*A quick reminder that the Substack Midterms Hub is up and running and full of content from Slow Boring and many of this platform’s other fine outlets today — it’s a great place to go for updates, results, commentary, analysis and more. *
The fact that the outcomes of the midterm elections are not knowable in advance is a real challenge for editorial planning.
I’ve been asking if they’ll just tell me who wins under embargo so that I can plan things out, but everyone I ask insists that they can’t actually forecast the future. At best, well-informed people give me sophisticated takes about which early outcomes they’ll be looking at as indicators of the national mood.
For example, a strong performance for Democratic House candidates in the Miami area is probably good news for the Democratic House candidates in South Texas, according to one friend who believes that Latino subgroups differ more in their baseline partisan preferences than in how their voting patterns shift from election to election.
Other qualified professionals, however, say that’s wrong and heterogeneity is understated — so even here we’ll need to count the actual votes. Disappointing!
That being said, I’ve been doing this long enough that I’ve seen pundits make certain kinds of predictable election night *errors *cycle after cycle. These analytic errors are largely inherent to the geography of the United States of America and the state-by-state variation in voting patterns, both of which change at a slow enough pace that we can be pretty confident that we’ll see these mistakes in 2026, just as we did in 2022 and 2018.
So here’s my exclusive Slow Boring guide to what not to watch for in tonight’s results.
This one is huge. A fundamental challenge of polling is that survey response rates vary by demographic subgroup, so if you just naively put a survey out in the field, you’ll get a skewed sample.
If you’re trying to get a poll that tells you something about all American adults or all registered voters, you can fix this with demographic weights (the Census Bureau will just tell you what share of the population is white or over 65 or another variable of interest).
You still need to be a little bit careful. There are lots of questions about exactly which demographics matter and how to weight them. You can pretty easily make sure you have enough Latinos in your sample, but are you only getting English-dominant Latinos when you need Spanish-dominant ones? Do you have the right balance of Cubans vs. Puerto Ricans vs. South Americans in your Florida poll? It’s a hard business. But it is possible.
Polling likely voters is harder, though, because you don’t actually know who will vote. But you can at least develop an explicit model explaining how you’re defining the likely electorate.
The exit poll is a real problem, however, because you do know the election result but you do not know who voted.
It’s the opposite of an all-adults poll where you know what the demographic composition of the population is but are trying to find out who they voted for. Since the exit poll is just kind of guessing what the electorate looked like, it is spectacularly unsuited for the problem of figuring out the behavior of different demographic subgroups. Suppose Black turnout unexpectedly craters in Georgia and causes Jon Ossoff to lose his re-election bid. The exit pollsters will know that Ossoff lost but won’t know about the turnout crash. They’ll weight their survey to the outcome they know and to the electorate they were expecting and end up giving Ossoff too little support among both Black and white voters because they don’t realize that what actually happened is few Black voters showed up.
Months later, Census and voter file information will come out about who actually voted, but by then the narratives are baked.
In the immediate aftermath of an election, the best way to obtain demographic information is (unfortunately) to commit the ecological fallacy. If Democrats really overperform in the most heavily Asian precincts in America, that is decent evidence of a Democratic surge in Asian support perhaps driven by backlash to “The Indian Question” right. Stat-heads will point out that this is not a fully valid inference. Most Asian Americans do not live in heavily Asian precincts, and lots of non-Asian people do live in heavily Asian precincts. But it really is the best we can do since the exit polls are a mess.
The polls close really early in Kentucky.
Kentucky is obviously not a swing state. But there is one House district in Kentucky that is kinda sorta maybe, if you squint at it, competitive, largely due to the fact that Democrats have a seemingly good recruit in Zach Dembo. Dembo is on the Democratic Congressional Campaign Committee’s official “red to blue” list, and he’s also a Blue Dog PAC endorsee.
That being said, when you look at the demographics and partisan composition of this district, I don’t totally understand how it got to be red to blue.
Or, rather, I think you have to understand this as in part about the fact that there’s an incentive to not permanently write off whole states. Kentucky has a Democratic governor and a lot of people like Andy Beshear. You want guys like Dembo who have strong bios to feel they’ll be supported if they take a crack at longshot races. I think when it comes to long-term party building, investing in races like this is the correct call. But I also think the conjunction of the investments with the polls closing at 6:00 p.m. Eastern generates a maybe higher level of awareness of this one district than it actually merits. Even in a D+9 national environment, it’s a real stretch. Dembo will probably lose, and there’s no real significance in that.
Speaking of the D+9 national environment, any analysis of an individual race needs to be contextualized in terms of the overall national vote.
If a guy like Dembo loses a district like that by one point in a D+9 environment, that’s a good race. If he loses by one in what’s actually a D+11 environment, that’s less impressive for him but better for Democrats overall. If he loses by one in what’s actually a D+7 national environment, that’s very impressive for him but means that Democrats are underperforming their polling.
Unfortunately, though, we’re not really going to know anything about the national popular vote tonight. It’s not just that we’re not going to know when the polls close in Kentucky at 6:00 p.m. We’re still not going to know tomorrow morning.
The problem here is California.
California accepts mail ballots that are sent by Election Day, which means that many ballots will not arrive until days after post-election takes have been written. This mail situation reflects an understandable impulse to make it easy for people to vote, but California compounds the delay by counting the actual votes slowly. They keep getting away with this because California is not a swing state and everyone knows Democrats are going to win the big races there. But something like 12 percent of Americans — and a much larger share of Democrats — live in California, so the slow counting of California votes meaningfully distorts our understanding of the national political climate.
We’re going to know how Sherrod Brown and Abdul El-Sayed and Josh Turek and Roy Cooper and James Talarico did before we know what the national popular vote was.
What we do know is that the popular vote in California will be way bluer than the national average, so whatever the national climate looks like tonight, the reality will be bluer than that. This then means that in the short term, every individual Democratic candidate in the South, Midwest, and Northeast will look more impressive relative to the national climate than they actually were.
Florida is on the East Coast, its polls close early, and it counts votes fast. Also, back in 2018, Florida moved considerably to the right relative to the national average, meaning that Republicans did well in Florida’s elections that year. This led to an early evening narrative about Democrats’ disappointing performance and the forecasted “blue wave” failing to materialize.
In fact, the 2018 polls were extremely accurate, and the overall national wave was exactly what had been forecasted.
Polling is never perfectly accurate, but on the whole, 2018 was probably the single most accurately polled election year of my career. It just happened to be the case that Republicans did well in one state where the votes were counted early in primetime.
In retrospect, I built a lot of my worldview around the opinion that the 2018 midterm result was sneakily bad for Democrats. After all, they lost Senate seats. These losses happened in deep red states, so at the time a disappointing performance was chalked up to a “bad map.” But the 2026 map is also bad and the 2028 map will also be bad. Basically all the Senate maps are bad, so you can’t grade Democratic electoral performance in any given cycle on a curve. They either win (as I hope they will tonight) or they lose (as they did in 2018).
The practical upshot of the 2018 election was that Trump was able to keep governing the country, and I’m very concerned that a “good night” for Democrats in 2026 isn’t good enough — they need the Senate. But to that point, we won’t really know whether Democrats have a good enough night until we know what happened in Alaska, which happens to be in an inconvenient time zone for those of us on the East Coast.
Instead of falling prey to nonsense and fallacies, what you really ought to do is join the election night watch party that Slow Boring, The Argument, Silver Bulletin, and GD Politics are collaborating on. We’ll be liveblogging plus live on video for many, many hours, and we’ll have guests and accurate quantitative insights and good takes!
