Democrats are hanging their whole midterm pitch on a stack of polls that show their Senate candidates pulling ahead.
The people who run those polls swear their methods are airtight.
And Nate Silver just caught one famous pollster doing something even he couldn’t defend.
Nate Silver Took a Hatchet to a Big-Name Survey
Nate Silver made his name crunching election numbers at FiveThirtyEight, and nobody has ever mistaken him for a MAGA cheerleader. He doesn’t think much of Donald Trump.
So when Silver goes after a pollster, nobody can write it off as Republican sour grapes.
On his Still Counting podcast with co-host Clare Malone, Silver ripped into the newest Maine survey from the Marist Institute for Public Opinion. That poll covers two of the hottest races in the country: US Senator Susan Collins (R-ME) against Democrat Troy Jackson, and Democrat Hannah Pingree against Republican Bobby Charles for governor.
He didn’t mince words. “Obviously the poll is incoherent and kind of indefensible in one way or another,” Silver said.
Marist interviewed Maine voters from October 22 through October 25. Among registered voters, Marist had Jackson ahead of Collins 53 percent to 44 percent. Once Marist narrowed the pool to likely voters, that nine-point cushion shrank to two points, 49 to 47, inside a margin of error Marist itself put at 6.6 points.
The governor’s race lurched even harder. Pingree led Charles by 11 points among registered voters. Among likely voters, Marist had the two dead even at 44 percent apiece.
Same survey. Same four days. An 11-point swing depending on which column a reader happens to look at.
The Undecided Voters Who Multiplied Out of Thin Air
Gaps between registered and likely voters happen. But Silver spotted a detail that made the governor’s numbers look downright goofy.
He took it to X: “For the Gov race, Marist lists 2% undecideds among registered voters. But that jumps to 6% among likely voters! That literally doesn’t make sense.”
Here’s why that set him off. Likely voters are the folks who’ve mostly made up their minds. Somebody still shrugging at the end of October usually isn’t the type who bothers to show up.
“Usually what happens is they take a likely voter poll, you have fewer undecideds because people who are undecided at this point probably are not going to vote,” Silver said on the show.
But Marist’s numbers ran backwards. The undecided pile tripled.
And he wasn’t done. Silver also flagged that Marist’s likely voter group carried a bigger share of college graduates than its registered voter group did.
“That does not happen in midterms,” Silver declared. “This poll is indefensible.”
Marist Apparently Didn’t Trust Its Own Sample
Silver floated two theories. Either Marist’s likely voter algorithm has a bug in it, or the outfit is “scared of their own shadow.”
He leaned toward the second one. Silver said he thinks Marist was “terrified to publish their actual result” and cobbled together some contraption to make the final numbers look respectable.
Then came the real tell. “You should not be using likely voters to fix your problems with your sample,” Silver said.
He suggested Marist may believe its registered voter sample is busted, and he told them to repair the sample directly “instead of saying there’s going to be an 11-point swing, which is quite close to statistically impossible.”
Follow that logic for a second. Marist’s raw registered voter pool spat out fat Democrat leads in both races. Marist’s likely voter screen then dragged both races back toward the middle. A pollster doesn’t yank the steering wheel that hard unless something in the raw pile looks wrong.
And the numbers Marist steered away from were the ones Democrats liked.
Silver never said in so many words which way the problem leaned. Maybe it really is just a buggy formula, the way he allowed. But somebody at Marist clearly decided the registered voter results needed an awful lot of help before they could face the public.
That’s worth remembering, because Marist’s early October Maine poll counted only registered voters. That survey had Jackson up 51 to 47 and Pingree up by 12, and outlets across the country ran it as fresh proof the Democrats had momentum.
The Senate race does look close in other surveys. Suffolk University found Jackson up 49 to 47 among likely voters in mid-October, and CBS News/YouGov had a 50-50 tie earlier in the fall. So Marist’s two-point likely voter number lands in the right neighborhood, even if the machinery behind it rattles.
The governor’s tie sits all by itself. Suffolk had Pingree up 11, and a Fox News poll taken in mid-October had her up 10.
Pollsters Keep Tripping Over the Same Rug
Silver has griped about scared pollsters before. In the fall of 2024, he complained publicly that pollsters were herding, bunching their results together so nobody would stick out if the race broke funny. Then Ann Selzer’s final Iowa poll that year showed Kamala Harris ahead by three. Donald Trump carried Iowa by roughly 13.
Maine voters have heard this tune, too. In 2020, survey after survey had Collins trailing, and she won by more than eight points.
The trouble usually starts with who picks up the phone. College graduates tend to enjoy answering surveys. A retired logger up in Aroostook County who gets a call from a strange number during supper hangs up and goes back to his beans and franks. Pollsters try to weight their way around that gap, and when the weighting gets fancy enough, it can cough up exactly the kind of mess Silver found.
Big samples don’t rescue a bad sample, either.
In 1936, the Literary Digest mailed out 10 million straw ballots, built mostly from phone books and car registration lists, and got about 2.4 million back. The magazine predicted Republican Alf Landon would beat Franklin Roosevelt 57 to 43.
Roosevelt took roughly 62 percent of the vote and 46 of the 48 states.
That skew ran toward Republicans, since phones and cars mostly belonged to the better-off in the middle of the Depression. Sampling errors don’t carry party cards. A pool stuffed with people who love answering surveys will mislead in whatever direction those people lean, and these days those people tend to hang diplomas on the wall.
Election night will grade Marist’s homework, same as everybody else’s. If the numbers miss, expect the excuses to roll in right on schedule.
But Silver already put his finger on the sore spot. A poll that has to operate on its own sample just to land somewhere believable leaves readers wondering which version anybody at Marist actually trusted.
When the votes came in that November, Alf Landon carried exactly two states, Maine and Vermont.
The Literary Digest stopped publishing two years later.
Sources: Mediaite, “‘Indefensible!’ Nate Silver Goes Nuclear on Major Pollster Over ‘Cockamamie’ Results”; Nate Silver on X (@NateSilver538); Marist Poll, “The Midterm Elections in Maine, October 2026”; Marist Poll, “Midterm Elections in Maine, October 2026” (October 1-4 survey); PollCheck, Maine Senate and Maine Governor 2026 polling averages; Newsweek, “Susan Collins vs. Troy Jackson: Candidates’ Final Appeal to Voters Compared”; Newsweek, “Collins vs. Jackson: What Latest Polls Say”; Political Wire, Suffolk University and CBS News/YouGov Maine polls; Emory University, “Famous Statistical Blunders in History”; ProQuest, “That Time the Literary Digest Poll Got the 1936 Election Wrong”; Pivotal Research, 1936 election white paper; arXiv, “Why Do Polls Fail?”