Since The Argument and Split Ticket released our 2026 midterms model, I keep being asked the same question: “Isn’t that awfully optimistic for Democrats?”
You can see why:
I get it. After all, the legacy, nonpartisan election analysts at Inside Elections and Cook Political Report don’t agree with us. Inside Elections has Democrats projected to only pick up between two and 10 seats total. Their analysis has Democrats ahead in just 207 seats, while Republicans are favored in 212, with the rest considered toss-ups. And if you look at The Cook Political Report, you’ll also see Democrats trailing Republicans in the net count of seats for the House, 206 to 208.
Meanwhile, our model gives just 11% odds of a Republican majority.
In fact, our median scenario is that Democrats take 232 House races, for a net gain of 18 seats and a relatively comfortable 29-seat majority. We’re not the only ones — Nate Silver’s model actually has Democrats winning 234 seats, which is even more bullish than ours.
Cook and Inside Elections do leave many more races as toss-ups than we do. But that isn’t necessarily what’s driving their relative pessimism on Democrats’ gains. Even if Inside Elections gave every single toss-up race to the Democrats, it would still produce a forecast with nine more Republican seats than our model is currently projecting.
In other words, modelers see a blue wave election. The expert forecasts suggest more of a ripple. Come November, somebody is going to be very wrong. And I don’t think it will be me.
I don’t think legacy forecasters are properly accounting for how blue the current polling environment is. If you look at RealClearPolling’s comparison of midterm cycles, Democrats’ lead this year is nearly 9 points — slightly larger than at this point in 2018 and far above the 2022 edge.
Our model projects Democrats to win likely voters by 8 percentage points, which would exceed 2018’s margin by nearly 1 percentage point. At that level, the House of Representatives isn’t a toss-up; it’s safely Democratic.
Things could get even worse for Republicans: A recent survey from The New York Times clocked the 2026 electorate at D+8, with undecided voters overwhelmingly disapproving of the president.
But the experts at Inside Elections and The Cook Political Report have all the same information we do; our forecasts are different because our methods are different.
Models like ours are quantitative, while the expert analysis you get from places like Cook is more holistic and qualitative. Models essentially operate solely based on the hard data they get, while experts (and pundits) make judgment calls that are informed by the public and private polling they see.
The two approaches don’t react to data in the same way, for better or for worse. That means that, although polling has never looked rosier for Democrats, experts and pundits alike don’t necessarily believe it translates to a seat-by-seat sweep yet.
Their ratings show a skepticism of the polling blueshift actually manifesting in battleground seats where the majority is contested. There’s also disagreement over how much redistricting will blunt Democratic gains, despite Republicans being likely to not net any seats this cycle due to a combination of the blue environment and Democratic retaliation. In any case, you’re seeing some political experts openly wondering if Republicans will really get the type of wipeout that the models suggest.
In several individual races, our model and the analysts disagree substantially about Democrats’ chances.
That includes the 2nd Congressional Districts in Alabama and Maine — both of which our model rates as pure toss-ups, but which Inside Elections and Cook grade as Likely Republican. It also includes Texas’ Senate race, which Inside Elections and Cook grade as Lean Republican and toss-up, respectively, while we have it at Lean Democrat. (This is due to both the underlying race fundamentals, such as the disproportionate Democratic gains with Hispanics and Ken Paxton’s litany of scandals, as well as polling favoring Democrat James Talarico.) And it includes New Hampshire, a state whose vote share has been several percentage points more Democratic than the nation’s in each of the last two elections; Inside Elections puts it as a toss-up, while our model still pegs it at Likely Democrat.
That’s to say nothing of Georgia or North Carolina, both of which are rated merely as Tilt Democrat by Inside Elections, even though Democrats Jon Ossoff and Roy Cooper hold commanding leads in the polls (to the Republicans’ increasing despondency). Our model rates both states at Likely Democrat.
In my conversations with political insiders, I sense some skepticism of the modeled predictions. And it’s not just Republicans or journalists; a lot of Democrats I’ve spoken to, including pollsters and campaign operatives, simply don’t believe the numbers they’re getting. This is not unreasonable; we saw infamously poor polling in 2016 and 2020, which badly misled audiences. It’s fair for people to simply guard against that, because nobody wants to make the same mistake twice (or three times).
But polling has been really solid when Trump is not on the ballot. In 2017, 2018, 2022, and 2025, we didn’t see any track record of polling overestimating Democrats. This appears to be a phenomenon centered around low-propensity voters that show up only in presidential years.1 Relying on this in ratings or assessments of 2026 would require proof against the evidence we have.
It’s important to note that the experts aren’t all skeptical of the Democratic wave. The Center for Politics’ Sabato’s Crystal Ball is much more bearish on Republicans than the other experts, with Democrats leading in 210 seats, compared to Republicans leading in just 203. What’s more, its founder, Larry Sabato, has explicitly said that seat-by-seat ratings are understating Democratic gains, likening it to a “microscope vs telescope” analogy. In other words, Sabato thinks that while nobody disputes that Democrats are heavily favored, it’s easier to make a case for each individual race potentially bucking the wave.
I have a lot of respect for the election experts. They’re not stupid. They had a phenomenal 2024, giving more accurate predictions than every quantitative model in existence, even as they had more incorrect calls. And though it should be noted that in 2024, their ratings aligned very well with polling, it’s also important to remember that they also see a ton of internal polling that models simply don’t, which means they’re usually operating off of a richer dataset than the other election forecasters are.
But their ratings are, at the end of the day, still qualitative opinions on what the quantitative data means. And when you look at those numbers as they currently exist, I really believe there is no way to see this election as anything other than a blue wave.
Recommended reading:
Four things I changed my mind on after a year of polling
When The Argument‘s editor-in-chief Jerusalem Demsas and I first began talking about building a polling program, we were both obsessed with one problem: Issue polling was dominated by interest groups.
The public housing miracle that never happened
If you're going to build a policy agenda out of historical analogies, at least do the reading
It’s true that there have been recent high-profile misses in primary polling, but primaries are much, much harder to poll than higher-turnout midterms.
It's funny, on Nate Silver, Galen Druke and Clare Malone's new podcast Still Counting (check it out! It's classic 538 again) Galen asked Nate this same question and he said the race rankers have no balls. They would rather gradually shift their rankings over time so that the final result shows Dems winning a blue wave than actually call it beforehand. Because no one cares what they say 2 months before. But if they accurately predict the outcome a week before, that's a feather in their cap. Granted they have different incentives than modelers. But it's funny to hear almost the same thing here as well
Your 'model' is based primarily on poling. This is unfortunate since poling data is inherently flawed for several reasons but mainly due to significant non-response bias. Non-response rates in public opinion poling range from high to ridiculously high - in excess of 90%. Thus 'conclusions' based on poling are always unreliable. Yet, here at The Argument data from your own poling is frequently cited. Why should we pay attention?