Polling

🚨 HOLY CRAP! Democratic socialists were OVERESTIMATED by the polls by nearly 25 points in Wisconsin

Francesca Hong loses to David Crowley in the governor's primary by just a few thousand votes, a fraction of a point

2AM ballot drops solidified the socialist's loss, and… pic.twitter.com/Yyc3fFn1z1

— Eric Daugherty (@EricLDaugh) August 12, 2026

I've been checking Real Clear Politics, and notwithstanding the sites's prestige comes from its polling aggregate, so far, there's not a peep about how wrong the polls were. And Frank Luntz, a silly individual, minimizes the problem:

Luntz said the math doesn't work with low voter turnouts that happen often.

"How," he explained, "when you get turnouts of 15% electorate, you can't make a conclusion or a projection. And pollsters need to stop doing this."

Luntz posted about the issue in the early hours of Wednesday in connection with the Wisconsin governor's race, where democratic socialist Francesca Hong had been declared as carrying large, double-digit leads over eventual winner David Crowley.

"I am calling for a total and complete shutdown of primary polling in the Midwestern United States," he wrote, "until we can figure out what the hell is going on."

But then he turns around and says low turnout isn't the problem:

Luntz told Henry that he was among those pollsters in 2016 who went all-in on Hillary Clinton winning the presidency.

"I believed that Hillary Clinton was going to win," he admitted, "based on the exit polling of all the networks.

.. . "And so, I put out a prediction that said, what makes you not understand the words President Hillary Clinton? I got it wrong because I believed those numbers. And I fixed it," Luntz said.

But that wasn't a low-turnout primary. And what did he fix? But let's get back to Real Clear Politics. I asked Chrome AI mode, "How accurate has the Real Clear Politics average been in predicting the 2026 primary cycle?" It answered,

The RealClearPolitics (RCP) polling average has seen a mixed but highly scrutinized performance during the 2026 primary cycle, facing notable accuracy challenges in key swing-state primary races. While the tracking averages effectively mapped broad national moods—such as an unfavorable macro-environment for the governing party—the final primary results in states like Michigan and Wisconsin exposed widening systemic errors.

In highly contested races (especially within Democratic primaries in the Midwest), standard polling models failed to properly weight low-propensity progressive or anti-establishment voters, leading to surprise double-digit margins on election night.

Primary polling naturally suffers from low sample sizes and a high volume of undecided voters who break late. RCP’s method of strictly averaging available public data meant that low-quality or outdated polls frequently dragged down the accuracy of the final metric.

With the primary cycle wrapping up, attention has shifted to the general election, where the RealClearPolling 2026 Generic Congressional Vote tracker currently shows Democrats holding a modest single-digit lead. However, given the primary misses, forecasters caution against relying too heavily on early general election averages.

Put more concisely, RCP averages garbage polls and gets garbage results, but it still has prestige. I asked Chrome AI mode a related question, "Do polling errors consistently overestimate Democrats?" It answered, predictably,

No, polling errors do not consistently overestimate Democrats, despite a recent run of high-profile presidential elections where they did. When looking at the full history of modern polling since the 1930s, polling errors are historically unbiased, underestimating Democrats and Republicans about equally often over the long term.

However, the perception of a permanent pro-Democratic bias exists because polls systematically underestimated Republican support across consecutive presidential election cycles involving Donald Trump.

But that leads to the question of why the polls were inaccurate when Trump was on the ballot, as well as why the Wisconsin and Michigan polls were so inaccurate this year, when Trump wasn't on the ballot. And the big question is why Real Clear Politics isn't reexamining why its polling averages simply pass the inaccuracies of individual polls into the aggregate, when the assumption ought to be that they would present a more accurate total that discounts outliers, Instead, the outliers skew the results.

But that seems to be their whole philosophy: post one piece from Victor Davis Hanson or Roger Kimball against another by Robert Reich or Harold Myerson and think you can mix them together and get a golden mean.