Welcome back to The Political Spectrum! Like everyone else, I am still trying to process what exactly happened in the midterms. There will be plenty of stories about this in the coming weeks and probably months. Reminder that paid subscribers will get a chance to ask questions for tomorrow’s special midterm mailbag. So become a paid subscriber, send a message, or comment below. Thanks. But for now, I want to tackle another subject: Polling and why it underestimated Democrats so terribly for the second year in a row.
I figured I would spend the first part of my Election Night in the State Ballroom Center at Marriott City Center in Raleigh, N.C. covering Roy Cooper’s Senate campaign party before decamping to my hotel to write.
I expected networks to call the race for Democratic nominee Roy Cooper fairly early. The most generous polling from AtlasIntel dropped the day before the election showed Cooper with a 15-point lead against Republican nominee Michael Whatley, but I discarded the survey given AtlasIntel’s uneven record.
In the end, Cooper won by 19.5 percent and caused a nearly extinction-level event for the North Carolina Republican Party, giving Democrats the majority in the state House of Representatives. It would be a preview of a bloodbath for Republicans.
Ironically, I had said publicly here and on social media that polling outlets were being overly cautious so as not to overstate Democrats and repeat their mistakes. The first piece I wrote in this space examined how people should read the polls this cycle and whether there would be a polling miss. I took a step further on Bluesky when I said last month that “I don’t do copium and I don’t Democratmaxxing. I am a journalist. But I will say this looking at the samples: polling is going to have a big miss and will underestimate Democrats. Pollsters are being skittish and weighting for 2024.”
It turns out that I was right, and if anything, I was more generous than others. The X account Politics and Poll Tracker showed how big the polling miss was. As election results poured in, VoteHub said that the national environment was shaping up to favor the Democrats by about 13 points, four points better for the Democrats than the national generic ballot polling average.
The trend repeated itself in Georgia, most polls showed incumbent Democratic Sen. Jon Ossoff polling in high single digits, only to steamroll Rep. Mike Collins (R-Ga.) by 18 points. In Texas, The New York Times/Siena University released two days before the election showed Democrat James Talarico beating Ken Paxton by only 7 points. When the ballots were in, Talarico would add about 5.1 points to that lead, beating Republican Ken Paxton by 13.1 percentage points.
So why was I so skeptical of the polls? The easiest answer is that pollsters feared another miss on the levels of 2016, 2020 and 2024, when pollsters underestimated both Trump and Republicans because they missed Trump’s pull with low-propensity and non-college-educated voters.
And Trump has proven to be litigious. In the days before the 2024 election, Ann Selzer, the most prominent pollster in Iowa, and The Des Moines Register dropped a survey showing Kamala Harris had a three-point lead in the Hawkeye State.
That turned out not to be the case and Trump would win Iowa by 13.5 points. Afterward, Trump attempted to sue Selzer and The Des Moines Register. That alone creates a chilling effect. That alone makes pollsters cautious.
There’s also just the credibility factor. Pollsters need to have a record of accuracy for people to continue speaking to them. But many of the polls in states with competitive races like in Texas used samples that assumed that the electorate would remain the same as it was in 2024.
Let’s take the final YouGov poll for Texas’s Senate race. That survey showed that 50 percent of likely voters supported Talarico and 44 percent supported Paxton. One reason why they missed? Look no further than the partisan breakdown in the sample, which was 41 percent Republican, 26 percent Democratic and 28 percent independent.
The survey also asked respondents how they voted in 2024 and 52 percent said they voted for Trump and 40 percent said they supported Harris.
But that simply did not match the national environment. The same survey showed 55 percent of likely voters disapproved of Trump’s performance as president. Even more, 17 percent of Trump 2024 voters said they disapproved of Trump’s performance. That strong anti-Trump sentiment will not produce a Republican +15 electorate.
And the Texas YouGov poll is not even the most egregious example. The Marist Poll for the Maine Senate race on Oct. 27 had the worst sample I saw. The survey showed that 49 percent of likely voters preferred Troy Jackson to Sen. Susan Collins, who clocked in at 47 percent. On the surface it seemed plausible given Collins’s advantage as a five-term incumbent and the fact she won in 2020 by eight points when Joe Biden won by nine points.
But the tell that the poll was drunk came in its survey for governor, which had Democrat Hannah Pingree tied with Republican Bobby Charles, each at 44 percent. That did not compute in a state that voted for Harris by almost seven points. And the real smoking gun came when comparing likely voters to registered ones. Jackson beat Collins 53 percent to 44 percent and Pingree beat Charles 51 percent to 40 percent.
But on Election Night, Pingree beat Charles by more than 17 points and Jackson beat Collins by about 6.43 points. It turned out the registered voter survey was more in line with the likely voter poll.
A reason for this? The likely voter sample broke down 31 percent Democrat, 31 percent Republican and 35 percent independent. But the registered voter sample was 36 percent Democratic, 31 percent Republican and 35 percent independent.
Democrats will likely not be calling for any pollster’s heads the way Trump’s team did. Still, these polls matter.
As Michigan state Sen. Mallory McMorrow said a few weeks ago, the bogus Wedgewood poll all but killed her campaign. Public opinion matters and in between elections, polling is one of the best ways for voters to express their thoughts and hopefully have elected officials respond. Fear of retribution from a self-obsessed president or even a little bit of embarrassment does not warrant this much hedging.
Writing hither and yonder: I wrote for MS NOW about Republicans lost Latinos because they misread their big victory in 2024.
Most eye-popping stat I read: Ben Briscoe for WFMY News 2 in Greensboro broke down how big Roy Cooper’s victory in North Carolina was.
Cooper received about 2.47 million votes, compared with Cunningham’s roughly 2.57 million in 2020: a decrease of about 102,000.
Whatley received about 1.64 million votes, compared with Tillis’ roughly 2.67 million six years ago: a drop of more than 1 million.
The pattern also shows up locally. In Randolph County, votes for the Democratic Senate candidate remained about the same, while votes for the Republican candidate fell from roughly 53,000 in 2020 to about 31,000 this year.
What I’m reading: Rachel Monroe in The New Yorker about Texas had one of the best kickers I have read in a while. Unsurprisingly, Beto O'Rourke has some insights about how Texas flipped. My former colleague Sophia Nguyen writes in The Atlantic about one underrated reason boys are reading less. Nick Miroff, one of the best immigration reporters in the country, and my fellow Tar Heel Matt Viser write in The Atlantic on how Texas flipped. Burgess Everett in Semafor writes about Annie Andrews’s most savage political move.
Your regular Bam Bam post.
That does it for The Political Spectrum. Thanks for reading. If you like this newsletter, please subscribe below.