How America's Accurate Election Polls Were Covered Up

SubscribeSign in

How America's Accurate Election Polls Were Covered Up<br>The Real Clear Politics National Average was removed by Wikipedia before the election and the New York Times denounced its failure to skew data. A solution to the "mystery" of crappy polling?

Matt Taibbi<br>Nov 15, 2024

1,752

653<br>126

Share

Note to Substack users: I’ll be answering questions from subscribers about this story on the “live” feature later this evening. Watch on Notes for a video appearance.

John McIntyre couldn’t believe it. The publisher of the Real Clear Polling National Average, America’s first presidential poll aggregator, woke on October 31st to see his product denounced in the New York Times. Launched in 2002 and long a mainstay of campaign writers and news consumers alike, the RCP average, he learned, was part of a “torrent” of partisan rubbish being “weaponized” to “deflate Democrats’ enthusiasm” and “undermine faith in the entire system.”<br>“They actually wrote that our problem was we didn’t weight results,” says an incredulous McIntyre. “That we didn’t put a thumb on the scale.”

The New York Times denounced a poll aggregator’s lack of subjectivity.<br>The Times ended its screed against RCP’s “scarlet-dominated” electoral map projection by quoting John Anzalone, Joe Biden’s former chief pollster, who said: “There’s a ton of garbage polls out there.” But being called “garbage” in America’s paper of record was nothing compared to what happened to RCP at Wikipedia.<br>Six months ago, when former Wikipedia chief Katherine Maher became CEO of NPR, video emerged of her talking about strategies at Wikipedia. She said the company eventually abandoned its “free and open” mantra when she realized “this radical openness… did not end up living into the intentionality of what openness can be.” Free and open “recapitulated” too many of the same “power structures,” resulting in too much emphasis on the “Western canon,” the “written tradition,” and “this white male, Westernized construct around who matters.”

In the context of poll averages, it seems even a track record of accuracy did not “end up living into the intentionality of what openness can be.” The ostensibly crowdsourced online encyclopedia kept a high-profile page, “Nationwide opinion polling for the 2024 United States presidential election,” which showed an EZ-access chart with results from all the major aggregators, from 270toWin to Silver’s old 538 site to Silver’s new “Silver Bulletin.”<br>Every major aggregate, that is, but RCP. McIntyre’s site was removed on October 11th, after Wikipedia editors decided it had a “strong Republican bias” that made it “suspect,” even though it didn’t conduct any polls itself, merely listing surveys and averaging them. One editor snootily insisted, “Pollsters should have a pretty spotless reputation. I say leave them out.” After last week’s election, when RCP for the third presidential cycle in a row proved among the most accurate of the averages, Wikipedia quietly restored RCP.

“THE INTENTIONALITY OF WHAT OPENNESS CAN BE”: Wikipedia’s poll page before (left) and after (right) the presidential election<br>“They just whitewashed us away three weeks before the election because we were a point or a point and a half more favorable to Trump, which as it turns out still underestimated him,” McIntyre said.<br>What happened in 2024 to RCP is emblematic of wider failures in data journalism, which has now turned in three straight cycles of obscene misses. Although problems in polling have been lavishly, even excessively covered, failures are inevitably presented as a Scooby-Doo whodunit, rooted in a magic invisibility power apparently unique to Trump voters. “If Trump outperforms the polls once again,” the Atlantic concluded this August, “something about his supporters remains a mystery.”<br>But it’s no mystery. The polling problem in America looks like good-old-fashioned lying, mixed with dollops of censorship and manipulation:<br>Start with a basic fact: major aggregators like 270toWin, Silver Bulletin, 538, and the Cook Political Report, while useful, are not the uncomplicated averages many customers imagine them to be. “A poll average is simple,” says McIntyre. “You add polls up, and you divide by the number.”<br>But the “reputable” or “nonpartisan” averages are different. “Those are really models, but they don’t call them that,” McIntyre says. “The New York Times had an average and the Washington Post had one, but they’re not really averages. They’re more like, ‘This is where we want to say the race is.’”<br>Most “averages” or “aggregates” have pages that explain how they weight results, though the exact recipe is usually a mystery. The Times in September explained in a general way how it weights pollsters. To an unspecified degree it favors track records of “unbiased and accurate results” (note the term “accurate” isn’t by itself). It also upgrades for “professionalism,” which to the Times among other...