I think Alaska and Maine will be the tightest ones, with Texas and Michigan not far behind.
Mary Peltola will win in Alaska without it going to second choice ballots. Is that overconfidence or just a rare (for me) strong opinion?
Re: Texas. I’ve been following the voting trends for over ten years and you could see the direction in which it was moving. I always thought 2028 was the absolute earliest that Dems would win a state wide election and it may yet not happen until later than that. But never let yourself get discouraged – Texas will be purple in the decade to come.
Here’s the question I asked a few AI’s:
“In what 21st century federal elections, excluding presidential election years, has polling underreported national GOP support.”
In this nearly meaningless sample size of six, I do not see support for the “polling underreports MAGA/GOP support” hypothesis. Your hypothesis does get support if presidential election years are included, but I do not see why to do that.
EDIT: It was a little unfair for me to use the phrase “sample size” above. It’s not a sample – it is all of them. So if there was a strong indication with, say, four of six pointing the same way, I should take it seriously. But we at most have three of six.
So, anybody who brings a complaint about voter fraud in blue districts will get a sympathetic ear. Red zones not so much.
For a long time it has been a known issue that a presidential election draws in voters who’d likely sit things out. So that changes overall behavior.
Osborn in NE has about 50/50 chance now, mostly because of Trump. A small percentage of GOP are not happy. His second plus is INDEPENDENT. I can see voters going for that.
My review reports:
2006: polls underestimated Democrats. In competitive Senate races, Democrats did about 2.25 points better than polls predicted. Silver also counts 2006 among the cycles where polls were biased against Democrats.
• 2010: polls overstated Republicans. GOP Senate candidates did about 2.65 points worse than expected, and the polls skewed Republican in all but two competitive races. Nevada and Colorado were the big misses.
• 2014: the largest miss. In close Senate races, Republicans beat the polls by about 5.3 points, and by nearly 2 points in governor races. HuffPollster called it the worst collective Senate polling miss since 1998. Suspected causes included late-breaking undecideds and flawed likely-voter models.
• 2018: little average bias. Nate Cohn found the polls were biased toward Democrats by only 0.4 points, the least since 2006. Misses like Indiana's Senate race didn't follow a partisan pattern.
• 2022: the most accurate cycle on record, with regional errors. FiveThirtyEight found more accuracy than any cycle since at least 1998, with almost no bias toward either party. But Axios found Republicans did 7 to 8 points better than polls in Florida, while Georgia polls were off by under a point. Pennsylvania polls, by contrast, underestimated Democrats
So last several cycles per Claude last two midterms cycles pretty good. One huge miss underestimating GOP performance. And one miss each way by two to three.
For what it is worth 2025 was a pretty big miss underestimating Democrats. I don’t think you can be confident if there will be a polling miss and in what direction based on the data. Polls do want to be right and they will adjust if they are off. Sometimes they will get more accurate sometimes not, and sometimes they overcorrect. The mean senate result is probably around 53 senate seats for the Democrats. If we have a Trump like miss then it gets close, though Democrats may still sneak out a majority. A miss for the Democrats and we are likely looking at 56 + seats.
The question was restricting to midterms specifically - which are their own beast.
That said yeah what does the special elections results foretell? Another underestimate of Democratic performance? Or an overcorrection of the turnout models swinging the other way? Or correcting just right?
I don’t pretend to know. I however do accept that of the last five midterms only two were pretty much on the money and three were off by more than two, and one of those by more than five. No consistent direction. So we should be shocked if there is a significant systemic correlated polling error or confident what direction it would fall if it happens.
Jesus Christ, Cenk, you’re not helping.
The ad comes from Rebellion PAC, which is fronted by left-wing political commentator Cenk Uygur. It depicts a couple in a supermarket saying they cannot afford groceries because they have to set money aside “for killing Palestinians.” It concludes by stating that the Michigan Democrat’s GOP opponent “is for they/them,” referring to Israelis.
“Why does Mike Rogers care so much about what Israelis want and so little about what you want?” the new ad asks. A voiceover then riffs on Republican ads from 2024 that attacked Democrats over LGBTQ issues.
“Mike Rogers is for they/them,” the voiceover states over photos of Israeli politicians Benjamin Netanyahu, Itamar Ben-Gvir and Bezalel Smotrich. “Abdul El-Sayed is for you.”
It makes sense when you remember he’s a right-winger in left wing clothing.
I think the ad will do an excellent job of reaching undecided voters in Michigan whose main issue of concern is Jews stealing their bananas and previously could not guess which candidate in the election is the better voice of the “Jews stole my bananas” movement.
It’s even dumber - Cenk recycled an ad he made from a '24:
2022 had a very accurate generic ballot but head to head polling was garbage. Much like this year a bunch of partisan pollsters (Quantus, InsiderAdvantage,Co/Efficient, Trafalgar) flooded the averages. So people were looking at averages and seeing Oz, Masters, Walker, etc overestimated by about 5 points.
So two comments to that:
The garbage partisan pollsters is why it’s better to look at an aggregator who ranks the pollsters and weights them, ignoring some even, than just a rolling polling average.
But more to the point - individual crappy polls are a separate item than correlated systemic errors; the latter was the concern under discussion.
It feels like you’re selling a one size fits all polling error definition, and I think that sacrifices accuracy for the illusion of uniformity.
There are correlated errors like democrats being generically overrated by 7ish points in 2020, that was apparent in places like Montana, Iowa, South Carolina, etc where competitive races turned out to be very comfortable wins.
Then there are district/state specific misses, like polling that showed Hillary (and Dems) with wide leads in the blue wall states but it didn’t really miss on the generic ballot or other close states like Florida or Arizona.
Now if the stance is “Polling errors are only uniform misses that can be applied to the generic ballot and individual races across the country and then conveniently dropped into a spreadsheet, and everything else is just a an issue of due diligence by third party aggregators.” fine.
But if that’s the case that these kinds of errors are exceedingly rare.
Systemic error in a cycle does not eliminate individual race variations: they are tide in or out on top of the individual race choppiness.
Individual race variation can be house effects, sample size , so on.
Correlated systemic errors are based on the different nonpartisan polling houses clustering on a specific model for likely voters and corrections for non response bias, weighting the responses in similar ways that lead to similar errors across the board on average. That happens to at least two points one way or the other fairly frequently.
Maybe think of as climate vs weather?
I’m in Kansas City, MO so get a shit ton of KS Senate ads. I think the GOP polling must be worrying them: In addition to bringing Trump in to campaign, they are trying to tie Hamilton to a pedophile scandal years ago at a church camp. From what I can tell, the response from the church was disappointingly slow, but they did do the right thing eventually. The Hamilton camp has responded with ads pointing out the GOP guy voted against releasing the Epstein files. I’m very surprised the GOP is bringing up pedophilia.
As to the metropolitan area - the Kansas side of the KC metro is basically two counties: Wyandotte (relatively poor, strongly Democratic) and Johnson (wealthy, traditionally ‘traditional’ Republicans). Johnson has been trending purple, though, and is what elected their Democratic governor twice (with the help of appalling GOP candidates).
VP Vance was in Topeka yesterday, which was his second visit to Kansas in the past few weeks.
More tellingly, DJT will be in the small city of Salina this Saturday, his second visit to the state in 12 days. I’m really not sure why Salina (population 46k) was chosen, other than the fact that it’s a city in the middle of farm country. I suspect that Trumpy will make additional promises to farmers during his speech this weekend.
Nevertheless, multiple visits to ruby-red Kansas by both the prez and the veep in the same campaign season is extremely telling as to the GOP’s forecasting of the Senate races.