, another Trump-won state with real question marks on the Democratic side; and defeat Susan Collins in Maine, a durable incumbent who has repeatedly survived difficult campaigns before. And even after all of that, Democrats would still most likely need to win two of Iowa, Texas, Ohio, and Alaska, four states Trump carried by double digits. Right now, we are projecting roughly a 9.4-point national swing toward Democrats. Trump’s net approval rating on Election Day in the 2018 midterms was -10, and at that number it seems like Democrats would have almost no chance of flipping Senate control in 2026. The good news for Democrats is that he has a long way to go to get back to -10.

2. Another important factor is that differential turnout is usually less extreme in the most competitive races, which helps whichever party is having a bad year. We saw this pretty clearly in 2024, and I expect we will see something similar in 2026. This will help Republicans quite a bit in the key Senate battlegrounds, where higher turnout and more attention will likely normalize some of the national wave effects. Republican turnout in places like Texas and Iowa is likely to hold up much better than it will in less competitive states like New York and California.

3. Democrats also have their own problems, and this comes through in the fact that their lead in generic ballot polling is not especially impressive given how unpopular Trump is right now. Honestly, knowing Democrats are (1) the out party, are facing (2) a president at -17 net approval, and have (3) a coalition that is unusually well-suited for midterms, you would probably expect them to be clearly favored to flip the Senate in a tsunami-like environment. But they are not. The fact that Democrats are only leading our generic ballot polling average by 6 points right now with a historically unpopular Trump does not reflect especially well on the party’s broader position with the American people right now.

4. The Iran situation seems like it could quiet down in the near future, and it looks like the Strait of Hormuz could reopen soon. This could help Trump’s approval a lot. At this point, Trump’s best path to a stronger midterm environment is probably to sit on his hands for the next six months: no major ICE escalations, no foreign policy chaos, and no self-inflicted distractions.

5. Republicans could gain a few U.S. House seats before the midterms with redrawn congressional maps after Section 2 of the Voting Rights Act was narrowed last week. In a best-case scenario for the GOP, those redistricting gains could nearly double their chances of winning the House in our forecast.
   

The largest declines in net international migration have occurred in Democratic-leaning urban areas—particularly in California, Chicago, and the Northeast, but also in parts of Texas, Colorado, and Florida. Given that the Census has already signaled that 2030 U.S. House reapportionment may already be unfavorable to Democratic states, a continuation of this trend—where domestic outmigration from blue states is not offset by international inflows—could further exacerbate those losses and frankly make the 2032 Electoral Map pretty scary for Democrats.

Trump’s administration has implemented a notably aggressive immigration policy, and I estimate that 31% of U.S. counties have experienced negative net international migration. This implies that total outflows—driven by both deportations and voluntary departures (including what is referred to by the Trump administration as “self-deportation”)—have exceeded total inflows, whether legal or illegal. The 15 counties with the largest net outflows are concentrated in predominantly Hispanic communities with high shares of unauthorized immigrants—along the border in California, Texas, Arizona, and New Mexico, as well as in California’s Central Valley.

Data and Methods

The U.S. Census Bureau’s yearly population estimates track net international migration as a single component of population change by county, capturing the net effect of all inflows and outflows regardless of pathway. As a result, shifts in these estimates provide the clearest available signal of how immigration is evolving at a national and local level.

Interpreting that shift, however, requires some care. The Census Bureau measures population change on a July-to-July basis, meaning that the Vintage 2025 estimates reflect a period that only partially overlaps with the current administration. In fact, those estimates include several months that precede both Trump’s 2024 election and his 2025 inauguration.

To avoid understating recent changes, I adjusted for this timing by assuming that migration flows in the three months prior to the election followed the average pace observed in 2023 and 2024, which were broadly similar years. From there, I worked backward to estimate the implied monthly migration rates for the nine-month remainder of the period.

Even under these deliberately conservative assumptions, the results point to a clear deceleration in net international migration. The slowdown appears both widespread and substantial, suggesting that the trajectory of U.S. population growth is already beginning to shift. If anything, this approach likely understates the magnitude of the change. It attributes some pre-inauguration migration to the current policy environment and assumes that any shifts occurred immediately, when in practice such effects often take time to fully materialize.

The more definitive test will come with the next data release. The Vintage 2026 estimates, expected to be released in early 2027, will be the first to capture a full year under current conditions, covering the period from July 1, 2025 to July 1, 2026. I expect a continued sharp decline in net international migration, a view that aligns with U.S. Census Bureau projections.

For now, the early evidence indicates Trump is, in fact, successfully limiting migration. Preliminary estimates suggest that net international migration has fallen sharply, to levels well below those observed in recent years. While it remains too early to draw definitive conclusions about what will happen over the remainder of Trump’s term, the initial data indicate that policy changes are already having a measurable impact on overall migration flows.

That pattern is often explained away as a byproduct of geography rather than political factors. Warmer climates, lower density, and suburban or rural lifestyles have long attracted movers, particularly in the Sun Belt and Mountain West. On its face, that explanation is intuitive, but it is incomplete.

Using county-level data, I examined domestic population change while accounting for several of the most commonly cited drivers of migration. I incorporated local climate by collecting average January high temperatures for every county, controlled for development patterns using population density (measured as median weighted census block group density), and accounted for demographic factors. These adjustments capture much of what we would expect to explain migration trends: people tend to move toward warmer, less dense areas, and population change varies meaningfully across demographic compositions.

Even after accounting for climate, density, and demographics, a substantial gap remains. Figure 1 shows the model-implied difference: counties in Trump-won states were still predicted to experience roughly 2.7 percentage points more domestic population growth over the past five years than otherwise similar counties in Trump-lost states. That does not, by itself, establish that partisan governance is the cause. But it does suggest that the divergence is not simply a story about warm weather, suburbanization, or demographic composition more broadly.

There is no single agreed-upon explanation. Instead, the debate tends to cluster around a few competing theories.

One argument, often advanced by center-left housing-focused policy analysts, centers on housing constraints. In many Democratic-leaning states, particularly in large metropolitan areas, housing construction has lagged demand for years. Restrictive zoning and local opposition to development have limited new supply, making sustained population growth more difficult. In that framework, out-migration is less a matter of preference than of affordability.

A second explanation, often offered by conservatives, focuses on public safety. Some observers point to differences in crime rates as a potential driver of migration decisions. While there are measurable differences across states, those gaps are more modest than often perceived once demographic and economic factors are taken into account. Even so, perceptions of safety can influence behavior independently of the underlying statistics, and may play a role at the margin.

A third line of argument, also offered by conservatives, emphasizes fiscal policy. Higher taxes and cost of living in certain states are frequently cited as reasons individuals and businesses relocate elsewhere. Whether these factors are decisive or simply contributory remains an open question, but they are consistently part of the broader narrative.

It is difficult to disentangle these mechanisms cleanly, and I do not want to attempt to assign causal weight to any single explanation because it’s not my expertise. What it does establish, however, is that the divergence in population growth between Democratic- and Republican-leaning states persists even after accounting for many of the most obvious structural differences between them.

Appendix: Full Regression Results

So why should you listen to my bracket advice? Over the last three years, I’ve made about $5,000 in net profit on bracket pools. I typically enter a handful of pools and spend around $300–$400 total each year on entry fees, so it’s been a pretty good run. Some of that is definitely luck. Even if my models were perfectly calibrated, I’d only expect to earn around 3–4x my entry fees in a typical year in revenue. Over a three-year period, that would usually come out closer to $2,500 in profit, not $5,000. So the recent results mean I’m on a bit of a heater.

But the underlying idea is still sound. If you follow the strategy in this guide, you should have a real edge in most bracket pools. In smaller pools, that edge is modest but fairly consistent. In larger pools, the strategy gets more aggressive—you’ll win less often, but your expected winnings edge is greater, as you can differentiate effectively through high-variance strategies.

Table of Contents:
Section 1: Math Thoughts
Section 2: Basketball Thoughts
Section 3: Algorithm Ranking and Ratings

Section 1: Math Thoughts

Key fact: The bigger your pool is, the more risk you should take.

This section applies to brackets every year and is mostly copied and pasted from last year.

First, understand the math of brackets. To maximize your chances of winning your pool, you need to do two things: pick good teams that are likely to win and differentiate yourself from the rest of your pool (especially in the later rounds). Differentiation matters a bit in earlier rounds too, but it’s usually smarter to pick “chalk” (the favorites). The goal is to pick up points where others lose them. You’re better off finishing first or last than somewhere in the middle.

Next, consider the size of your pool. The larger the pool, the more risks you should take to differentiate yourself. In small pools (fewer than 25 people), it’s often best to pick mostly chalk, perhaps deviating slightly with your title pick. In mid-sized pools (around 100), you’ll still want to stick with a lot of chalk, with maybe a single deviant pick in the mid or later rounds. Once you get above 100, having some lower-seeded teams in the Final Four starts to make more sense.

The pool’s scoring system is also key. In seed-bonus pools, where you get a bonus for picking lower-seeded teams to win, take advantage of that by picking strong double-digit seeds in the first two rounds. However, you should still pick a top seed to win the title. In flat-scoring pools (where the point structure is something like 1-2-3-4-5-6 instead of 1-2-4-8-16-32), picking chalk in the later rounds is crucial since differentiation is less important. Focus on accuracy, with a little differentiation in the early rounds.

(The rest of the general advice is for people who take this more seriously—skip to the next section if you’re getting bored.)

How “sharp” your pool is also matters. If you’re in a casual pool with friends, family, or work colleagues, it’s fine to pick trendy teams, such as teams with strong traditional (KenPom) analytics profiles for their seed. But if you’re in a sharp pool with more serious players who will be looking at analytics and think they know basketball, you want to turn the tables on them and avoid those trendy teams.

My competitive edge comes from looking beyond surface-level “efficiency” analytics to uncover the deeper stories behind why teams may outperform or underperform their statistical profiles. The #1 most important consideration is to view college basketball as a small-sample exercise in a Bayesian world. We start with priors—such as the preseason AP poll and KenPom rankings—and update them based on in-season results. Many analysts mistakenly rely 100% on the data (season results), ignoring the importance of incorporating priors into their posterior beliefs about team strength. That approach throws away valuable information.

This concept is fundamental to how you should view NCAA brackets. It might seem like one of those spurious correlations you should ignore, but it’s actually rooted in a sharp Bayesian understanding of college basketball. The streak will eventually end, and I suspect it matters less than ever anyway. In the NIL and transfer portal era, preseason expectations carry much more uncertainty. Still, it’s an important concept to understand.

This also extends to conference strength. Teams from conferences that were stronger than expected during the regular season (which isn’t really the case for any conference this year) tend to underperform in March, while teams from conferences that were weaker than expected (like the Big East) tend to overperform. The reason? Cross-conference results early in the season provide only a small sample of data, and failing to account for the importance of the prior can lead to misjudging team strength.

Other key factors I consider:

• Health: Are there teams that struggled with injuries but are finally healthy?

• Experience vs. Talent: Have highly talented but less experienced teams improved significantly over the past two months?

• Fake Team Strength: Did some teams inflate their efficiency metrics by dominating weak opponents while struggling against tougher competition?

• Key Absences: Are there teams missing a key player due to injury who won’t return?

• Rotational Changes: Have lineup adjustments significantly altered a team’s performance?

• Matchups: Does a team have an especially favorable or unfavorable draw?

• Opponent Shooting Variance: Have opponents hit an unsustainably high or low percentage of jump shots or free throws?

• Team Shooting Variance: Are there players with a proven track record struggling this year but improving late?

Finally, for the tiebreaker score, a good rule of thumb is to enter 144 (e.g., 76–68), which is considered the optimal total. If your title pick is a high-scoring team like Michigan, Arizona, or Florida, adjust the total upward. For lower-scoring teams like Duke, Iowa State, or Houston, adjust it downward. The tiebreaker matters, especially in large pools. A massive proportion of winning outcomes will come down to tiebreakers. Don’t overlook it.

BASKETBALL SECTION

The 1 seeds are really, really good this year (especially Duke, Michigan, and Arizona). Unfortunately, the NIL era has led to more dominance at the top and fewer upsets in March. That doesn’t mean this year will be as boring as last year, where basically all the favorites won, but it does mean the underlying analytics for the top seeds are the strongest I’ve ever seen relative to the field.

Title Favorites

Duke (1 seed, dislike)
Duke is widely regarded as the best team in the country, and they had a great year. They are very good and are led by the best player in the country, Cameron Boozer.

Losing Caleb Foster to a broken foot against UNC on March 7 will dominate my analysis of this team. Foster should be out until at least the Final Four, and potentially for the entire season. Without Foster, they not only lose his 25 MPG and 9 PPG, but also the floor spacing he provides. Duke only had eight good players (with Foster), and his loss will probably force them deeper into their bench and into playing Darren Harris more, who isn’t very good.

This Duke team’s biggest (arguably only) weakness is three-point shooting, and without Foster they’re going to have to rely on guys like Dame Sarr and Nikolas Khamenia to make shots. Stud center Patrick Ngongba also sat out the last few games of the year and has been ruled out for the first NCAA Tournament game, and although the injury isn’t expected to be serious, it is still concerning.

Duke is really good, and arguably the favorite to win the title in my head, but the issue is how many people are picking them. It’s hard to win pools when you’re picking a team that a huge portion of the field is also picking that isn’t even clearly the favorite. I think the Foster injury is a pretty big issue.

Michigan (1 seed, like)
Michigan is also really good and dealing with an injury of their own. Backup guard LJ Cason tore his ACL against Illinois. However, I’m less concerned about this injury for two reasons: (1) Cason’s injury leaves Michigan with eight strong players (compared to seven for Duke without Foster), and (2) Cason is more of a redundant piece than Foster.

Michigan does now only have a single lead guard in Elliot Cadeau, but he can play ~30 MPG, and the team has been fine without either Cadeau or Cason on the court this year. Trey McKenney, Roddy Gayle, and Nimari Burnett can all handle the ball some, and Yaxel Lendeborg is an elite passer.

Another thing that works in Michigan’s favor is how good their draw is. Their 5 seed Texas Tech lost JT Toppin for the year to a torn ACL, and their 4 seed Alabama had their second-best player Aden Holloway arrested and charged with a felony for being a drug dealer on Monday (oops).

Michigan’s analytics were, if anything, actually slightly stronger than Duke’s this year. And I believe the Cason injury is less of a big deal than the Foster injury. They did close the season really poorly, which makes you think the Cason loss might be a bit of an issue. Still, the public is really out on them. Michigan is my favorite national champion pick this year in small pools.

Arizona (1 seed, like)
Arizona is a really good team with only two losses against a tough schedule this year (and one of those came with star Koa Peat injured, which is nice). They don’t have the most talented roster ever, but they have good players and are very strong defensively across the board.

This team is very, very good, and they are healthy—unlike Michigan and Duke. A lot of people are picking them to win the title, but I still think picking them to win is a very reasonable play, especially in small pools.

Florida (1 seed, like)
Florida is a bit of a conundrum. They started the season poorly, then had a very strong second half, before getting completely ripped to shreds by Vanderbilt in the SEC Tournament.

The defending champs return their entire frontcourt from last year, and it’s elite. They also spent a lot of money on elite guards Xavian Lee and Boogie Fland, who are solid all-around players but seem to have forgotten how to shoot the basketball this year. Lee has been a pretty big disappointment but has been playing better as the year has gone on, which you could interpret as adjusting to a higher level of basketball.

If Florida makes shots, they can win the title. The public really hates them, which is surprising given they are the defending champs and a 1 seed. Either way, they are undervalued. I think it shows how overvalued Duke is that I think all three of Michigan, Arizona, and Florida are good title picks.

Houston (2 seed, like)
Houston is pretty much just as good as a 1 seed Florida, especially after considering the added bonus of getting to play their Sweet 16 game (potentially against Illinois) and their Elite 8 game (potentially against Florida) at home in Houston. This is pretty ridiculous, and the NCAA should be embarrassed about it. However, the public is really really in on Houston this year and people are picking them a ton to go to the Final Four. Illinois is the value play to the Final 4 in larger pools in this region, while all three of Houston, Florida, and Illinois are being underrated in their chances to make the title game.

Other Teams

Iowa State / Illinois (2/3 seeds, like)
ISU and Illinois are the only truly “viable” title picks in pools under 500 people outside of the top 5. Purdue is also roughly as good as these teams (deep down I think they’re better), but a lot of people are picking them.

ISU and Illinois both have good early-round draws (Illinois gets UNC in the second round, who is very weak, while ISU’s 3 seed Virginia is very weak), are very solid teams (probably the 6th and 7th best teams in the country), and are being underpicked by the public.

If anything, I’m probably lower on both teams’ chances to win than basketball sharps are (especially Illinois, whose “matchup-hunting” offense has allowed them to post some really gaudy efficiency ratings but has also made them perform worse against better teams relative to weaker teams for three years in a row now—I don’t think that’s a coincidence). Illinois’s best path to a title is to take care of business against weaker teams and hope for chaos. But people filling out brackets just aren’t picking ISU and Illinois as much as they should, meaning they are title-winning options in 150+ person pools and can go to the Final Four/Title Game in ~100 person pools.

Purdue (2 seed, meh)
I wrote a section saying I really liked Purdue before they won the Big 10 tournament last weekend and got hyped up a lot. I still think they are the sixth best team in the country, but they aren’t as good of a value as Illinois and Iowa State.

Purdue has huge depth problems—they have five great players, but Omar Mayer and Daniel Jacobsen have been huge disappointments. I actually view this as a positive in the tournament, where teams tighten rotations. Just like UNC in 2022, I think Purdue will essentially try to play five guys the entire tournament.

Gonzaga (3 seed, ???)
It looks like Braden Huff, probably Gonzaga’s second-best player, is going to miss the first two games, but then has a pretty good chance of returning for the second weekend.

This creates a weird dynamic. They are much weaker early but significantly stronger later. So I both feel the urge to pick against Gonzaga early (if I had to pick one double-digit seed to the Sweet 16 in the entire field, it would be Texas over BYU, then Gonzaga) and also play them to the Final Four or title in very large pools where I want to differentiate with a high-variance strategy—especially against “sharp” players who rely heavily on traditional analytics that aren’t especially fond of Gonzaga.

They are also the worst early round survivor pick imaginable because they get stronger as the tournament goes on.

Wisconsin (5 seed, kinda like)
Wisconsin is the hottest team in the country, oddly without Nolan Winter (their star center) for the last four games. He should be coming back, and I’m optimistic they can keep playing well with him integrated back in.

However, the public is picking them a decent amount, and their draw is tough (Arkansas as their 4 and Arizona as their 1). I’m not especially optimistic about the value in picking Wisconsin deep. I like them, but Illinois, ISU, Vanderbilt, and even UCLA are better options for deeper runs.

Vanderbilt (5 seed, love)
Vanderbilt has the best chance of any 4 or 5 seed to win the title. They have high-end talent and are playing very well recently.

They also have the weakest 1 seed (Florida) in their region, and Nebraska is a weak 4 seed. They can be taken to the Elite Eight in large pools (150+) and even to win the title in massive pools (500+). My favorite true sleeper.

Louisville (6 seed, hate)
Louisville is interesting. They have a very talented roster that has underperformed their ceiling.

They get back Mikel Brown, an NBA-level guard who has missed 12 of their last 23 games. Before his injury, they looked like a 2–4 seed team. I also think Ryan Conwell is due for positive shooting regression and will benefit from Brown taking pressure off him.

However, their traditional analytics profile is inflated by dominating weak competition while struggling against strong teams, which is something I really don’t like to see.

North Carolina, Texas Tech, BYU, Alabama (hate)
The the most important thing to understand about this tournament is what happened to Caleb Wilson (UNC), JT Toppin (Texas Tech), and Richie Saunders (BYU). All three teams lost top-30 caliber players to season-ending injuries.

And then Alabama lost Aden Holloway after he was arrested and dismissed from the team.

All four teams are meaningfully worse than their seed suggests. Alabama vs. Texas Tech, in particular, has turned into a total mess that I’m pretty indifferent toward.

UCLA (7 seed, love)
UCLA has been really good recently and has a great path to the Elite Eight. I don’t think very highly of either MSU or UConn in their region.

Ohio State (9 seed, like)
I really like this team. They’ve improved as the season has gone on and are very talented.

Like Purdue, they have major depth issues, which I actually think helps them in a tournament setting as rotations tighten. Another reason I’m lower on Duke is that I think Ohio State is very, very good.

Specific Earlier Round Picks

2nd Round:
Texas over BYU is my favorite upset. Utah State, Ohio State, Georgia, UCLA, and St. Mary’s are must-picks in my book. Clemson/Iowa I don’t have a strong stance on.

If you want to be a little spicy, I like VCU, Mizzou, and Santa Clara, though they are still underdogs. If you want to be really spicy, I like Northern Iowa and Hofstra (~20–25% win probability).

A lot of people are picking South Florida over Louisville and Akron over Texas Tech, which I actually respect—these are exactly the types of 11/12 games where the underdog wins more often than people expect.

Sweet 16:
It’s a good year for slightly spicy picks in the second round. I do like every 5 seed over every 4 seed in larger pools, but in smaller pools I think you want Arkansas over Wisconsin.

I like UCLA and Tennessee as upset picks in the second round, but mostly in larger pools since they are clear underdogs.

Elite 8:
Not a great year for spicy Elite Eight picks. I like MSU over UConn, but other than that I do really think picking every 1 and 2 seed to the Elite Eight is a must in any pool under 100 people.

In larger pools, I like Illinois and St. John’s (both can make the Final Four). In very large pools (250+), Vanderbilt, UCLA, Tennessee, and Wisconsin are also strong high-variance options.

ALGORITHM RANKING AND RATINGS

Tournament Odds

Team Ratings

I’m well aware that state-level data is not always the cleanest unit of analysis, and I would not present a plot like this and imply it’s evidence of an individual-level relationship without more granular data to support it. That said, with the release of the new 2025 U.S Census Bureau population estimates this week, it’s worth noting that fertility rates in Republican-leaning states are substantially higher than in Democratic-leaning states. The relationship is pretty strong. The correlation between Trump vote share and fertility at the state level is now 0.84, up from 0.79 last year, which is unusually high for demographic data of this kind.

At a high level, the gap reflects systematic differences in family formation between conservatives and liberals. Conservative women end up having more children than liberal women, but why? Upstream, they report wanting more children, are more likely to marry, and conditional on marrying are even more likely to marry at younger ages. Since married women have significantly higher fertility than unmarried women, and conservative women are more likely to be married at nearly every age, these patterns compound over time.

The underlying drivers of these trends and preferences to me are not especially mysterious. Much of this appears downstream of the long-running divide between feminist norms and family-centered values. Liberal women, on average, are now very, very likely to embrace the idea that fulfillment should be defined independently of their “traditional” role of marriage and motherhood, a view that has become increasingly dominant in the last half-century. It is therefore not surprising that young liberal women, having internalized these norms, place less value on marriage and childbearing, marry at lower rates, and ultimately have fewer children. This outcome seems like a fairly natural consequence to me of stated preferences and social incentives. There is good survey evidence supporting these dynamics, including some recent polling work from NBC and this cool poll on dating from Daniel Cox.

But you say, “This partisan fertility gap is not new. Feminism and its downstream cultural effects have been shaping family behavior for decades, and the divergence between liberal and conservative fertility has been visible for roughly 30 years. So why does this dynamic feel as though it is coming to a head now? Couldn’t you have written the same piece ten years ago?”

Good question. The key change this year that really is blowing my mind isn’t really related to anything I discussed above. For many years, higher fertility among Black women helped offset lower fertility among White liberals, which mattered electorally because Democrats have historically won overwhelming margins among Black women across ideological lines. That buffer is now eroding rapidly as Black fertility rates declined sharply in 2025 while White fertility rate actually increased. This data is preliminary and subject to revision, but if it’s real and this decline persists, it poses a meaningful problem for Democrats. For much of the past several decades, generational churn has been relatively favorable to Democrats: conservative fertility exceeded liberal fertility, but Millennials were overwhelmingly liberal, and higher Black fertility rates helped offset the partisan birth gap.

That equilibrium appears to be shifting. Gen Z is currently far more conservative than Millennials, though that could certainly change. At the same time, Black fertility rates are declining relative to White fertility rates. If those two trends continue in parallel, the demographic dynamics that previously cushioned Democrats against lower overall fertility among White liberals may no longer operate in the same way.

Taken together, this creates a genuine long-run challenge for Democrats. Children tend to inherit their parents’ partisan leanings at meaningful rates, and Republicans are having more children than Democrats. And that gap is widening, not narrowing. Sustaining a competitive political coalition becomes more difficult when one side is, quite literally, not replacing its voters over time.

I didn’t expect those estimates to become a point of contention in a high-profile redistricting case, but they have. In the NY-11 litigation, my demographic estimates were introduced by Dr. Stephen Voss as a validation check about what the district’s political coalitions actually look like. The Petitioners’ Summation Memorandum attempts to discredit my estimates and peddles an alternative analysis by Boston University Professor Maxwell Palmer, but unfortunately does not engage substantively with my ecological inference framework or my claim that the independence assumption is strongly violated in many contexts. It does, however, take a brief detour to note my age, complete with a citation to my LinkedIn profile.

This post is my attempt to explain the key empirical dispute as clearly as possible: what NY-11 looks like politically, why traditional ecological inference methods struggle in places like Staten Island, and why I’m confident my estimates better reflect reality than traditional ecological inference models, including the one Professor Palmer employs.

(This is not meant to be a full methods piece. If you’re interested in a more technical explanation of how I produced my estimates and why I built the model the way I did, I’ve outlined the methodology publicly here: [link].)

NY-11 Background

NY-11 is a heavily Republican district centered on Staten Island. In 2024, it was Trump +25, and the district is about 54% White. Politically, the geography is also straightforward: Southern Staten Island is extremely Republican, northern Staten Island is more diverse and votes more Democratic, and the portion of Brooklyn included in the district (Bay Ridge, Dyker Heights, and Bensonhurst) is mixed.

Plaintiffs argue that NY-11 should be treated as a VRA-protected coalition district, and that the current lines disenfranchise minority voters by preventing them from forming an effective coalition. The requested remedy is a redraw that would likely flip control of the district from red to blue.

In the Petitioners’ Summation Memorandum, plaintiffs attempt to defend a specific empirical claim based on Prof. Palmer’s Report: that Hispanic and Asian voters in NY-11 form a strongly Democratic bloc that can anchor a coalition district. I do not believe this claim is supported by the data.

The Core Issue: Hispanic and Asian Voters in NY-11 Are Not a Blue Bloc

In my estimates, Hispanic and Asian voters in NY-11 are close to evenly divided, roughly splitting 50–50 between Trump and Harris in 2024.

That should not be surprising given broader national trends. Hispanic and Asian voters have shifted noticeably toward Republicans over the past eight years, and in 2024 they were not overwhelmingly Democratic groups nationally. NY-11 is also a Trump +25 district, so it would be unusual for the district’s non-Black minority groups to vote like they do in a place like San Francisco.

My topline estimates for NY-11 are:

• White, Hispanic, and Asian voters were collectively around Trump +34

• Those groups made up roughly 91% of the electorate

Where the Dispute Comes From

The opposing analysis comes from a report using traditional ecological inference methods by Prof. Maxwell Palmer. I respect a lot of Prof. Palmer’s work, but I strongly disagree with his conclusions in this case.

His report implies that 88.1% of Hispanic voters in NY-11 supported Kamala Harris in 2024. That number is hard to reconcile with both national benchmarks and basic facts of American political geography.

To put the number in context:

• VoteHub’s national estimate for Hispanic Democratic support in 2024: 55%

• Catalist’s national estimate for Hispanic Democratic support in 2024: 54%

Prof. Palmer finds that Hispanic voters in a very conservative House district were 30+ percent more Democratic than Hispanics nationally.

That gap should immediately raise skepticism, because it conflicts with a simple and well-established fact about U.S. elections: Hispanic partisanship, when measured at the precinct level, is strongly related to the surrounding local political environment. In other words, Hispanic voters in very Republican areas tend to be less Democratic than Hispanic voters in very Democratic areas, and vice versa.

So the claim is not merely that NY-11 Hispanics were a bit more Democratic than average. The claim is that they were dramatically more Democratic than national benchmarks in a district whose overall electorate is deeply Republican. It would place NY-11’s Hispanic vote among the most Democratic Hispanic electorates anywhere in the country, despite NY-11’s broader voting behavior pointing strongly in the opposite direction.

The Missing Piece: Aggregation Bias

The most important statistical issue here with Prof. Palmer’s analysis is the presence of aggregation bias, which occurs in ecological Inference when relationships inferred from aggregated data do not hold across different local contexts.

Traditional ecological inference often relies on assumptions that are violated when a demographic group’s political behavior varies sharply depending on the type of precinct they live in (known in the literature as the Independence Assumption, as described here in Section 2). Staten Island is a textbook example of that problem.

In NY-11:

• Hispanic voters in Northern Staten Island are often located in more diverse precincts that also have substantial Black populations and vote strongly Democratic overall

• Hispanic voters in Southern Staten Island are located in precincts that are overwhelmingly White and overwhelmingly Republican

We see this phenomenon across the country in voter registration data, such as here. Those two populations are very different. We should not expect that Hispanic voters living in very red suburban areas to vote the same way as those living in very diverse and blue urban cores.

If an ecological inference method extrapolates too heavily from the most Hispanic precincts in the district, it can produce exactly the kind of distortion seen here: a Democratic estimate for Hispanics that fits northern Staten Island fairly well, but becomes implausible when extended across the full borough.

What About Confidence Intervals?

A common response I’ve gotten in the past is that ecological inference models can generate confidence intervals and uncertainty bounds. In fact, Prof. Palmer’s lower bound in his CIs for Hispanic support of Kamala Harris in NY-11 is 81.1%.

That is a good feature in principle, but confidence intervals are only meaningful when the model is correctly specified. Unfortunately, even the low end of the CI bounds from Prof. Palmer’s analysis are implausible. When a method depends on an assumption that is violated, the confidence intervals do not represent valid uncertainty—they represent uncertainty around a distorted estimate. In other words, confidence intervals cannot compensate for systematic bias introduced by model misspecification.

The issue in NY-11 is not that we are missing precision. It is that the core assumptions used to infer group behavior in traditional ecological inference are not well-suited for the political geography of the district.

A Less Technical Summary

The simplest way to describe the independence problem in NY-11 is this:

Hispanic voters in very White, very Republican precincts vote differently than Hispanic voters in more Hispanic, more Black, heavily Democratic precincts.

Any approach that assumes those voters can be pooled into a single district-wide relationship will struggle. And that struggle will systematically pull estimates toward the behavior of the precincts that provide the clearest signal, which tend to be the most Hispanic precincts (which are blue in NY-11).

Ways to Evaluate the Competing Analyses

One reason this disagreement is difficult to resolve publicly is that I am currently the only analyst releasing national precinct-level demographic estimates. In practice, this is because extending traditional ecological inference methods to congressional districts and then scaling them nationally, as in Prof. Palmer’s approach, can produce results that are difficult to defend when examined across diverse political and geographic contexts.

Having previously applied traditional ecological inference methods to both election results and partisan registration data in regions like NY-11 and across the country, and having validated those results against voter file data, I have repeatedly encountered the same problems present here. That experience ultimately motivated me to develop a different approach, one specifically designed to remain reliable when the independence assumption is violated.

If precinct-level Hispanic vote estimates from Professor Palmer’s analysis were made available for NY-11, particularly across southern Staten Island, it would be even easier to evaluate the results transparently.

On VoteHub’s Model and Transparency

VoteHub’s model itself is not admissible in court because it is proprietary. Full disclosure would require the release of implementation details that the company is not prepared to make public, as the model is considered commercially valuable.

More generally, it is notable that traditional ecological inference methods such as those employed by Prof. Palmer continue to be characterized in legal filings as “state of the art.” While these approaches have a long and established history in the literature and form the basis for my model, we believe VoteHub’s model represents a meaningful methodological advancement and produces more accurate estimates in many contexts. It is precisely this improvement in performance and reliability that underlies the model’s commercial value.

That said, I want to be clear about what this does and does not mean. “Proprietary” does not mean that the model is unexplainable or opaque. VoteHub and I have been quite transparent for a private-sector model: the methodological structure is laid out publicly, minus a small number of constants and implementation details that are sensitive from a commercial standpoint.

That math is available here, down to a few constants introduced on Page 10: [link].

What I Am and Am Not Claiming

It’s important to separate two questions:

1. Whether the Voting Rights Act should apply here, and what the remedy is

2. What the voting behavior in NY-11 actually is

I am not a voting rights lawyer, and I do not have an opinion on how the legal question should be resolved. My focus is on the empirical claim of voting behavior underlying the case presented by the petitioners.

And empirically, the conclusion is clear:

Hispanic voters in NY-11 do not vote as a Democratic bloc. Traditional ecological inference analyses that suggest otherwise conflict with national benchmarks, with the district’s political reality, and with what we know about how demographic partisanship changes across political geography from voter file data.

Closing Thoughts

This case illustrates why ecological inference—and, in particular, careful treatment of violated independence assumptions and aggregation bias—has real-world consequences. More broadly, my work on ecological inference models and national estimates is driven by both intellectual interest and a commitment to producing analyses that are informative, transparent, and accessible to a wider audience, even as they invite public scrutiny.

Thanks for reading Data and Divergence. Subscribe for free to receive new posts and support my work.

2025 was also a big year for me personally. It marked the end of long chapters at Yale and Decision Desk HQ, and the beginning of new ones at Harvard and VoteHub. When I was five years old, I stayed up all night glued to TV coverage of the 2008 presidential election, reading every number on the screen and begging my parents to let me stay up far past my bedtime. A year ago today, my New Year’s resolution was to put more energy into sharing my work with a broader audience, and in 2025 that effort resulted in 60 million impressions and 2 million total engagements. I am deeply grateful to everyone who has read, shared, argued, and learned alongside me here and on X. Being able to share my passion with others has been incredibly rewarding, and I am excited to see where it leads in 2026 and beyond.

2025 Data & Divergence Year in Review

1. D&D in the NYT

My breakdown of Zohran Mamdani’s shock primary win in June 2025 was among my favorite work this year. His victory in the New York City mayoral race was by far the biggest electoral story of 2025, and as Mamdani’s mayoralty begins at midnight on January 1, 2026, my thoughts below still are relevant. They help frame what may be the central question of 2026: how well does Mamdani actually perform as mayor?

2. VoteHub National Precinct Map

Definitely the coolest project I’ve ever been a part of. It was especially rewarding to fold my senior thesis into something tangible and public. Here’s my guide of how to use VoteHub’s first publicly available, precinct-level estimates of presidential support by demographic on a national scale, and still the only complete national precinct map of the 2024 U.S. presidential election.

3. What’s in a name?

With the help of the legend Havish Netla, I also ran a whole series on the partisanship of first names using voter registration data that people really enjoyed. It’s silly fun, but there are also some genuinely interesting takeaways, like how much of the emerging new GOP base is made up of male Zoomers.

4. Dems Collapse With Hispanics

The single biggest story of the 2024 election was Hispanic voters’ dramatic shift away from the Democratic Party, and covering it was especially fascinating. It was even more rewarding to apply my senior thesis directly to the 2024 presidential election and see just how large and consistent those Hispanic shifts away from Democrats were across virtually every major U.S. city.

5. Youth Voters Abandoning the Democratic Party

The second-biggest story of the 2024 election was young voters’ sharp shift away from the Democratic Party, evident across polling, voter registration data, and election results on college campuses. Tracking and covering this trend throughout the year was especially fascinating to me and quite surprising.

6. Youth Voters Coming Back to the Democratic Party

I’ve also focused on highlighting a growing body of evidence, from both polling and 2025 election results, showing that young voters are turning against Trump and Republicans at a striking pace. The gains the GOP made with young voters in 2024 don’t appear to be sticking, and I see this as the second-most important story heading into 2026.

7. NYC Early Vote Tracking

The New York City Board of Elections is outstanding to work with. Every time we’re in touch, they’re responsive, helpful, and generous with high-quality data for analysts. Tracking the massive youth turnout surge and Zohran Mamdani’s transformation of the NYC electorate alongside them was fascinating, genuinely important, and a lot of fun. How often do you see young voters outvote older voters? Almost never. And Mamdani’s campaign may have been the most effective use of social media to mobilize young people we’ve ever seen in American politics. It really felt like the start of a new era and a great preview for the 2028 Democratic Presidential Primary.

8. Marriage Pact

I will always be partial to projects that are fun, creative, and a little off the wall. One example was scraping data from the Marriage Pact survey, an annual matchmaking experiment on U.S. college campuses where students fill out compatibility questionnaires to be paired with a hypothetical future “safety” spouse if they end up unmarried. I pulled data from about 50 schools, including Yale, where many of my friends (and I) filled it out at one point, and shared content based on the results. Was it a randomly selected sample? Definitely not. But as a funny, solid supporting data point that complemented other analyses, sure.

9. The Starbucks Divide

Another great off-the-wall stat: even after controlling for population density and race, precincts with a Starbucks are still far more Democratic.

10. Street Cameras

More in the out-of-the-box category: a few weeks before Election Day, I searched for snow across central Tennessee to see how much the state actually got on the morning of December 2, since weather was a real question going into the election. Republican Matt Van Epps needed good weather to boost Election Day turnout, and in Tennessee even a little snow can matter since it rarely sticks and people are generally unprepared for it. Central Tennessee did see some snow that morning, but not enough to accumulate or meaningfully affect turnout.

That’s all for this year. Happy New Years.

Texas was the centerpiece of that strategy. Gov. Greg Abbott and Texas Republicans passed an aggressive new gerrymander that, if implemented, would likely shift three to five seats from blue to red. Republicans moved quickly in other states as well, enacting more favorable maps in Ohio and North Carolina and pursuing a similar push in Florida, Missouri, and Indiana.

Democrats, of course, weren’t going to sit still. In California, Gavin Newsom pushed through Prop 50, replacing the state’s relatively fair commission-drawn map with a partisan Democratic gerrymander that effectively offsets the gains Republicans hoped to lock in with the new Texas map. And in Virginia, after winning full control of the state government in 2025, Democrats now look poised to draw a very aggressive map that could flip three to four seats blue.

A big twist came yesterday: a panel of federal judges blocked Texas’s new map from being used in 2026. State officials immediately appealed to the U.S. Supreme Court, and I think it’s more likely than not (but far from certain) that the Court issues some sort of stay that keeps the Texas map in place. Even so, the momentum has shifted. With Democrats picking up wins in CA, VA, and UT, the overall picture looks much less like a Republican firewall and much more like a chaotic, state-by-state scramble that could end up benefiting Democrats just as much—or even more—than Republicans.

2026 US House State of Play

What This Chart Shows

This chart projects how many House seats would lean towards Republicans in 2026 under different electoral scenarios and redistricting outcomes.

The X-axis (bottom) shows the national US House vote margin—ranging from Democrats winning by 6 points (D+6) on the left to Republicans winning by 4 points (R+4) on the right. “Even” means a tied national vote.

The Y-axis (left side) shows the number of House seats Republicans would be expected to win out of 435 total. The dashed line marks the majority threshold—whichever party gets 218 or more seats wins control of the House.

The three lines represent different map scenarios:

• Gray (2024 Maps): The maps used in 2024

• Red (Expected 2026 With TX): Expected 2026 maps if Texas redistricts along with OH, VA, CA, NC, UT, FL, and LA

• Blue (Expected 2026 Without TX): Expected 2026 maps if Texas doesn’t redistrict while OH, VA, CA, NC, UT, FL, and LA do.

Key Takeaways

1. The Mid-Decade Redistricting Arms Race Is On Track to be a Wash

To measure the structural bias of the House under each scenario, you can look at where the lines cross the 218-seat majority threshold. On the 2024 maps (the grey line), that crossover happens at roughly D+1.6 (R+1.6 structural bias), meaning Harris needed to win the national popular vote by about 1.6 points to carry a majority of House districts. The expected 2026 maps with Texas’s new gerrymander (the red line) push that bias to about R+2, while the scenario without Texas (the blue line) brings it closer to R+1.

Texas is the biggest swing factor in the mid-decade redistricting cycle, but the overall effect is on track to be roughly a wash. Whether the new Texas map survives or not, the structural bias of the expected 2026 landscape—if applied to 2024-level results—ends up looking pretty similar to the bias we saw under the 2024 maps. In other words, neither party is on track to gain much structural advantage from redistricting alone heading into the midterms.

2. Mid-Decade Maps Are Making the House Less Competitive Than Ever

Redistricting after the 2020 US Census was already packed with aggressive gerrymandering on both sides, and while this gerrymandering largely canceled out in partisan terms, it sharply reduced the number of competitive seats. This second round of mid-decade redistricting has pushed states even further: both Democratic and Republican legislatures have adopted more severe maps that again mostly offset each other, but further shrink the pool of competitive districts. In the figure above, you can see this in the red and blue lines—both are noticeably flatter than the grey line—meaning the number of seats that could realistically swing between a strongly Democratic year and a strongly Republican one is much smaller. The battleground will be narrower than ever, and the House will almost certainly be decided by extremely tight margins, even after accounting for the national political environment.

Too Many Unknowns to Call the Fight Yet

It’s still highly unclear which states will even be allowed to redraw their maps, given the mix of ongoing federal court battles, the possibility of Supreme Court overrides, potential ballot measures, state legislatures deciding whether to join the national redistricting push, and the still-pending Section 2 Voting Rights Act ruling in Louisiana v. Callais. As of November 19, Kalshi shows nine different states with between a 20 to 80 percent chance of redistricting again before the 2026 elections. There are many independent events that could break better for Democrats or Republicans, but the key takeaway from our current trajectory is twofold: (1) we’re heading toward a very similar, slightly Republican structural bias as we saw in 2024, and (2) we’ll see a massive decline in competitive general elections due to increased gerrymandering across the country.

New York has nine days of early voting, and the figures and data below reflect totals through Day 5. I’ll continue to post my analysis of Early Voting on X.

Summary

The outcome of the 2025 New York City mayoral race is looking less and less in question. Zohran Mamdani enters the final stretch with a consistent lead in the polling, and early voting data doesn’t challenge those numbers. Curtis Sliwa continues to stay in the race and peel away anti-Mamdani voters from Andrew Cuomo, leaving the former governor with little room to maneuver.

Cuomo’s only realistic path requires a combination of several unlikely events: a collapse in Sliwa’s support, a surge in Republican turnout, and a polling miss that restores his primary-era strength with Black Democrats. So far, none of those boxes are being checked. Early voting data suggests a bluer and reasonably young electorate that plays directly into Mamdani’s strengths. Mamdani appears on track for a decisive victory that would cement his status as the national left’s most prominent new face in city politics.

The Good for Mamdani

1. He’s Leading Comfortably, and It’s Not Changing

Mamdani is consistently polling ahead by double digits, and the electorate projected in those polls appears in line with what we’re seeing so far in early voting. Candidates leading by this much rarely lose. Sliwa continues to play the spoiler role effectively, drawing a sizable share of the anti-Mamdani vote from Cuomo. Mamdani is polling strongly with registered Democrats (including Black Democrats, a group he lost handily in the primary) who are expected to make up nearly three-quarters of the electorate on Tuesday.

2. The Age Mix Is Shifting Younger

There’s been a lot of buzz this week about how different November turnout has been from the June primary in terms of age, largely thanks to this good article from Gothamist. Older voters led the first three days of early voting, which seems good for Cuomo as he did much better with older voters in the Democratic Primary. However, younger voters have been showing up in greater numbers over the past couple of days, making the age gap less of a problem for Mamdani.

3. Bad Republican Turnout

To have any real chance at beating Mamdani, Cuomo needs to assemble a coalition of moderate Democrats (particularly older voters) and Republicans who back him over Sliwa. Compared to Mamdani, he’ll likely win Republicans and lose Democrats. That makes lower Republican turnout especially damaging, given how much he’s relying on them to stay competitive.

Early voting in the 2024 general election:
🔵 69.9% Registered Democrats

Early voting in 2025 (through 5 days):
🔵 74.0% Registered Democrats

New York City was Harris+39 in 2024, and based on early vote data, we’re currently tracking toward a Harris+45 electorate this November by my estimate. That makes the math even tougher for Cuomo.

The Path (and Problems) for Cuomo

There definitely have been some good datapoints for Cuomo over the past week. To stay competitive, he needs to roughly match his June primary performance with registered Democrats, winning around 40% of them, and the Democrats turning out so far look more favorable to him demographically. Turnout among older voters and Black voters, both groups he won comfortably in June, has been strong. The early-week data sparked a bit of momentum for him, reflected in rising prediction market odds and a string of GOP endorsements backing him over Sliwa.

But there have also been clear negatives for Cuomo, and a mix of positives and negatives isn’t great when you are consistently trailing by double digits in polling.

For Cuomo to make this a real race, several things would need to break his way. First, Sliwa would have to badly underperform his polling and act less as a spoiler than he has so far. That is not impossible, since candidates with no realistic chance often underperform polling when voters make strategic choices, but it is far from guaranteed. Second, Republican turnout would need to rise significantly. The problem with relying on reluctant anti-Mamdani voters is that they’re not particularly motivated to show up, and GOP turnout has been weak so far. That could always change if Republicans start turning out in force, but as things stand, the math is much easier for Mamdani since he is winning Harris voters and only losing among Trump voters. Third, Cuomo would need a fairly large polling error. Recent polls suggest he has lost a lot of ground among groups he performed well with in the primary, especially Black Democrats. For him to have a shot, that would need to be wrong.

If all three of those things happened—Sliwa fading hard, Republican turnout surging, and Cuomo holding his primary-level support with Black voters—Cuomo would have a chance. But right now, it looks like he is 0-for-3.

ZD Final Prediction

(This reflects my own personal view, not a model or a forecast from any organization I’m affiliated with.)

Mamdani 51%
Cuomo 33%
Sliwa 14%

(This is not meant to be a methods piece, but if you are curious about how I produced the estimates I’ve linked a more formal explanation here)

I will frame this analysis through two towns: Westerville, OH and West Covina, CA.

1. Westerville, Ohio — a family-oriented suburb outside Columbus with manicured lawns, strong schools, and golf courses — is the kind of place once seen as reliably Republican but now increasingly Democratic. It exemplifies the suburban areas where Democrats are gaining ground.

2. West Covina, California — a diverse community near Los Angeles with many residents who are immigrants or children of immigrants from Mexico and Asia — was long considered safely Democratic but is now edging toward toss-up status. It exemplifies the diverse, once-solidly blue areas where Republicans are making inroads.

The Great American Political Realignment

Over the past eight years, America has undergone a major political realignment. Democrats have traded their diverse working-class coalition for a new base anchored by just two groups: cosmopolitan White voters and Black voters. Meanwhile, Republicans have maintained their stronghold among rural and exurban White voters while massively cutting into Democratic margins with Hispanic and Asian voters, particularly across the Sun Belt and parts of the urban Northeast.

The degree of White educational polarization in the United States cannot be overstated. The relationship between the share of adults with a bachelor’s degree and the estimated White Democratic vote share in White-majority precincts is strikingly clear and consistent.

I think most people reading this have probably heard before about the dynamics I’m describing here, but geographically, here is what this has looked like:

Northern suburbs have driven the shift toward Democrats, while heavily Hispanic and Asian urban cores have trended toward Republicans. I focus on two towns that embody these opposing movements. In 2012, West Covina, CA was over 40 points more Democratic than Westerville, OH. By 2024, the two communities voted almost identically.

The Great American Political Realignment (R to D): Westerville, Ohio

Westerville, OH is an inner northern suburb of Columbus, only about a 20-minute drive from downtown. Biden carried it by 16 points in 2020, and Harris actually expanded that margin to 18 points in 2024—even as she lost ground nationally. Why? Westerville exemplifies all four traits that made for the strongest Democratic shifts in 2024: (1) White, (2) Anglo-Saxon, (3) adult, and (4) upper-middle class voters.

1. First and foremost, Westerville is very White—82% compared to 58% nationwide. Only 3% of the population is Hispanic and 4% is Asian. Across the country, Hispanics shifted heavily toward Trump, and Asians trended that way too (which is why Harris performed worse in nearby Columbus suburbs like Dublin and Orange Township, which have much larger Asian populations than Westerville). Nearly all of the places where Harris improved on Biden’s margins were overwhelmingly White, and Westerville fits that pattern.

   

2. Westerville isn’t just White—it’s very Anglo-Saxon. The two most common White ancestries in the suburb are German and English. In 2024, “White ethnic” voters such as Italian, Irish, Jewish, and Portuguese shifted much more strongly toward Trump than voters of German, English, Dutch, or Scandinavian descent, even after controlling for education. This was especially evident among Italians and Irish across the Northeast, but it also appeared in White ethnic enclaves around Chicago and northeastern Ohio, and even in the Irish mining town of Butte, MT, which swung four points towards Trump than any other town of similar size in the state. To me, this is one of the most fascinating and puzzling trends of 2024. One possible explanation is that Biden at the top of the ticket boosted Democrats among White ethnic Catholics in 2020, but even comparing 2016 to 2024 shows a consistent Republican drift among these groups. Just as Asian and Hispanic voters consist of diverse subgroups with distinct patterns, White voters also vary internally, and ethnicity remains an important dividing line.

3. Westerville is primarily a “family suburb.” It has about 40,000 residents, a small downtown, and 77% of housing structures are single-family units compared to 55% in Columbus. About 60% of residents are married, compared to just 37% in Columbus. Democrats struggled significantly with young people, especially young men, in 2024 as I have written about constantly and as highlighted in post-election reports by David Shor and Catalist. Harris did great in Westerville, but just 10 miles away, shifts on the Ohio State campus (which is also dominated by highly educated White voters) were very unfavorable for Harris. Westerville, however, is more about parents casting ballots, with kids either too young to vote or that have already moved out.

4. Westerville is an upper-middle-class suburb, but it does not have a large concentration of ultra-wealthy residents. Harris actually performed a few points better there than in some of the wealthiest enclaves in the country—places like Greenwich, CT, the Upper East Side of Manhattan, Highland Park, TX, and Bel Air, CA—where Trump made major gains. This points to a broader development in 2024: education polarization still defined the electorate, with college-educated voters (especially White ones) leaning Democratic, while non-college Whites and an increasing number of non-White working-class voters leaned Republican. That basic pattern held, but something new emerged at the very top of the income ladder. Trump made meaningful inroads with a narrow group of highly educated elites in business, tech, finance, and crypto. Figures such as Elon Musk, David Sacks, Chamath Palihapitiya, Bill Ackman, and Miriam Adelson were either openly supportive of Trump or clearly sympathetic to him. Much of this shift stemmed from frustration with Democratic positions on regulation, DEI, and the general direction of cities and governance. These individuals are not “typical” college graduates—they represent a small slice of the electorate, but their influence over money, media, and political narratives (especially online) is enormous.

   

   That influence is already visible. One of the most undercovered stories of 2025 is how badly Democrats are lagging in fundraising ahead of the 2026 midterms, a sharp reversal from 2018 and 2020 when they had such a cash advantage that they poured resources into longshots in places like South Carolina and Kentucky. So even though education and class polarization still broadly benefit Democrats, the nuance matters. In 2024, suburbs like Westerville were a few points stronger for Harris than the ultra-rich enclaves. It is an early sign that Democrats are no longer reaping the full benefits of educational polarization among a group that, though small in number, wields outsized political and financial clout.

The Great American Political Realignment (D to R): West Covina, California

The worst groups of voters for Democrats in 2024 were Hispanic voters and Asian voters. I chose West Covina as a prime example of this shift because the city includes both populations. West Covina is 51% Hispanic and 27% Asian, and it moved from Biden +32 in 2020 to just Harris +18 in 2024. Nationally, Hispanics and Asians make up only 14% of voters, yet they accounted for 42% of the entire shift toward Trump (in raw votes) between 2020 and 2024.

West Covina itself is a middle-class city with both working-class neighborhoods and more affluent areas, but consistent with the national trend, the movement toward Trump among Hispanic and Asian voters cut across class lines.

According to my precinct-level estimates, there is essentially no evidence of educational polarization among Asian voters. I find this especially interesting because it clashes with my prior from smaller-scale research and casual scans of precinct results. The likely explanation is that highly educated Asian voters often cohabit with Whites, which masks how strongly highly ed. Asians themselves shifted toward Trump. Among Hispanic voters, there is some evidence of educational polarization, but it is fairly limited. Even in higher-education areas, Hispanic voters still moved sharply toward Trump.

Ideological Polarization

One key way to understand minority shifts to the right is through the lens of ideological polarization. It is not that minority voters are becoming much more conservative overall, but rather that Republican candidates are now performing better controlling for the political ideology voters identify with, as shown in the plot below. In practical terms, Democrats once won a majority of non-White voters who described themselves as “conservative” or “very conservative,” but in 2024 they captured only about a quarter of them. Among Hispanics and Asians who identify as “moderate,” the Democratic margin has shrunk from roughly D+50 to around D+20. These shifts are especially pronounced among younger voters.

A evenly ideologically polarized electorate, controlling for race, would have the four lines in the figure below stacked on top of each other. The 2024 graph (mostly converging except for the dark blue line) suggests Republicans still have significant room to grow with Black voters, and a bit more room to gain among Hispanics and Asians.

Hispanic Voters

As in the country as a whole, West Covina’s minority shift to the right was primarily driven by Hispanic voters. My estimates suggest Hispanics there moved from about 65 percent Democratic in 2020 to just 54 percent Democratic in 2024. Nationally, Hispanics are now close to a tossup group, and the implications for Democrats are sobering. On one level, it is a rhetorical blow: after years of emphasizing diversity and DEI in their messaging, Democrats are watching minorities, especially young minorities, move decisively toward Donald Trump. On an electoral level, the consequences are bad but not the end of the world. While the shift hurt Democrats in Arizona and Nevada, its larger effect has been to take the massive states of Texas and Florida off the map for the foreseeable future, given how underrepresented Hispanics are in the Senate and Electoral College battleground.

It is also crucial to remember that “Hispanic” is not a monolith but an umbrella covering many distinct groups. Democrats fared poorly with Mexicans, who are concentrated in the West, Texas, and Chicago; Puerto Ricans, particularly in Florida and Pennsylvania; and Cubans, in Miami and Tampa. But even more dramatic losses came among other communities: Salvadorans in the DC area and northeast of Boston, Dominicans in places like Lawrence, Massachusetts, Paterson and Perth Amboy, New Jersey, and Providence, Rhode Island, and Ecuadorians in Queens and New Jersey. The shifts among Mexicans and Puerto Ricans were serious, often up to 15 to 20 points, but the swings among Dominicans, Central Americans, and South Americans reached extraordinary levels, often 35 points or more. In Corona, a heavily Ecuadorian neighborhood in Queens with about 20k Hispanic voters, Biden carried 76 percent in 2020, while Harris won only 54 percent in 2024.

Asian Voters

The implications of the Asian shift to the right are complicated. First, it is concerning for Democrats who hope that racial depolarization, meaning minorities moving toward Trump, can be mostly explained by educational polarization, meaning high-education voters moving toward Democrats and vice versa. Asians are one of the most highly educated groups in the country, yet Democrats are struggling with them, as well as with highly educated Black and Hispanic voters, as my precinct-level analysis shows. Second, Asians are badly underrepresented in American politics. They are concentrated in uncompetitive coastal states like California, Washington, Virginia, New Jersey, and New York. I have written before about how underrepresented Jews are in the Electoral College and the Senate, and the same logic applies to Asians. Since their political strength is expressed most directly at the House level, many of the consequences will be felt there. For example, rumored Democratic gerrymanders in California are now much harder to execute because of the Asian exodus from the Democratic Party.

As with Hispanics, Asian voters are not a monolith but rather an umbrella category that covers many different ethnic communities. Predominantly Muslim South Asian groups, such as Bangladeshis and Pakistanis, appear to have shifted the most to the right. Muslim Indians likely did as well, although the data is thinner. Koreans, non-Muslim Indians, and Chinese Americans all moved strongly toward Trump, with Koreans perhaps shifting the most, though it is difficult to say definitively with the available data. Vietnamese Americans likely shifted the least from 2020 to 2024, and in fact some Vietnamese areas in California actually moved slightly to the left. But this is somewhat misleading, because Vietnamese voters had already swung hard to the right between 2016 and 2020, so in 2024 they were simply ahead of the curve.

Andrew Cuomo is the former Democratic Governor of New York who was forced to resign in August 2021 following a state attorney general’s report that found he had sexually harassed multiple women, including current and former government employees, in violation of state and federal law. Zohran Mamdani is an insurgent and progressive Democratic member of the New York State Assembly, representing the 36th District in Queens since 2021. Mamdani leads 43%-36% with 96% of the 1st round vote counted according to DDHQ, and once ranked-choice voting is tabulated, Mamdani is widely expected to surpass the 50% threshold needed to win the primary. Brad Lander, who’s running a strong third with more than half of the remaining vote, has endorsed Mamdani.

The Central Role of Young Voters

First and foremost, young voters carried Mamdani to victory in the primary. In Democratic primaries, younger voters (especially millennials) consistently lean far more progressive than older generations, and that played out here. The precinct-level data makes that pretty clear.

As you can see above and below, Cuomo actually won the oldest precincts while Mamdani won the youngest precincts in a 30 point landslide.

This pattern held for both white and non-white voters. Cuomo was widely expected to dominate with non-white voters, especially Black voters, in pre-election polling. But he lost non-white voters in younger areas, which really did him in. This data is a little more noisy (you can see the thicker confident interval shading on the plot below) but you can see that this positive slope holds even after stratifying by race, and is especially notable in Black majority areas.

Upper-Middle-Class Voters Continue to Anchor Urban Progressive Blocs

Second, upper-middle-class voters were the engine behind Mamdani’s win. Cuomo performed best in the city’s poor and working-class areas—though those are relatively limited outside the Bronx and Orthodox Jewish neighborhoods. His second-best showing came in the ultra-educated, high-income parts of the city, which tracks: the NYC elite mobilized against a democratic socialist.

I posted a much worse version of this chart on Twitter, and it got tons of love because the underlying pattern is pretty interesting and so much hate because it was *terrible* data viz—the massive mob of blue points was apparently very distracting to people, who would have guessed. I like keeping the points in the plot to show that a Cuomo+10 average in low-education areas actually reflects a mix of tons of Mamdani+20 precincts and tons of Cuomo+50 precincts. But if this keeps causing confusion, I might have to drop them entirely. For now, I’m hoping that making the points smaller and lighter helps shift the focus to the red line—and especially the SE ribbon around it.

Expansion Beyond Traditional Progressive Coalition

As I noted above, it’s totally expected that young and upper-class voters of all races were the core of Mamdani’s progressive coalition—that’s standard in Democratic primaries. The big new story here is how well Mamdani did with Hispanic and Asian voters. Cuomo was polling strong with both groups and was widely expected to win them on Tuesday. But that didn’t happen. He lost Asian voters pretty much across the board, and got crushed with more assimilated, younger, higher-education Hispanic voters (especially in areas where Hispanics weren’t the majority, like a lot of Manhattan and Brooklyn).

What’s notable is that both groups swung hard toward Trump in 2024. I think that’s connected for two reasons: (1) there's deep frustration with the Democratic establishment, and a lot of voters aren’t as ideologically consistent as people think—voting for both Trump and Mamdani can be consistent as anti-establishment choices. And (2) many of the truly moderate or conservative Hispanic and Asian voters have already left the Democratic Party, so they’re not there anymore to support moderates like Cuomo. That’s not the case with Black voters. Conservative and moderate Black Democrats are still a major force in NYC primaries, and they anchored Cuomo’s support.

Mamdani General Election Chances

Mamdani is widely expected to win the Democratic primary next Tuesday when RCV is computed and I consider him a heavy, heavy favorite to win the general election. Sometimes I feel guilty because in D&D I just tell people that prediction markets are roughly right, but I think Mamdani has a significantly better than 74% chance to become the next mayor of New York City.

The current field includes Andrew Cuomo running on an alternative ballot line, incumbent mayor Eric Adams as an independent, Republican nominee Curtis Sliwa, and a genuinely strong independent bid from attorney Jim Walden. In a five-way race like this, with the opposition split, Mamdani will win easily. Even in a three- or four-way contest, he’d be heavily favored. It’s possible that a couple of Cuomo, Sliwa, and Walden drop out, but unless the field clears to a two-way race, he’s in a strong position.

If people actually wanted to coordinate to stop Mamdani, it wouldn’t be that hard. Being mayor of Manhattan would be like being mayor of San Francisco or Boston, but NYC as a whole isn’t the most Democratic city in the country. It went from Biden +54 to Harris +39—so a candidate with Republican backing would start with a solid 25–30% base and only need to win about a third of Harris voters to beat Mamdani.

If the field cleared for him, Jim Walden would probably be favored in a 1v1 matchup against Mamdani. But that level of coordination seems pretty unlikely right now. (Others also agree this is by far the clearest route to beat Mamdani but is incredibly unlikely to happen). Adams himself could have a shot at beating Mamdani if he managed to clear Cuomo, Sliwa, and Walden from the field—which is a more realistic path. But even then, I think Mamdani would still be favored. Adams would be running with a pretty unusual coalition: Trump voters and Black voters. That’s an interesting proposition, but I think he’d have a harder time than someone like Walden, especially with the technocratic Kathryn Garcia-type voters who are likely to be turned off by Adams because of his corruption issues and relatively friendly relationship with Trump.

Throwing Some Water on the Fire

As I said in the title, this is definitely set to be the left's biggest win since AOC defeated Joe Crowley, and potentially even bigger. But let me play devil’s advocate and throw some cold water on the excitement.

It's nothing new for the left to prove they can win Democratic municipal primaries in the most lefty cities in the country. Sure, this was against a massive disadvantage in name recognition and fundraising, and NYC is a much bigger fish to fry that gets far more national media attention. But, Brandon Johnson won in Chicago just two years ago and now sits at a 6% approval rating as mayor. Progressive DA George Gascón won in Los Angeles in 2020, only to get absolutely demolished by an independent in 2024, running a staggering 53 points behind fellow Democrat Kamala Harris in the same county. The Bay Area tells a similar story. Progressive prosecutors Chesa Boudin and Pamela Price both became district attorneys, but their tenures were such disasters they couldn't even finish their terms before getting recalled. Then Oakland turned around and elected progressive Barbara Lee as mayor in 2025.

This is shaping up to be the left's biggest electoral win in forever, but it could all be for nothing (and potentially really damaging to the gaining momentum of the movement) if Mamdani can't break the recent pattern of his progressive predecessors having spectacularly unsuccessful tenures actually governing.

These races come on the heels of a string of special elections for state legislative seats, where Democrats have been performing exceptionally well. On average, Democratic candidates have outperformed Kamala Harris’s 2024 showing by 12 points, including a headline-grabbing flip of a State Senate seat in central Pennsylvania that Trump carried by 15 points last November. But, to be clear, this trend isn’t new — Democrats have been dominating in low-turnout elections for over a year and a half now.

I broke down these recent Democratic overperformances and previewed Tuesday’s special elections during my segment on NewsNation on Friday. You can watch the full clip here, especially if you’re interested in the Florida specials.

A Brief History of Voter Turnout Dynamics

For decades, political analysts and strategists believed that higher voter turnout favored Democrats, based on the assumption that nonvoters—often younger, lower-income, and more racially diverse—were more likely to support Democratic candidates. This idea became particularly salient during the Obama years, when Barack Obama's 2008 campaign mobilized millions of new and infrequent voters. Black voter turnout reached a historic high, matching white turnout for the first time, and surged in key swing states, helping Democrats win the presidency and make gains down-ballot. The success of Obama-era turnout operations seemed to reinforce the longstanding belief that Democrats benefit from expanded access to the ballot.

Republicans have historically tried to take advantage of the idea that the electorate in lower turnout elections would be more conservative, while liberals have complained about it as disenfranchisement. Georgia’s runoff rule, which requires a candidate to win a majority of the vote to avoid a runoff, dates back to 1964. It was implemented by segregationist Democrat Herman Talmadge to dilute Black voting power after the Supreme Court’s one person, one vote ruling. More recently, Georgia Republicans have maintained and adjusted the rule—most notably shortening the runoff period in 2021—amid criticism that it disproportionately affects minority voters and can suppress Democratic turnout in high-stakes races.

Changing Coalitions

I’ve been tweeting about this nonstop, but it’s worth emphasizing again: since 2020, we’ve been living through a major political realignment. Republicans have made significant gains with voters of color, particularly Hispanic and Asian voters, who tend to have lower turnout and political engagement. At the same time, Democrats have improved with well-educated White voters, who historically turn out at much higher rates.

I'm including one of the most insightful charts I came across last year — one I cited constantly throughout the summer and fall of 2024. It shows that high-engagement Americans, those who voted in 2020, didn’t shift much at all between 2020 and 2024. But among low-engagement Americans, meaning those who sat out in 2020, the movement was dramatic: a sharp swing from Blue to Red in just four years.

David Shor of Blue Rose Research estimates that if every eligible voter had cast a ballot in 2024, Trump would have defeated Harris by about 5 points, rather than the actual margin of 1.4. In other words, the race was only close because many conservative-leaning voters — particularly those with lower turnout propensity — stayed home. One particularly striking detail from the same analysis: the 2022 midterm electorate was roughly two points more Democratic than the 2024 electorate, despite 2022 being a relatively strong cycle for Republicans, who gained seats in the US House as the opposition party.

Broadly speaking, we’re moving toward a dynamic where lower-turnout elections tend to favor Democrats. If everyone voted, this would be ideal for Republicans. Their best-case realistic scenario is a high-turnout presidential election, with midterms coming in a close second. But the further turnout drops — especially in these random special elections — the more the environment tilts toward Democrats. This pattern is particularly visible in the ultra-low-turnout special elections we’ve seen recently. Democrats tend to thrive in these contests, especially in lower-profile races like state legislative specials, where media attention is minimal and the electorate skews more engaged and left-leaning.

What does this mean for 2026 or 2028?

Democrats have been performing well in recent special elections, and they’re likely to notch more wins on Tuesday — potentially flipping the Wisconsin Supreme Court seat and racking up a major overperformance in Florida. But I’d be cautious about reading too much into what that means for 2026 or 2028.

Historically, special election results have been a pretty good indicator of the national political environment heading into the next general election. But that pattern broke down in 2024. Democrats ran up the score in specials throughout 2022 (which I wrote about here), 2023 and early 2024, only for Harris to flop against Trump in November. Why? Because the electorate in these really low turnout specials is fundamentally different from the one that shows up in higher-turnout elections.

Right now, the highest-propensity voters in the country are disproportionately well-educated, white, and left-leaning — the kind of "resistance liberal" Democrats who are especially activated when Trump is back in the spotlight. They dominate special election turnout, but they don’t represent the full electorate you see in a midterm or presidential cycle.

There are still good reasons to think Democrats could do well in 2026 — most notably the historical trend that presidents almost always take a hit in their first midterm, and some lukewarm public reception to early moves by the Trump administration. But I wouldn’t count great special election results as one of those reasons. The Democratic edge in these contests feels more like a structural feature of the current turnout dynamics than a sign of broader political momentum. It’s getting a bit ridiculous at this point after Trump’s victory — and a close race in FL-06 on Tuesday would only add to the surprise. But it’s worth keeping perspective.

I've had a pretty good run with UConn lately - their back-to-back titles have helped me make over $4k combined from bracket pools in the past two years, though I'll be the first to admit luck played its part. Last year's less public version of this post seemed to help quite a few people with their pools, and while past performance doesn't guarantee future results (and I might be due for some mean reversion), I figured I'd share my thoughts this year.

From a statistical standpoint, if you follow my advice in bracket pools, you’re probably looking at an expected (average) return of 2-5x on your investment, with a bigger edge (but more variance because you’re going to hit big super rarely) in larger pools and a smaller edge (but less variance) in smaller pools.

Table of Contents

SECTION 1: Summary

SECTION 2: Math Thoughts

SECTION 3: Basketball Thoughts

SECTION 4: Full Model Probabilities

SECTION 5: Bonus Yale Thoughts

If anyone has questions abt a specific team I didn’t cover (e.g. why are you so low on Purdue?) feel free to leave a comment.

SECTION 1: Summary

Compared to what the general public (the people you’ll be playing against in your bracket pool) is picking, I am pretty high on Duke and Houston to win the title and Texas Tech to advance to the E8 or even F4/Title Game, and I am low on St John’s and Kentucky to advance to the E8, Michigan to advance to the S16, and Memphis to advance to the 2nd Round. I recommend picking more favorites from a seed perspective than you are used to both early and late in the tournament, since there are pretty big tier breaks between the 4 and 5 seeds and the 8 and 9 seeds this year which is convenient for picking teams. This is the strongest set of 1 seeds compared to the field in the history of the NCAA tournament. It’s a boring year for making brackets, because in most situations you want all 1 seeds in the Final Four. If you’re in a small pool (<100 people) you want to be picking all 1 seeds to the F4 (really anyone can win but I would recommend Duke or Houston) with the potential exceptions of Texas Tech or even Maryland. If you’re in a larger pool (>100 people) it opens you up to some more fun stuff like Louisville/Gonzaga/Iowa State/Arizona to the F4 or Colorado State/UNC (if they win their play-in game) to the Elite 8. In really large pools (>500 people) you can have Tech/Maryland/Louisville/Gonzaga/Arizona/Colorado State win the title or lose to Duke/Florida or something.

SECTION 2: Math Thoughts

These tips apply to brackets every year. First, understand the math of brackets. To maximize your chances of winning your pool, you need to do two things: pick good teams who are likely to win and differentiate yourself from the rest of your pool—especially in the later rounds. Differentiation matters in earlier rounds too, but it’s usually more smart to pick “chalk” (the favorites). The goal is to pick up points where others lose them. You’re better off finishing first or last—not somewhere in the middle.

Winning money through bracket pools is subtle and long-term. You’ll lose more often than you win, especially in big pools. Success comes from finding small edges that build up over time. Remember, for a team to reach a later round, they must win every game leading up to it. A super easy second-round matchup can make a team a great title pick.

Next, consider the size of your pool. The larger the pool, the more risks you should take. In small pools (fewer than 25 people), it’s often best to pick mostly chalk, perhaps deviating slightly with your title pick. In mid-sized pools (around 100), you’ll still want to stick with a lot of chalk. When you are getting above 100 then having some lesser teams in the F4 starts to make more sense.

The pool’s scoring system is also key. In seed-bonus pools, where you get a bonus for picking lower-seeded teams to win, take advantage of that by picking strong double-digit seeds in the first two rounds. However, you should still pick a top seed to win the title. In flat-scoring pools—where the point structure is something like 1-2-3-4-5-6 instead of 1-2-4-8-16-32—picking chalk in the later rounds is crucial since differentiation is less important. Focus on accuracy with a little differentiation in the early rounds.

(the rest of the general advice is for people who take this more seriously, skip to the next section if you’re getting bored)

How “sharp” your pool is also matters. If you’re in a casual pool with friends, family, or work colleagues, it’s fine to pick trendy teams, such as teams with strong traditional (KenPom) analytics profiles for their seed like Gonzaga or Missouri to make a deep run or early-round upsets like Baylor or VCU. But if you’re in a sharp pool with more serious players who will be looking at analytics and think they know basketball, you want to turn the tables on them and avoid those trendy teams.

My competitive edge comes from looking beyond surface-level “efficiency” analytics to uncover the deeper stories behind why teams outperform or underperform their statistical profiles. The #1 most important consideration is to view college basketball as a small-sample exercise in a Bayesian world. We start with priors—such as the preseason AP poll and KenPom rankings—and update them based on in-season results. Many analysts mistakenly rely 100% on the data (season results), ignoring the importance of incorporating priors into their posterior beliefs about team strength. That approach throws away valuable information.

This concept is fundamental to how you should view NCAA brackets. It might seem like one of those spurious correlations you should ignore, but it’s actually rooted in a sharp Bayesian understanding of college basketball. The streak will be broken at some point—St. John’s or Michigan State could do it by reaching the Final Four this year—but it still offers critical insights. For example, it flags Florida (preseason #21 AP / #26 preseason KenPom) and, to a lesser extent, Texas Tech (preseason unranked AP / preseason #12 KenPom) as teams to be wary of. On the flip side, it suggests that teams like UNC, Kansas, UConn, and Baylor—who underperformed their preseason expectations—are more likely to succeed in the tournament.

This also extends to conference strength. Teams from conferences that were stronger than expected during the regular season (like the SEC this year) tend to underperform in March, while teams from conferences that were weaker than expected (like the ACC and Big East) tend to overperform. The reason? Cross-conference results early in the season provide only a small sample of data, and failing to account for regression to the prior can lead to misjudging team strength.

Other key factors I consider:

• Health: Are there teams that struggled with injuries but are finally healthy?

• Experience vs. Talent: Have highly talented but less experienced teams improved significantly over the past two months?

• Schedule Strength: Did some teams inflate their efficiency metrics by dominating weak opponents while struggling against tougher competition?

• Key Absences: Are there teams missing a key player due to injury—one who won’t return for the tournament?

• Rotational Changes: Have lineup adjustments significantly altered a team’s performance?

• Matchup Factors: Does a team have an especially favorable/unfavorable draw?

• Shooting Variance: Have opponents hit an unsustainably high or low percentage of jump shots or free throws, distorting perceptions of a team’s true strength?

Finally, for the tiebreaker score, a good rule of thumb is to enter 144 (e.g., 76-68), which is considered the optimal total. If your title pick is a high-scoring team like Alabama, Auburn, Florida, Illinois, Missouri, or Gonzaga, adjust the total upward. For lower-scoring teams like Houston, Texas A&M, or Tennessee, adjust it downward. The tiebreaker matters, especially in large pools. Massive proportion of $ winning situations are going to include tiebreakers. Don’t overlook it!

SECTION 3: Basketball Thoughts

I’ll start with two teams where my analysis differs significantly from both the analytical consensus and the market (Vegas/Polymarket betting odds). I’m willing to take a stand on these teams—not just because my original gut view is different, but because the more I look the more I believe others are overlooking something important. I’m factoring in the widespread disagreement and still confident that my perspective is the right one.

HATE: St. John’s (2)

I’m much lower on St. John’s than both the market and other analysts, but this is one of the rare teams I’m very willing to take a strong stance against. Their draw is also brutal—Kansas is an incredibly strong 7-seed.

As I mentioned earlier, college basketball is a small-sample exercise where priors matter, and St. John’s is attempting to break a long-standing trend: 39 straight teams that have started the season unranked and earned a 1 or 2 seed have failed to reach the Final Four. That alone is a red flag. But even worse, they’ve played a soft schedule and haven’t beaten a single Top 25 KenPom team all year (unlike MSU which dominated the 2nd best conference in the country and returned a team who’s underlying analytics last year were actually pretty decent and were probably systemically underrated in the preseason, but I digress). So what do we actually have here? A team that (1) wasn’t expected to be elite coming into the season and (2) hasn’t proven they can beat top-tier competition. Yes, they’re physical, great defensively, and force a ton of turnovers—but teams like this often inflate their ratings by dominating weaker opponents while struggling against better ones. St. John’s has yet to prove they can beat high-level teams, and I hate their chances of making a deep run. I wouldn’t pick them past the Elite Eight under any circumstances. In larger pools, you could even justify taking Kansas over them early, though in smaller pools, that’s riskier since St. John’s is still the more likely team to advance.

LOVE: Texas Tech (3)

With ISU down, Tech is probably the best dark horse in the field that I want my readers to be in on. I think Texas Tech is the 5th best team in the country (behind the four one seeds) and even better they have a soft S16 draw against St. John’s and E8 draw against Auburn. To start this section on Tech, check out this comparison detailing the three best players in college basketball this year on a per minute basis.

You’ve probably heard of Player A (Cooper Flagg) and maybe even Player B (Johni Broome), but Player C—JT Toppin—belongs in that conversation. He’s the best basketball player you haven’t heard of, yet somehow, at just 19 years old, he had only one Power 5 offer just a year and a half ago. The biggest difference between him and guys like Flagg or Broome isn’t talent—it’s playing time. He hasn’t been on the court as much this year, but when he’s healthy, he’s proven he can handle heavy minutes in big games (logging 33, 30, 34, 38, 30, 44, 34, 37, 32, and 35 in various matchups this season).

Texas Tech has been extremely banged up all year but still managed to perform well. Almost all of their key players have missed time, and Toppin himself played less than half of their minutes through their first 23 games due to a lower-body injury, an ejection, and other setbacks. Despite this, he’s an absolute force, and the supporting cast has held up surprisingly well in his absence. Now, with the team getting healthier in time for the tournament, they’re going to be exceptionally dangerous.

The only concern? Darrion Williams and Chance McMillian—likely Tech’s second- and third-best players—missed their last Big 12 Tournament game. While both are expected to play in the first round, their minutes may be limited. For Tech to seriously compete for a title, they’ll need them fully back and effective by Saturday’s key matchup. We may not get much more injury news before Thursday, but the fact that they were cleared to play four days before the 3-14 game is a strong sign for their availability moving forward.

One other thing to watch: when Tech has been short-handed in the backcourt, they’ve given Kerwin Walton significantly more minutes. His playing time has fluctuated from as little as 5, 7, 8, or 9 minutes to as much as 31 and 38 minutes in recent months. And frankly, I think he’s awful. First, the eye test tells me he’s a terrible defender and he doesn’t provide that much on offense besides making wide open threes at a slightly above average clip. Second, the analytics back up what I’m seeing—which is always a great feeling.

I’m hoping Williams and McMillian return so Tech can shift back to a seven-man rotation—something they’ve done often earlier in the season—with Toppin, Hawkins, Federiko, Anderson, and Overton, while essentially cutting Walton out entirely. That would be a major upgrade. Walton has played 44% of Tech’s minutes this season—only 13% fewer than Toppin—which is wild considering how much of a black hole he is on the court. If they tighten up the rotation and minimize his role, this team could take a huge step forward.

St John’s vs. Texas Tech

It’s a happy accident that the 2 seed I hate and the 3 I love could face off in a potential Sweet 16 matchup. That said, I want to be clear: I only give Texas Tech a 41% chance to make the Elite Eight and St. John’s 29%, meaning there’s a 30% chance I look like an idiot and a 40% chance I look like a genius—pretty even either way. The key edge here is that the public is overwhelmingly picking St. John’s. If you take Texas Tech and they advance, you’ll gain 8 points on a lot of people. Even though 41% is still below 50% (meaning it’s more likely to be wrong than right), it’s the safest choice in terms of expected value for that slot.

Love: Duke (1)

Pick them in every pool, no matter the size! They’re the best team in the country, and while plenty of people are choosing them, not enough to make me hesitate—and it shouldn’t stop you either!

Like: Houston (1)

A very strong team and deserving 1-seed, but they got unlucky with Gonzaga as their 8-seed and Duke as their opposing 1-seed. If you're picking them to get past Duke, you should go all the way and pick them to win the title. They’re the second-best title value pick after Duke in smaller pools (<100 entrants).

Meh: Florida/Auburn (1)

They’re on the opposite side of the bracket from Duke, and both are strong Final Four and runner-up picks (especially to Duke!). However, too many people are picking them to win the title, and there isn’t much reason to do so in most pools. In smaller pools (<100 entrants), Duke is a better option, while in larger pools (>100), Duke over a riskier team like Texas Tech, Maryland, Iowa State, Colorado State, or Louisville—or Houston over Auburn or Florida—makes more strategic sense.

Don’t Like: Alabama/MSU/Tennessee (2)

Small fades. They just aren’t good enough to be a final four pick in any pools basically. It is just a weak year for 2 seeds and the 1 seeds are all a tier ahead.

Hate: Kentucky (3)

Big fade. They play a style of basketball—strong offense, weak defense—that has historically been a negative predictor of tournament success, even when controlling for overall team strength. On top of that, their roster profile isn’t ideal: they’re extremely experienced but lack high-end NBA talent, which tends to be a bad combination for deep tournament runs. Injuries are another concern. Their fourth-best player, Jaxon Robinson, is out for the season, and their second-best player, Lamont Butler, has been dealing with persistent shoulder issues. He still looks banged up heading into the tournament, and even if he plays in Rounds 1 and 2, he likely won’t be at full strength. That said, their early path isn’t terrible—Illinois, Xavier, and Texas aren’t exactly world-beaters—so you can justify picking them through to the Sweet 16. But beyond that? No shot. Definitely not to the Elite Eight.

Hate: Memphis (5)

Memphis simply isn’t a very good basketball team. From a pure team quality standpoint, they shouldn’t even be in the tournament. In my ratings, they’re closest to Yale among teams in the field, and in KenPom, they’re sandwiched between two 10-seeds and actually rated lower than Northwestern—a team that’s gone 7-13 in its last 20 games. Yet, they’ve somehow built their resume on an unsustainable 10-1 record in games decided by six points or fewer. Even worse, their second-best player, Tyrese Hunter, will likely be out for at least the first weekend. And to make matters worse, they drew Colorado State—a team I’ll break down below…

Love: Colorado State (12)

Colorado State did something more teams should consider midway through the season: they asked, “What happens if we cut our two worst players out of the rotation?” The answer? They’re now 10-0 in their last 10 games and have covered in all 10. Since mid-February, they’ve been playing at the level of a 1-seed. Now, some of this surge has been fueled by unsustainably good shooting, so I’m not saying they’re actually a Top 10 team. But they’re almost certainly a Top 25 team right now. The key is that this isn’t just a random hot streak—it’s a direct result of cutting Jorgensen and Williams from the rotation, validating what the analytics had been saying all along (which isn’t always the case). On top of that, Nique Clifford is a legitimate Top 25 player in the country. For a mid-major, having the best player on the court in multiple tournament games is rare, but that’s exactly what CSU might have. They’re clearly favored over Memphis in the first round and pose a serious threat to Maryland in the second. That being said Maryland is pretty good and then Florida in the next round is quite good so it’s not an ideal draw beyond Memphis.

Hate: Michigan (5)

Michigan isn’t as good as their seed suggests—they’re more like a 7-seed in terms of actual team quality but have built an inflated resume by winning a lot of close games. Their 13-4 record in games decided by six or fewer points has made them the fourth-luckiest at-large team in the field (behind Memphis, Oregon, and Utah State). To make matters worse, they drew a brutal 12-seed in UCSD—an awful stylistic matchup for them. It almost feels like the committee is playing a joke on Michigan fans. Michigan ranks in the 9th percentile nationally in ball security—the worst among all at-large teams—while UCSD ranks in the 100th percentile in forcing turnovers—the best among all at-large teams. There’s strong empirical backing that high-pressure defenses thrive against teams that struggle to take care of the ball, and this is a textbook case. Because of that, I only give Michigan a 53% chance to win in the first round, which is terrible for a 5-12 matchup.

Like: Iowa State (3)

With the devastating news that Keshon Gilbert is out for the year, Iowa State's tournament outlook has dramatically shifted. What once looked like a potential 1 seed now faces a significant challenge. Iowa State was one of college basketball's most complete teams when fully healthy. The Cyclones demonstrated elite potential before and after Milan Momcilovic's 7-game absence, where the team clearly missed his 6-8 frame and premier spot-up shooting that creates crucial spacing.

However, losing Gilbert is a crushing blow. As Iowa State's best player for much of the season, his exceptional guard play before the groin strain was pivotal to their success. The Cyclones were already showing vulnerability with him missing 4 of their last 7 games, including that concerning loss to BYU in the Big 12 quarterfinals. This team had legitimate national title aspirations when healthy. While Tamin Lipsey's minor injury shouldn't be a long-term concern, Gilbert's absence creates a void that's nearly impossible to fill. The Cyclones remain dangerous, but they're definitely only a small value (they can be picked to the E8 over MSU some) where with Gilbert they would have been at the top of my list for dark horses to watch.

Love: Gonzaga (8)

Absolute smash as a Top 10 team stuck with an 8-seed. They have elite analytics, a strong talent profile, and have been on the wrong side of a lot of close games. It’s rare to see a team this good seeded this low—not just because they’re at full strength now, but because they’ve played at this level all season. The issue? Their second-round matchup is brutal, and I only give them a 20% chance to make the Sweet 16. But in large pools (300+ people), taking a shot on them upsetting Houston and making a deep run—even to the Final Four—could be a great high-upside play. Most brackets will have them exiting early, so if they pull it off, you’ll gain a ton of points on the field. I generally don’t like Cinderella runs for teams on the right side of the bracket (since Duke and Houston are there and are better values than Auburn/Florida), but Gonzaga is one of the two exceptions (along with Arizona).

Like: Arizona/Maryland (4)

Arizona and Maryland are strong 4-seeds, but both face tough Sweet 16 matchups against the two best teams in the country, Duke and Florida. In larger pools (100+ people), you can justify taking them on deep runs to the Final Four or even the title game. And in really large pools (500+ people), they’re my second and third favorite dark horse title picks—assuming you’re not rolling with Texas Tech. As I mentioned in the Gonzaga section, I slightly prefer Arizona as a team, but since they’re on the right side of the bracket, I’d lean toward Duke/Maryland title games over Florida/Arizona in bigger pools.

Like: Louisville (8)

Louisville is a really, really good 8-seed—more like a 4- or 5-seed in actual team quality. But unlike some of my other under-the-radar picks, everyone who follows college basketball closely already knows this. Their underseeding is brutal, and while I think most of the committee’s decisions were fine (including leaving WVU out), this one is a complete head-scratcher. That said, picking them over Creighton in the first round is a must. Creighton is solid, but Louisville is clearly favored (~60% chance to win) and is still only being picked in about half of all brackets. In really large pools (500+ entries), they also have Cinderella potential for a Final Four or even a title run. And I love that they’re on the left side of the bracket, which makes that kind of deep run even more viable.

Love: North Carolina (11, play-in)

We’ll see tonight if they win the play-in—they’ve got about a ⅔ chance to do so. If they do… as I mentioned in the math section, it’s crucial to factor in start-of-season priors when evaluating teams. UNC has both a strong talent profile and great recent form—they were ranked #9 in the country preseason and have played at a Top 14 level over their last nine games. That’s an incredible profile for an 11-seed. It’s funny—many people were upset about them getting into the tournament, yet they’d actually be favored in their 11 vs. 6 matchup against Ole Miss. In large pools (300+ people), it might even be worth taking them to the Elite Eight since their 3- and 2-seed matchups are relatively weak. And in really large pools (500+ people), making a Final Four or even title game run is totally reasonable—especially since they’re a left-side team.

SECTION 4: Full Model Probabilities

(assumes UNC and Xavier win play-in games, they are the stronger/more notable teams. don’t pick SDSU to go anywhere if UNC goes down)

SECTION 5: Bonus Yale Thoughts

Since I go to Yale and many readers will be Yale-affiliated, I’ll share my thoughts on Yale’s first- and potential second-round matchups. Yale is an excellent mid-major team and arguably the best Ivy League squad in the last eight years. Bez Mbeng, John Poulakidas, and Nick Townsend are all extremely strong players, and we drew the weakest 4 seed in Texas A&M. That said, I’m concerned that TAMU is a tough stylistic matchup for Yale.

Texas A&M is an offensively unskilled but extremely athletic and physical team that plays elite defense and dominates the boards—especially the offensive glass. There’s plenty of past data that shows teams with this statistical profile tend to struggle more against high-level teams but overwhelm weaker opponents, which isn’t ideal for Yale in the first round. TAMU generates the highest percentage of its offense in the country from second-chance points, and Yale relies on four wings (Mbeng, Townsend, Isaac Celiscar, and Casey Simmons) ranging from 6’4” to 6’7” for much of its defensive rebounding. While not having a true big lead your defensive rebounding works in the Ivy League, Townsend and Celiscar will be undersized vertically by a couple inches against TAMU’s bigs, and Simmons will be down 40 pounds on his matchup.

Yale also isn’t well-equipped to exploit one of TAMU’s main weaknesses—turnovers. The Aggies are a turnover-prone team (5th percentile in live-ball turnovers), but Yale doesn’t force many turnovers (16th percentile in forcing live-ball turnovers), which means they’ll have to get stops through defensive rebounding instead. That’s a problem against a team that thrives on offensive boards. Center Samson Aletan, Yale’s elite shot-blocker and rim protector, does not contribute much as a defensive rebounder because he often prioritizes help-side defense and shot-blocking over rebounding. The Yale coaching staff will likely need to adjust his approach against TAMU or take the risk of relying on the wings to rebound at a decent rate on the weak side.

The one stylistic silver lining for Yale is that TAMU runs much of its offense through two high-volume low efficiency guards, Wade Taylor and Zhuric Phelps. Mbeng, the Ivy League Defensive Player of the Year, is an elite perimeter defender who matches up well against them.

For Yale to pull off the upset, four things need to happen:

1. They need to take and make more threes, as scoring inside will be tough against TAMU’s elite interior defense. TAMU allows a higher percentage of points from beyond the arc than 98% of teams nationally, making outside shooting a key factor. Poulakidas, Yale’s only high-volume three-point shooter, will be especially crucial—he needs to get hot.

2. Mbeng must not only lock down Taylor and Phelps, forcing them into inefficient shooting nights (think 5-for-17), but do so without as much interior help Yale’s perimeter defenders have relied on from Aletan all season.

3. Yale has to hold its own on the defensive boards, which requires Aletan adjusting his playstyle to some degree or Townsend/Simmons somehow holding their own on the against stronger and taller opponents. Yale has rebounded 74% of their opponents missed shots this year, and the goal is to just hit 60% against TAMU.

4. Yale needs to take care of the ball, something they’ve done well this season. However, TAMU’s athleticism and defensive intensity will be at a level Yale hasn’t encountered before. Given how ineffective TAMU is offensively, Yale cannot afford to let them generate easy points off live-ball turnovers and runouts.

My model gives Yale just a 24% chance to win, which is solid for a 4-13 game. That said, if Yale pulls off the upset, the second-round draw is promising. Michigan is a weak 5-seed, and their struggles against teams with a legitimate big to guard Goldin and a strong wing defender like Mbeng on Wolf play right into Yale’s strengths.

This will end up being a bit of a long read, but I’m tossing some plots/maps/figures in to summarize key info, so I recommend scanning through as looking at them if you want something shorter.

I’ll break this piece into sections, talking about the 6 different groups of voters and how they shifted, and then about some random stuff at the end.

1. Hispanic Voters

The biggest story of this election cycle is the mass exodus of Hispanic voters from the Democratic Party. With Hispanics as a growing voter bloc, the long-term implications are huge. In one night, they dashed a decade-long Democratic dream of flipping Texas, cemented Florida’s status as a solidly red state, and helped deliver Nevada and Arizona to the GOP. Beyond the direct electoral consequences, it’s also a bit of a rhetorical gut punch for the Democratic base. Since Summer 2020, much of the party’s evolution has focused on progressive racial messaging and the expansion of DEI initiatives. Yet, this shift has coincided with significant defections among voters of color, many of whom have moved toward Trump—ultimately prompting a reevaluation of these policies by corporations and the executive branch in the weeks after Trump’s inauguration.

We knew this was happening to some extent on election night. The question now is: what else have we learned over the last few months? For starters, we’ve got some hard estimates on the magnitude of the Hispanic shift to the right—and it’s remarkably consistent across different subgroups. Cubans, Mexicans, and Puerto Ricans all moved sharply to the right by a pretty similar amount, and I’ve seen convincing anecdotal evidence that Venezuelans and Central Americans followed suit. Regionally, the trend held pretty steady across the board. Among the nine areas where I’ve generated strong estimates, urban Ohio (Columbus, Cleveland, Cincinnati) saw the smallest shift, but even there, Hispanics swung 18 points to the right.

Nationally, Hispanics went from D+42 in 2016 to D+26 in 2020 and my back of the napkin math based on the areas I’ve looked at has them ending up around D+5 in 2024.

We’re also starting to see the early signs of educational polarization among Hispanic voters. Historically, there’s been a U-shaped relationship between education levels and Democratic support among Hispanics in northern/western cities—Hispanics in both the least and most educated areas were the most Democratic. But that pattern is beginning to break down, with the shifts now looking pretty linear as Democratic support declines the most with the lowest education voters and vice versa.

2. Asian Voters

Asian voters also swung sharply to the right in 2024, dropping from D+34 in 2020 to around D+15 in 2024 based on my best estimates. While much of the rightward shift among Black and Hispanic voters can be attributed to growing educational polarization, Asian voters serve as an important reminder that even highly educated minority groups are increasingly leaving the Democratic Party in droves.

I’m usually one of those people who love to criticize the broad umbrella categorization of “Asians” and instead focus on the distinct ethnic subgroups for this type of demographic electoral analysis. That said, 2024 was actually a big win for the “umbrella” approach. Nearly every subgroup, aside from Vietnamese voters, showed the same general trend: a hard rightward shift towards Trump. Still, it’s both interesting and productive to break that shift down by subgroup, so I’ll do that below.

Among Asian subgroups, predominantly Muslim South Asian communities—especially Bangladeshi and Pakistani voters—registered the largest rightward shift from 2020 to 2024. It’s likely that Muslim Indian voters followed a similar trend, although I don’t have sufficient data to confirm this. Vietnamese Americans saw the smallest rightward shift, with some precincts in Vietnamese-heavy areas of Orange County, CA, even moving left. However, this shift is somewhat deceptive—Vietnamese Americans had already made a substantial rightward shift between 2016 and 2020 while other Asian groups didn’t, positioning them ahead of this broader trend. Korean, non-Muslim Indian, and Chinese American voters all swung significantly to the right, with working-class Koreans and Chinese driving the largest shifts. Precincts in Palisades Park, NJ (heavily Korean), and Sunset Park, NYC (heavily Chinese), were particularly difficult for Harris for example. Filipinos also shifted notably to the right, which is especially significant given that they are by far the largest Asian subgroup in Nevada—a key battleground state—making up about half of the state’s Asian population. While it’s pretty tight among these groups in terms of total swing, my best summary is that high-education, non-Muslim Indians shifted right more than other high-education Asians, though still less than working-class Koreans and Chinese, who experienced extreme rightward shifts.

Regionally, the shifts align with expectations. They are much larger in areas with more working-class Asians, such as NYC and Chicago, and somewhat larger in places with higher South Asian populations, like Northern Virginia, Columbus, and Raleigh. In California, where the Asian population is predominantly affluent Korean and Chinese—especially in the Southland—the shifts to the right are noticeably smaller.

A map of Asian shifts in San Francisco nicely highlights growing educational polarization among Chinese voters. Wealthy Asians in the Inner Richmond district barely shifted toward Trump for example, while middle-class Asians in the city’s southeast corner swung heavily to the right.

3. Cosmopolitan White Voters

Another major story from this election is how well Democrats performed with highly educated, cosmopolitan white voters. In particular, Democrats excelled with white city dwellers. While urban areas overall trended right, that shift was largely driven by minority voters. Among white urban voters, Harris actually performed quite well.

Certain “resist lib” enclaves somehow managed to become even more Democratic in 2024. These are the quintessential urban upper-class liberal bubble neighborhoods—places like Park Slope in Brooklyn, where moms sip $8 matcha lattes on their way to spin class, chatting about Montessori schools and how their CSA box was particularly fresh this week, or Noe Valley in San Francisco, where dads compare Tesla configurations, brag about their sourdough starters, and debate which boutique wine bar feels the most “authentic.” Despite Biden winning over 95% of the vote in some of these areas in 2020, White voters there still shifted even further to the left.

Meanwhile, Democratic success in affluent white suburbs was highly variable. In some suburbs, Democrats struggled, while in others, they exceeded expectations. Democrats saw notable challenges in two types of areas:

1. Ultra-wealthy suburbs, like Beverly Hills and Greenwich, which shifted significantly to the right.

2. Heavily “white ethnic” suburbs with large Irish, Italian, Jewish, or Portuguese populations—such as parts of Long Island—where voters also moved right.

But in many suburbs, Democrats did exceptionally well with white voters. Take the Columbus suburbs, for example. Worthington just passed Columbus in Democratic support, and at this rate, Westerville is on track to be more Democratic-leaning than Columbus by 2032. It’s genuinely mind-blowing, even as someone who’s only followed politics closely for a decade or so. The same trend holds across much of the Midwest and beyond—suburbs around Omaha, Kansas City, St. Louis, Milwaukee, Indianapolis, Louisville, Cincinnati, Raleigh, Pittsburgh and Charlotte all showed similar patterns. Worthington and Westerville are perfect examples of the type of place that Democrats are doing great in—white voters around the 80th national income percentile are trending hard to the left. Once you move above that into ultra-wealthy areas like Greenwich, CT; Alpine, NJ; Sands Point, NY; and Beverly Hills, CA, Trump actually performed quite well. But in upper-middle-class suburbs, Democrats are doing well.

4. Racial Depolarization in General

We’re reaching a point where white voters in many affluent suburbs are more pro-Democratic than Hispanic and Asian voters. In Northern Virginia’s DC suburbs, for example, white voters collectively backed Harris at higher rates than people of color in the same region. Aside from Black voters, White voters were actually the most Democratic group.

5. Less Cosmopolitan White Voters

Republicans made big gains in historically Democratic areas that have been trending GOP for years, but 2024 was the year the bottom truly fell out for Democrats in a lot of these places. The Driftless Area in northeast Iowa and southwest Wisconsin, the coalfields of West Virginia and Kentucky, and southern white rurals and small towns that had already been shifting right all swung hard to the GOP. Nationally, Trump clearly performed better with white voters in 2024 than in 2020, driven largely by these kinds of areas. Lawrence County, AL, is a perfect example—already deep red and somehow getting even a lot redder.

6. Black Voters

It’s not a great time for Democrats with Black voters, especially when thinking about support on a logit scale rather than a raw unit scale1. Still, relative to expectations, this was actually one of the few bright spots for Democrats on election night. Pre-election polling crosstabs and media coverage made it seem like collapsing Black support would define 2024, but Black voters turned out to be far more resilient than expected. Instead, the real story of 2024 was the massive Hispanic and Asian shifts to the right.

Nationally, Democratic support among Black voters looks like it’ll drop from about 90% to 86%—an 8-point shift to the right. On a linear scale, that’s not especially alarming, considering the overall national shift was about 6 points to the right. But in logit space, it’s much worse. If I had to guess, I’d expect Black Democratic support to keep dropping for a bit and eventually stabilize around 80%, but honestly, it’s hard to know.

7. Ethnic White Voters

The Northeast was a disaster for Democrats and Harris, and she particularly struggled with White voters. While Democrats held up reasonably well with White voters in the Midwest and West and did passably in the South (outside of historically Democratic areas), the Northeast stood out for its sharp rightward trends among Portuguese, Irish, Italian, and Jewish voters. Long Island and the Lower Hudson Valley were brutal for Harris, as was South Jersey. Massachusetts was, if anything, even worse than New York relative to my expectation. What’s puzzling is that this shift with white ethnic voters in the Northeast occurred across both lower-education and higher-education areas. It feels like there’s something deeper driving it, but I don’t have a well-formed theory for why just yet.

8. Harris Over-performing in Swing States

I’ve written about this at length very recently, but it's a wacky development. Not clear why it happened or whether it will stick, but it's one of the better pieces I’ve written in the past year so obviously I gotta plug it here since it’s a notable situation.

9. Moderate Democratic House Candidates

This tweet sums up my thoughts on the issue perfectly. It highlights something pretty notable about the 2024 elections, and I’m honestly surprised it hasn’t been a bigger part of the discourse. The effect size is substantial and highly statistically significant.

It Wasn’t Supposed to be This Way

Five years ago, after Democrats secured sweeping victories in suburbs nationwide during the 2018 midterms—and appeared poised to replicate this success in the 2020 presidential election—the media was filled with stories about the growing political influence of Hispanic and Asian voters moving into suburban areas and shifting them blue.

This New York Times story by Sabrina Tavernise and Robert Gebeloff was a good read at the time, I remember enjoying it while bored during high school on the day it came out and it always stuck with me as outlining the biggest political theme of 2008-2018. In summary, Northern Virginia (NoVA) was once a stronghold for affluent white Republicans. However, between 1990 and 2018, an influx of Hispanic and Asian voters transformed the region, shifting it decisively blue. NoVA became a vibrant symbol of the new, diverse Democratic coalition, expanding from urban areas to suburbs.

The Wikipedia page on the Great Replacement theory in America describes it as a “conspiracy theory,” which is a fair characterization. However, it did stem from two very real trends: (1) the United States was becoming increasingly diverse over time, and (2) immigration helped Democrats as they were consistently improving their performance with these voters. Republicans diagnosed these problems in the much famed “RNC Autopsy” of 2013. It warned that the growing non-white population in the United States could make it increasingly challenging for Republicans to win elections unless the party made efforts to engage with these voters. Trump won in 2016, sure, but his victory just relied on boosting GOP support among white voters to another level. However, this did little to address the GOP’s long-term challenges with non-white voters, who continued to grow as a share of the electorate.

So, what’s happened since 2016? Affluent white voters in Northern Virginia have become increasingly liberal, while Republicans have started to win over both affluent and working class Asian and Hispanic voters. By 2024, we reached an inflection point: Harris won a higher percent of white voters than non-white voters in Northern Virginia. The realignment of political coalitions in 2024 was striking and dramatic, and it will likely take years for people to adjust to the idea that Democrats are becoming less and less of the party of minorities and more and more of the party of elite white voters over time. In 2016, only about 10% of Trump’s voters were people of color, but by 2024, that number had grown to around 25%.

Estimates

Modeling suggests that Harris actually outperformed Biden’s 2020 performance with white voters in the Northern Virginia area. However, she faced significant challenges with Black voters and saw sharp declines in support among Asian and Hispanic voters.

When analyzing county-level data, this result might seem surprising, as Harris actually gained ground in very few areas. However, if you look at the precinct-level data, it makes more sense. Harris flipped Purcellville, outperformed Biden by 8 points in McLean, and ran it up in the Prince William County exurbs near Dale City. Basically, she ran particularly strongly in the whitest areas of Northern Virginia.

The plots below illustrate the modeled raw margin shifts by race, normalized for the overall turnout decline in the region. Harris made significant gains among well-educated, affluent white voters in the inner-ring suburbs, particularly in areas like McLean and Vienna. Meanwhile, Trump improved in Prince William County and held steady in Arlington and inner Loudoun County. Trump made gains among Black voters across the region, though the overall impact was limited due to the smaller Black population. In the diversifying suburbs around Dale City, Harris increased the Black margin primarily due to growth in the raw Black population. Democrats suffered heavy losses with Hispanic and Asian voters across the region, particularly in eastern Loudoun County and central Fairfax County. Middle-class Vietnamese voters in the Springfield area shifted away from Democrats significantly, as did affluent Indian voters in Loudoun County and affluent Korean voters around Centerville.

Ecological Inference

Ecological inference (EI) is a statistical method used to infer individual-level behavior or characteristics from aggregate-level data, particularly when direct individual data is unavailable. In this case, the central question revolves around how different demographic groups voted in the 2024 election. While exact data isn’t available, EI allows us to estimate voting patterns using precinct-level data. The graphs below illustrate the core concept of EI, which examines the relationship between racial demographics and vote percentages across precincts to uncover correlations.

The estimates above are based on my recently completed Statistics Senior Thesis at Yale, which focused on developing a new method for ecological inference. You can read the full thesis here for a detailed explanation of the methodology. All the estimates presented above are based on election and demographic data sourced from Dave’s Redistricting App, which outlines its data sources here.

Local Estimate Disclaimer

It’s important to emphasize that these are not national estimates. The populations of White, Hispanic, Asian, and Black residents in Northern Virginia (NoVA) differ significantly from their national counterparts. For instance, the highly educated white voters in NoVA are far more liberal than white voters nationally and likely represent one of the strongest regions for Democratic gains among white voters in the country—comparable to areas like the inner northern suburbs of Salt Lake City, Utah, and the inner western suburbs of Milwaukee, Wisconsin.

Similarly, Black voters in NoVA are markedly different from Black voters nationwide. They tend to be more affluent and highly educated, which actually makes them more conservative. Inner-city Black voters, who tend to be Democrats’ strongest supporters, shifted rightward by a smaller margin. Based on my preliminary estimates, inner-city Black voters were about D+90 compared to D+73 for NoVA Black voters in 2020, and D+83 compared to D+60 for NoVA Black voters in 2024.

The Hispanic and Asian voter samples in NoVA are somewhat more similar to their national counterparts, particularly in how they voted in 2024, but notable differences remain.

Virginia Precinct Data Disclaimer

In Virginia, absentee ballots (AB) are reported at the countywide level. Results collection services then allocate these ballots to specific precincts, often using their own methodologies. For the 2020 election, I relied on estimates from the Harvard VEST dataset, which describes its approach for Virginia as follows:

"Absentee ballots and provisional votes were reported at the county or city level throughout the state. These were distributed by candidate to precincts based on their share of the precinct-level reported vote."

This methodology is widely regarded as reliable, but there were known issues in Prince William County (PWC). Specifically, the allocation process led to an overestimation of Democratic support in the western parts of the county and an overestimation of Republican support in the eastern parts. These discrepancies can be addressed using a simple linear regression model (R² of approximately 0.96) to recalibrate the 2020 results based on data from the 2018 Senate election and the 2021 gubernatorial election using precinct results from surrounding areas. By applying this approach in PWC, I significantly improved the accuracy of the allocations.

To further investigate potential precinct aggregation errors in PWC’s 2020 data, I analyzed the correlations between precinct-level Democratic and Republican vote shares in the 2018 Senate election and the 2020 presidential election results. Compared to other counties in Northern Virginia, PWC exhibited notably weaker correlations, suggesting potential aggregation issues. Specifically, the correlation for Democratic vote number was 0.613 in PWC versus 0.771 in neighboring counties, while the correlation for Republican vote number was 0.743 in PWC versus 0.918 elsewhere.

This R+0.2% bias highlights a significant shift from the Electoral College’s biases in previous elections, which were R+2.9% in 2016 (when Trump beat Clinton despite losing the popular vote) and R+3.8% in 2020, reflecting a decline in the GOP's structural advantage in national electoral dynamics. One way to understand this contraction is to note that, although Trump’s overall national performance improved by about 6% in 2024, his margin only improved by 2.4% in the tipping-point state—amounting to a 3.6% leftward trend in Pennsylvania relative to the country.

Why did this happen? In simplest terms, Republicans have been trading heavily white suburban and exurban voters (who are overrepresented in today’s battleground states) for Hispanic and Asian voters clustered in places like Florida, Texas, New York, and California. Because none of those big states are core battlegrounds at this point in time, this new coalition proved much less electorally efficient for the GOP.

Below are the three primary reasons why the GOP’s structural advantage collapsed between 2020 and 2024—from a R+3.8% to just R+0.2%. I also estimate the share of that 3.6-point drop attributable to each factor.

1. Random Variation in State Partisan Leans

In 2020, the electoral map aligned extraordinarily well for Republicans, as a large number of medium-sized states ended up 2–5% more Republican than the nation. Of course, each state’s partisan lean relative to the country moves around from cycle to cycle. Controlling for the national environment, these state-by-state shifts have historically followed roughly a normal distribution with mean 0 and a standard deviation of 2.9%.

To see how much the GOP’s 3.8% bias might drop simply by normal flow of state partisan leans from cycle to cycle, we can perform a simulation using the 2020 results as a baseline. For each simulation, we add a single draw from a normal distribution with a mean of 0 and a standard deviation of 2.9% to each state’s partisan lean, recalculate the Electoral College outcome, and then find the tipping point state and the corresponding structural bias. We then average over all simulations to find the mean expected bias for 2024. The result is R+3.3%, which represents a drop of 0.5% from the initial 3.8%—despite 2020 Census electoral vote reapportionment alone suggesting a move from R+3.8% up to R+4.0%. Hence, purely random shifts among the states would have contributed roughly 0.7% toward shrinking the GOP’s Electoral College Bias edge, accounting for about 19% of the total decline (3.8% down to 0.2%).

2. Increasing Demographic Inefficiency of the GOP Coalition

After 2012, Republicans thrived in the Electoral College by winning over ancestrally Democratic, white working-class voters in crucial Northern states like Pennsylvania, Ohio, Wisconsin, and Michigan. Meanwhile, Democrats were piling up high margins in big coastal states—California, New York, Maryland, Massachusetts, Washington—where diverse urban cores and highly educated suburbs concentrated the Democratic vote.

However, this changed in 2024. Predominantly white areas, especially those with higher education levels, trended left. At the same time, Hispanic and Asian voters trended decisively to the right, fueling huge Republican gains in states such as Florida, Texas, New York, and California. However, those states were nowhere near the tipping point—making the GOP’s gains in them largely ‘wasted’ from an Electoral College perspective.

To estimate how demographic changes alone reduced the GOP’s advantage, I developed a county-level retrodictive regression model to predict 2024 results based on 2020 results and demographic data (including race, education, income, and population density). I then aggregated these county-level predictions into state-level projections. Overall, these retrodictions performed well, explaining 72% of the variance in state-level partisan lean shifts from 2020 to 2024. Using these state-level projections, I estimated a mean Electoral College bias of R+2.0% for 2024, both with and without the inclusion of a small amount of noise. By comparison, the “random-variation” simulation had already reduced the bias to R+3.3%. This additional drop to R+2.0% suggests that demographic shifts accounted for a further 1.3% reduction in the GOP’s advantage—representing approximately 36% of the total decrease in Electoral College bias from 2020 to 2024.

3. Non-Demographic Harris Overperformance in Battlegrounds

Harris generally overperformed the demographic model’s expectations in key swing states. Of the seven major battlegrounds, four ranked among her top six state-level overperformances. This pivotal dynamic neutralized much of the GOP’s remaining systemic edge, setting up a notable realignment in presidential politics.

One significant factor behind these battleground-specific overperformances was inter-state migration patterns, as detailed in a fantastic NYT piece by Ronda Kaysen and Ethan Singer. In states like Pennsylvania, Michigan, Wisconsin, and Georgia, new movers to the state were more Democratic than the populations they joined; in Arizona, however, new movers were 26% more Republican than existing residents. Across all seven battleground states (weighted by electoral votes), these incoming voters collectively leaned 6% more Democratic than their new neighbors. Harris underperformed the demographic model in Michigan (struggles with Muslim voters in Wayne County plus lingering problems in some white working-class areas) and in Arizona (GOP-friendly migration).

When I attempt to explain Vice President Harris’s resounding success in key swing states such as Georgia, North Carolina, Wisconsin, and Pennsylvania I come up with two primary factors. First, her campaign’s substantial financial advantage enabled a highly coordinated strategy that combined targeted advertising with robust ground operations, which proved particularly effective in these pivotal states. This advantage not only allowed for precise voter outreach but also contributed to improved Black voter turnout relative to 2020 in these key states, especially when compared to lower turnout in states such as South Carolina, Mississippi, and Alabama. This disparity highlights the strategic deployment of resources to mobilize key demographic groups in areas where their participation would have the greatest impact. Second, Harris’s supporters, who generally exhibited more moderate levels of enthusiasm compared to the fervor seen among Trump voters in polling, may have been especially motivated to turn out in battleground states. This heightened engagement likely stemmed from the perception that their votes would carry significant weight in determining the outcome of the election, underscoring the critical role of these states in shaping the broader electoral landscape. Either of these factors may be the entire reason, or maybe I’m missing something and there’s a third reason, but these are the two that comes to mind for me.

Overall, her net battleground overperformance was enough to drive the actual 2024 Electoral College Bias down to R+0.2%, rather than the R+2.0 predicted by demographics. This means that overperformance in key states beyond demographic expectations explains 50% of the total decrease.

Future of the Electoral College Bias

In 2024, the Electoral College advantage for Republicans hit a low point of R+0.2%. Yet one must note that the map aligned quite favorably for Harris, especially compared to 2020’s extremely GOP-friendly environment. If one assumes state-level random noise for each state partisan lean with mean 0 and standard deviation 2.9% for 2028, the projected bias moves higher to around R+0.6%, although any reversion of Harris’s state-specific overperformance could push it up more.

Then there is 2032. Modeling two cycles’ worth of random shifting (with an effective standard deviation of about 4.1% by 2032) suggests an expected GOP bias of about R+0.7 if we continued using the 2020-era electoral vote distribution. However, Republicans will benefit greatly from the 2030 Census reapportionment. States Trump carried in 2024 are currently expected to gain nine electoral votes on net, while Harris-won states will lose nine, worth roughly +0.7% in bias under present conditions. If the GOP begins the 2032 cycle at R+0.7%, the reapportionment effect will boost it to around R+1.4%.

At the same time, demographic trends may continue eroding this Republican advantage, particularly given that many decisive battlegrounds (like those in the Upper Midwest) remain heavily white and reasonably well-educated. Most analysts expect Democrats to keep gaining ground with white voters relative to the nation as a whole over the coming decade.

In short, after 2024’s extraordinarily tight outcome, the Electoral College bias for Republicans—once as high as 3.8%—is now much narrower. Future cycles will bring further state-by-state fluctuation, continued demographic changes, and reapportionment-induced realignments. Given those layers of volatility, neither party can bank on a sizable Electoral College advantage in cycles to come, but it is certainly possible that one develops.

Republican Gains

Before diving into specific trends or notable observations, it’s essential to keep in mind that Trump improved his performance across nearly every demographic and throughout the country compared to 2020.

Nationally, we’re looking at about a 6-point shift in Trump’s favor. For any county or group to show a shift toward Democrats, they would have to ‘trend’ (meaning shift relative to the national average) by at least 7 points toward the Democrats—a substantial shift. Still, examining which groups trended toward Democrats or Republicans is fascinating, as it reveals potential coalitions for future elections, such as in 2026, when the national map may resemble 2020 more closely.

White Voters

Heading into election night, Democrats were hoping to offset minority losses by making gains among white voters, particularly well-educated whites. But that shift didn’t materialize; in fact, white voters swung hard toward Trump.

The only notable exceptions were in historically Republican, well-educated northern suburbs, where Democrats made slight inroads compared to 2020. Areas like Waukesha, WI; Carmel, IN; Overland Park, KS; and Westerville, OH—some of the whitest, most affluent, and best-educated parts of the country—trended 7–10 points toward Democrats. These places are going to come out with a vengeance in 2026, but it will be very interesting to see how blue they get.

More diverse suburbs, like Somerset County, NJ; DuPage County, IL; and Collin and Fort Bend Counties in Texas, were disappointing for Democrats, a stark contrast to their strong performance in these areas over the past decade. While we’re still waiting on precinct-level data for confirmation, it I’m guessing the predominantly white areas remained fairly stable (similar to trends in Carmel and Westerville), while the more diverse precincts, particularly those with Asian and Hispanic populations, showed significant declines for Democrats.

White voters in working-class towns, suburbs, and cities abandoned Democrats en masse. In 2020, Biden managed to stop the bleeding, but 2024 has seen a realignment on par with 2016, with Harris experiencing a sweeping loss among working-class white voters in densely populated areas. Trump flipped Bucks County in Pennsylvania and surged in Macomb County, Michigan.

Democrats suffered major losses among rural white voters. This was especially harmless to Democrats in the general Driftless Area, which hurts Democrats in a battleground state (Wisconsin) and in key congressional districts (WI-03 and IA-01). Many areas swung nearly as far to the right as the nation overall, which is remarkable given that they were already close to Republican saturation. And it wasn’t just about vote share—turnout surged in these areas as well. As I talked about on NewsNation on Thursday, rural voters turned out at a much higher rate than urban voters in 2024 compared to 2020. This was especially noticeable in Pennsylvania and Ohio, where each county did not shift that much towards Republicans but the statewide margin still shifted a lot since Republicans composed a much higher proportion of the electorate.

Hispanic Voters

Hispanic voters shifted dramatically towards Trump for the second consecutive election. Exit polls show that Trump won Hispanic men nationally, a remarkable shift. Miami-Dade went from a Clinton +30 county in 2016 to a Trump +13 county in 2024. Trump flipped Webb, Hidalgo, and Cameron counties in the Rio Grande Valley, as well as every county in California's Central Valley and Inland Empire. His support surged in Hispanic communities in New York City and Chicago. Hispanic voters now appear set to become a lasting part of the Republican coalition and a critical swing voting bloc nationally, much like Catholic voters.

Trump’s coalition in 2016 was overwhelmingly white—about 90% of his supporters were white voters, reflecting his strong appeal among white working-class and rural voters. However, by 2024, the composition of his coalition had significantly diversified, with the share of white voters dropping to around 75%. The share of minority voters in his base has nearly tripled in just eight years, with Hispanic voters playing a major part in this.

Black Voters

Democrats have spent the year worried about Black voters switching to Trump, but in 2024, that concern turned out to be unfounded. While precinct-level data is still needed for precise numbers, Democrats held steady with Black voters overall. In fact, Harris appears to have done slightly better with Black voters than Biden did in 2020, as seen in predominantly Black counties like Coahoma County, MS. Most of the Democratic losses in Philadelphia seem to be from Hispanic voters in the city’s Near Northeast, while in Wayne County, it appears to be from Arab voters.

The real challenge for Democrats with Black voters wasn’t support but turnout. Black voters remain the backbone of the Democratic Party and the most consistent voting bloc in the country for either party. While Pennsylvania’s 29 rural counties saw a 2% increase in votes compared to 2020, Philadelphia’s turnout fell by 5%, with even sharper drops in predominantly Black neighborhoods. Despite Harris’s campaign having virtually unlimited funding for ground operations and turnout drives, the results are sobering for Democrats.

Jewish and Muslim Voters

Ironically, both Jewish and Muslim voters shifted sharply to the right in this election, largely influenced by the Gaza conflict. In Dearborn, Michigan—the largest Arab-American community in the U.S.—Harris’s performance was disastrous (it shifted from Biden+50 to Trump+6 with Jill Stein winning 18% of the vote). In some precincts, Harris even placed third behind Trump and Stein. Interestingly, Stein may have drawn more votes from Trump than from Harris in Dearborn, contrary to expectations.

This shift among Muslim voters makes sense when considering that, while Muslim Americans tend to hold socially conservative views, they have historically leaned Democratic due to a perception of Republican hostility toward Muslims. With many now viewing the Democratic Party’s stance on the Gaza conflict as equally unwelcoming, they gravitated toward Trump as the more conservative option if they didn’t choose Stein.

Meanwhile, Democratic losses among Jewish voters were also significant. Many Jewish Americans, perceiving Republicans as more staunchly “pro-Israel,” turned right as well. In NY-17, Republican Mike Lawler effectively captured the support of swing-voter Jews in the Hudson Valley, and Rockland County saw a pronounced shift toward Trump at the top of the ticket. This dual shift among both Muslim and Jewish voters highlights the unexpected electoral impact of the Gaza conflict across communities with typically different political alignments.

Asian Voters

Asian voters are a growing part of the electorate and have leaned strongly toward Democrats for most of recent history. While we're still awaiting precinct-level data to confirm the trends, preliminary county and district data suggest that Democratic support among Asian voters has significantly declined—almost as much as it has among Hispanic voters. Initial results show that in NY-6, a majority-Asian district in Queens (most notably including Flushing), shifted massively from a Biden +30 in 2020 to Harris +5 in 2024. Similarly, results in Asian communities in New Jersey and California look equally concerning for Democrats.

You always need to wait for precinct data to really figure out how Asians shifted (since there aren’t any exclusively or majority Asian counties) but it won’t be good for Democrats. Based on the initial county and district-level data, the shift toward Trump among working- and middle-class Asian voters is extreme, but even the shift among affluent Asians in the Bay Area looks like it will be a bit larger than the national average.

Florida Amendments Failure (Repro. Rights and Pot Go Down)

Since Florida is a bit of a second home state to me, I want to address Tuesday’s results, which have pushed Florida Democrats to the brink of political hopelessness. Democrats were wiped out across the state. It’s remarkable that, in just four years, Florida has shifted from a battleground where Biden was favored to win to a state with as conservative governance to anywhere in the country.

All of Florida’s liberal ballot amendments failed, including significant ones on abortion and marijuana legalization, leaving the state entirely in the hands of Republican lawmakers with virtually no checks and balances. In other red states, Democrats have had some success using constitutional amendments to pass popular liberal policies (most recently with Ohio voting in favor of marijuana and abortion in 2023 after preventing Republicans from raising the threshold for ballot issues to 60%). However, the challenge in Florida is the 60% supermajority requirement to pass amendments—a rule that ironically passed with just 57% simple majority support in 2006 as backlash against a ridiculous pregnant pigs amendment on the Florida ballot in 2002. This threshold ultimately stymied progressive measures that might have otherwise had strong public support.

Florida Republicans have recently discussed raising the amendment threshold even higher, which would effectively dismantle the citizen-initiated constitutional amendment process. Conversely, if Florida Democrats are smart, they should invest substantial resources into a 2026 amendment campaign to lower the threshold back to 50%. The 2026 midterms, likely to take place in a favorable environment for Democrats under a Trump administration, would be a uniquely good opportunity to get something like that passed.

If successful, Democrats could then follow Ohio Democrat’s approach by introducing amendments for issues like abortion and marijuana legalization only needing a simple majority—two major policy areas increasingly decided at the state level. However, if Democrats fail to get this threshold amendment on the ballot in 2026, or if it doesn’t pass, Florida liberals are pretty screwed. With heavily gerrymandered state House and Senate maps and a strong Republican tilt statewide, the pathway back to competitiveness would be severely obstructed.

Ticket-splitting

If you're a Democrat and think Tuesday was rough, you’re absolutely right. But it could have been much worse. Had Americans voted for Republicans in Congress at the same rate they voted for Donald Trump, the GOP would have secured 57 seats in the Senate and likely around 230 in the House—a commanding majority. Instead, the GOP will end up with around 53 seats in the Senate and 220–221 in the House. Expect to hear a lot more from moderate Republicans over the next two years. Trump will hold only the narrowest of trifectas, and no one can really predict what’s coming in the years ahead.

The GOP’s overperformance suggests one of three things: (1) Donald Trump is a strong candidate, (2) Kamala Harris is a weak candidate, or (3) Democrats generally overperform in down-ballot elections because they field better-funded, higher-quality candidates. Differentiating the first two through data is challenging since they’re interconnected, but both likely hold some truth. Trump clearly has a unique appeal among Hispanic voters, driving turnout for him, while Harris seems particularly weak with middle-class and affluent suburban voters, as well as less cosmopolitan white voters.

There are some striking results to consider. For example, Zach Nunn (super strong GOP moderate candidate) didn’t outperform Trump in Polk and Dallas counties, likely due to Harris’s unpopularity there. In New Hampshire, Harris trailed behind Goodlander and Pappas by 4.5% and particularly struggled in more educated areas—New Hampshire, one of the least Trump-friendly states, saw swing voters simply preferring anyone over Harris at the top of the ticket.

The third point also has some validity: Democrats notably outraised Republicans in competitive House races. Yet even in races like Michigan’s Senate contest, where funding and candidate quality were closely matched, Democrats still saw roughly a 2% down-ballot boost, underscoring overperformance across the ticket.

Final Prediction Review

I released my final predictions a couple of weeks ago, and I performed well (or, let’s say, got lucky) this year. I correctly predicted 49 out of 50 states in the presidential election, missing only Wisconsin, where I had picked Harris. Although she lost, it was her narrowest defeat among the seven battleground states, so I’ll take some comfort in that.

In the Senate, I went 33 for 34—no one would have reasonably predicted Casey’s loss. In the House, I look set to finish with around 423 correct out of 435, which I’m very pleased with. Some of my calls missed the mark (MT-01 and NE-02 flopped), but a few of my upset picks, like PA-10, WI-3, and CA-47, turned out to be closer than expected. 

Overall, I got about 97% of the races right, as well as 6 out of 7 presidential battlegrounds and 10 out of 11 competitive Senate races. From a topline perspective I got within 10 electoral votes, 1 senate seat, and 3-4 seats in the US house.

2026 Senate Map

Democrats will need to flip North Carolina in 2026, but beyond that, viable pickup opportunities are scarce. It would be surprising if Collins loses in Maine; she’ll likely have plenty of chances to oppose Trump and build her credentials in his administration. As for other possibilities, three likely Republican states—Iowa, Kansas, and Ohio—will be up for grabs, with Ohio being the most promising.

Ohio could be winnable if the GOP nominee is anyone but Matt Dolan (considering Vance won a relatively close race in the Biden midterm). Sherrod Brown could even run if he chooses, but he’s 71 so I think there’s a pretty good chance he retires. These races would be competitive in a D+8 environment, like 2018, but without a strong Democratic wave, flipping any of these seats will be challenging. Realistically, Democrats might reach 52 seats at best—or 51 if they manage to unseat Collins.

Democratic Losses in Liberal States

It’s no surprise to me that Democratic losses were especially concentrated in liberal states. Abortion rights were the only major issue that played strongly in Democrats’ favor, so in states where abortion rights were already protected by liberal state governments, this issue was much less salient. Consequently, voters in these states shifted more toward Republicans, as Democrats' strongest issue held less sway.

International Elections

Some would say Democrats have been the best-performing incumbent party in a major economy since 2021. Macron’s party in France suffered major losses, Canadian Liberals are in a precarious position (about to get crushed), the UK’s Tories got destroyed, and even Modi’s party in India experienced setbacks. The economic strain from the 2021 COVID recovery has been challenging globally, and the U.S. isn’t the only country facing inflation issues. Incumbent parties worldwide are feeling the effects, and Democrats are actually doing better than most of them.

I'll focus primarily on the presidential race (because let's be honest, that's what most of you are here for), but I'll drop some down-ballot insights at the bottom.

All times are Eastern.

The Basics

The first results will start coming in at 6 PM, with key results expected by 7 PM. The busiest period will likely be from 7 PM to 3 AM. There’s about a 70% chance we’ll know the winner by 3 AM, but if not, it could take several days to reach a final call.

The outcome of 43 states at the presidential level is largely irrelevant going into Election Day, seven battleground states will ultimately decide the Electoral College. I’ll focus primarily on these key states, as well as states that report results early. If Trump wins Minnesota or Harris flips Iowa, it likely won’t matter much since either candidate would have already secured a dominant performance, reaching 270 Electoral Votes anyways. Here’s a (very rough) map of states Harris or Trump are favored in or those that are considered very close.

Follow results on Twitter @DecisionDeskHQ and online here.

Presidency: Indiana and Kentucky

The first results will be reported after 6 PM with poll closings in Indiana and Kentucky. While neither state is competitive (Trump will dominate both), their early results could provide valuable clues about national trends. If either candidate significantly exceeds expectations in key areas that report quickly, it could indicate broader momentum, as overperformance in one states has a strong correlation with overperformance in other states.

Presidency: Florida

Florida, the first competitive state to fully report, closes at 7 PM and counts votes really fast. Roughly 65% of votes—mainly early and absentee ballots that could slightly favor either Democrats—are counted within 30 minutes. A race call should come right at 8 PM if Trump performs well, as the western panhandle closes. Florida's efficient reporting provides an early glimpse of the national picture, making it a focal point of analysis.

While Trump’s win in Florida signaled strength in 2016 and 2020, Harris was never expected to compete here that seriously in 2024. If she wins Florida, she has a 99% chance of winning the presidency. Trump is projected to win by about 7.2%, so a tighter margin would be good for Harris. 9:00 AM update: Election Day turnout looks so bad for Harris in Florida so she’ll probably lose by more, but this is probably isolated to the Sunshine State.

I’ll especially be interested in results in…

Northwest Jacksonville (Black voters):

Early returns from Duval County’s inner-city Black precincts could signal national Black support. If Trump exceeds 8% here, he may reach 15% nationally among Black voters—this would be rough for Democrats.

South Tampa (Affluent White suburb):

An ancestral Republican area, South Tampa’s more normal age demographics than the rest of Florida mean it’s a valuable signal for affluent suburban shifts. A 5+ point gain for Harris here would indicate similar trends nationwide.

Osceola County (Puerto Ricans voters): 

Puerto Ricans have been under a lot of discussion since the “garbage” comment from comedian Tony Hinchcliffe at a Trump rally. Unfortunately for the Trump campaign, Pennsylvania has a massive population of Puerto Ricans in Reading, Allentown, and Bethlehem. Osceola County also has a massive PR population, which will be predictive of Trump’s performance in PA.

Presidency: North Carolina and Georgia

Polls in GA and NC close at 7 and 7:30 PM, respectively, and should report votes quickly (especially in Georgia). Georgia, thanks to new legislation requiring absentee and early votes to be counted within an hour, will be the first key states that reports a lot of votes. North Carolina follows a similar pattern but faces a potential hour-long delay in counting early votes due to new state laws. 

Absentee and early votes (combined) are expected to slightly favor Trump in GA and slightly favor Harris in NC. If NC only reports absentee ballots first however (which I expect because of a new state law postponing the count of EV), before EV, Harris could build a massive lead across NC in the first 20 minutes, before she comes back down to earth.

Trump will gain in both states in the middle of the counting process, as Election Day votes from more rural areas are counted. Harris should close especially hard in Georgia since the final votes counted are from in and around the Atlanta urban area.

Ohio closes at 7:30 PM as well, which most notably has a key senate race in Sherrod Brown vs. Bernie Moreno. Early results (AB/EV is reported first) could look pretty good for Harris.

Presidency: Pennsylvania

Pennsylvania closes at 8 PM and is the most important state in this election. If Harris is off to a bad start in the Sun Belt (Florida, North Carolina, and Georgia) she will need to win Pennsylvania to keep her hopes alive.

Early results are likely to heavily favor Harris due to urban absentee votes being counted early. Then, Trump will gain and most likely take a lead as rural areas report from 9pm to midnight. 

Unlike North Carolina and Georgia, Pennsylvania will not count all of its votes on election night. Harris is expected to gain substantially (by about 2%) on Wednesday, meaning she can be losing by a considerable margin in Pennsylvania at 4 AM and still win the state.

Presidency: Michigan and Wisconsin

If PA, GA, and NC “split” the election will mostly come down to MI and WI.

These Midwest battlegrounds close at 8 and 9 PM and count votes pretty slowly.

Wisconsin typically finishes counting around 2-3 AM ET. A crucial and heavily Democratic batch of Milwaukee City absentee votes, dropping between 1 and 3 am, could swing the result. Harris should gain about 60k votes on Trump in that drop, so she can make up a pretty sizable deficit late.

Michigan counts more slowly and less transparently, with Wayne County (Detroit) mostly reporting between 11 PM and 4 AM. Wayne County is a massive proportion of the state population and is quite Democratic, so she can make up a lot of margin late in Michigan too.

Both states should fully complete their counts by Wednesday night, but both should be called by decision desks in the early hours of Wednesday morning.

Presidency: Nevada and Arizona

Don't expect a quick call in either state. Arizona's first results will report an hour after 9 PM ET polls close, and should slightly lean towards Harris compared to final results, but Trump is expected to gain ground as Election Day votes are counted. Nevada, closing at 10 PM ET, won't release any results until all polling places are clear, and only about 70% of votes will be counted on Election Night. 

If the election comes down to Nevada and Arizona, there’s a good chance it takes multiple days, and maybe even as long as a week to declare a winner.

US Senate

The Senate battlegrounds largely mirror the presidential map, with two additional crucial races. Ohio's contest between Brown and Moreno should be called by midnight - if Brown trails by less than 2% with 80% reporting, he's likely to win as remaining urban votes typically favor Democrats. Montana's race could stretch into Wednesday, with Republicans expected to gain about 3% after Election Night concludes.

US House

US House control won't be determined on Election Night. Roughly 25 key districts will remain uncalled on Wednesday morning. This delay stems from slow counting in states like California, Arizona, Washington, and Oregon. The final balance of power might not be known for several weeks after Election Day.

Before I launch in, I'd like to emphasize that I generally avoid trying to making these types of predictions. A key principle in my approach to election forecasting and communication is acknowledging and highlighting the inherent uncertainty involved in the process. I still stand by my probabilistic forecasts you can find here, which will give you a better understanding for the range of outcomes that is possible to happen on Election Night than my “predictions” below. I also urge you to read my Election Night Guide article coming in a week or so, which will make you the most knowledgeable person in the room about what’s going down on November 5th (and the week after) and is probably more of a value add than my predictions below. And, follow Decision Desk HQ on Twitter, for fast and reliable analysis on Election Night.

Still, everyone thinks it's fun to see some predictions and I’m sick of people groaning at me when they ask what’s going to happen and I say “idk man it's pretty close” so without further ado let's dive in. 

Trump Wins the Presidency and Clears 300 Electoral Votes

Another Disclaimer: I think experts basically don’t have any more of an idea about what’s going to happen on Election Day than anyone else, so I don’t want to put my prediction on a pedestal because who the hell knows. I think Kalshi nails the current situation; 58-42 Trump is almost exactly what my priors are heading into Election Night. If I’m really splitting hairs I slightly disagree with some of the state markets (they are slightly underrating her in the Sun Belt and slightly overrating her in the Rust Belt, relative to current polling. Polling errors have tended to be better for Democrats in the Sun Belt than Rust Belt this decade, so it’s unclear why they’re going in the other direction, but I digress).

My gut is with Trump as well, and I have him winning 6 of 7 key battleground states in my prediction. I think it's going to be an election of radical racial depolarization, and I think the Dem turnout machine is operating really well in Wisconsin right now post Dobbs, so I think she holds onto the Badger State.

I’ll say Trump wins AZ by 3, FL by 7, GA by 3, MI by 0, NC by 3, NV by 2, TX by 8 and PA by 2.

Republicans Win the Senate and Flip all the Red States

I think Republicans hold FL and TX and flip MT by pretty healthy margins, which gives them 51 seats and control of the chamber.

No one cares that much about the seat AFTER Republicans clinch control, but the most interesting US Senate race in 2024 is obviously Sherrod Brown in Ohio. I’m increasingly optimistic about Kamala Harris’s prospects in Ohio, which could be a major boost for Brown. While I think Harris will lose by about 9-10%, that’s still a slight leftward shift compared to the country as a whole and at least gives Brown some chance, halting two consecutive rightward trends in Ohio. I believe the bleeding for Democrats in SE Ohio has mostly stopped, as it’s hard to imagine Republicans doing even better in Appalachia. Meanwhile, about a quarter of the state is made up of highly suburban areas that I expect to swing left again.

One reason I’m not as confident in Brown’s chances of significantly outperforming Harris is Trump’s unpopularity in these suburbs. It’s hard to picture Brown outpacing Harris by much in the suburban areas or in the inner-city Black communities, which make up around 10% of the electorate. For example, I expect Harris to win Westerville (a white, upper-middle-class Columbus suburb) by 23% and Mason (a southwest Ohio suburb) by 1%. I can’t see Brown winning these suburbs by the 33% and 11% margins he would need. In 2022, despite Democratic gains in suburbs nationwide, Westerville and Mason actually shifted 2% to the right compared to the state overall, even with a right-wing populist like JD Vance on the ballot.

My prediction: Trump wins Ohio by 9.5%, and Brown loses by 2.5%. If Brown manages to keep the race close, it will likely be because Trump performs poorly in the suburbs, preventing Brown from needing a large overperformance.

Looking beyond Ohio, I’d say there’s a 50% chance Republicans win at least one of Wisconsin, Michigan, Pennsylvania, Nevada, or Arizona. I’m skeptical of the polling that suggests significant split-ticket voting in these states. If Trump wins Wisconsin or Michigan by 2-3%, that could be enough to push Hovde and Slotkin to victory. On the other hand, I’m not convinced by the hype around McCormick in Pennsylvania; calling it a tossup seems a bit too optimistic. It would still be a massive upset if Casey goes down.

I’d be shocked if moderate Republican Larry Hogan flips Maryland or if Independent Dan Osborn wins in Nebraska, but both will likely overperform fundamentals by a ton and lose by around 8-12%. I think Osborn will come closer than Hogan, as Nebraska is less Republican than Maryland is Democratic.

I’m not overjoyed about the (lack of) transparency surrounding Osborn’s internal polling that shows him ahead. The main reason we think this race is close is internal polling from Osborn, and it doesn’t at all meet the standards I look for from a methodology, crosstabs, and presidential toplines standpoint. It’s also an odd situation for Fischer, since releasing internal polling showing her ahead by 15 points wouldn’t help her fundraising efforts. So we’re left in a bit of a gray area in terms of reliable internal polling, and we have very limited public polling of the race.

Democrats Narrowly Clinch the House, Avoiding a GOP Trifecta

In a similar manner to 2020, I have Democrats really struggling in NY and CA, really hurting their math to flip control of the chamber. However, I have them performing quite well across the rest of the country, and flipping barely enough seats in NY and CA (CA-27, CA-45, NY-03, and NY-22) to win the chamber.

As I said above, I think there’s going to be a racial alignment where Republicans do really well with Hispanic and Black voters (almost flipping TX-34, and winning NC-01, CA-13, and CA-22) and Democrats gain a lot with White voters, especially secular and middle/higher education white voters in the Great Plains area (winning MT-01, IA-02, and WI-03).

Of course, this map is a bit aggressive and has some upsets for kicks and giggle (CA-47 going red), so I encourage readers to take these with a grain of salt and investigate my probabilistic house predictions more.

I’ll use this entry of D&D to share the “Zachary Election Season BLOG” which you can access at tinyurl.com/zd2024 for the next few weeks. You can get access to my thoughts in a pretty live fashion from this doc before, during, and after Election Night.

I’ll attach today’s entry below for a taste of the insightful and concise updates I’ll be sharing the next couple weeks.

10/22

Republicans are ripping the roof off in early voting. Disregard most early voting discourse, but I have the good stuff. The trick is to stratify by past turnout, which can be done effectively in states like Arizona and Nevada. Republican early vote turnout is high in both the Sun Belt and Rust Belt, which was largely expected, though perhaps not at this magnitude. Two key questions arise from this data regarding how much it increases Trump’s chance to win: (a) will this surge continue through the rest of the early voting period? There's already some indication it may have been a first-week phenomenon that has already tapered off in today’s data; and (b) how much of this early voting strength is simply cannibalizing GOP Election Day votes? While this is typically the case, the data still looks strong for Republicans even after accounting for recent voting history in AZ/NV (GOP beating Dems with lower propensity voters). Although this is clearly a positive sign for Republicans, it still doesn’t push Trump’s odds much past 60% in my mental priors. It’s important to remember that just two years ago, Republicans had an even stronger turnout nationwide than what we’ll see in 2024, yet Democrats still managed to win Senate races in key states through persuasion.