Can Polls Measure the True Strength of the Left?

Progressives had high hopes of dominant victories in marquee primaries in Michigan and Wisconsin, but in the end their candidates fared much worse than expected. The results would have been a lot less disappointing if the pre-election polls hadn’t been so far off.

The polls helped set extraordinarily high expectations for progressives: Dr. Abdul El-Sayed was ahead by double digits in the Senate primary in Michigan, and Francesca Hong led by more than 20 points in the Wisconsin governor’s primary.

Dr. El-Sayed ultimately won by one percentage point. Ms. Hong lost.

Primary elections have always been hard for pollsters. They’re uncertain and volatile. Turnout is low, and voter preferences can switch on a dime. This is very different from a general election, when people know whom they’ll support based on whether a candidate is “D” or “R.”

Even so, it’s pretty unusual for polls to badly overestimate the same political faction twice in two weeks. It’s fair to wonder whether the errors betray a deeper problem with measuring the strength of moderates against the progressive left.

Although it’s hard to say based on just two races, it’s possible they do.

Give the polls a p*** in Wisconsin

The polls missed by a lot more in Wisconsin than Michigan, but pollsters deserve a p*** in the state where they did the worst.

Wisconsin was a cl***ic example of why primaries are so hard to poll. Progressive Democrats quickly consolidated behind Ms. Hong, a democratic socialist. The party’s establishment struggled to put forward a candidate at all. Over the final month or so of the race, three establishment-backed candidates dropped out. Incredibly enough, this included the eventual winner, David Crowley, who re-entered and wound up winning by less than a point.

With progressives unified and moderate voters divided and bewildered, Ms. Hong built an enormous lead in the polls — and the ingredients came into place for a major polling error. If moderate voters were to consolidate behind Mr. Crowley, a relatively unknown candidate who nonetheless was the last moderate standing, the race could quickly tighten. In the end, that’s exactly what happened: Mr. Crowley’s support surged as moderates coalesced behind his candidacy.

Almost all of the polling error was because the polls underestimated Mr. Crowley’s support. The polls did slightly overestimate Ms. Hong — she won 39 percent of the vote, while the polls showed her at 44 percent on average — and this may suggest a deeper kind of error. But it’s equally possible that she lost support down the stretch as the final week of the campaign was dominated by stories highlighting her general election vulnerabilities, like her apparent opposition to … Thanksgiving and Halloween.

Michigan raises more serious questions

The pollsters in the Michigan race don’t have easy excuses.

Dr. El-Sayed and Haley Stevens, the establishment-backed candidate, made up a clear top two. They offered an obvious contrast and received plenty of national media attention. Tens of millions of dollars were spent on their behalf, and voters had every opportunity to know the candidates and the stakes. Neither candidate had major missteps in the final week of the campaign. It’s hard to believe many voters changed their minds.

Nonetheless, the polls badly overestimated Dr. El-Sayed. What happened? Some explanations are more troubling than others.

One relatively innocuous one would be if pollsters were wrong about the makeup of the electorate. Maybe they overestimated the turnout among young voters, for instance, who overwhelmingly backed Dr. El-Sayed.

This is a real possibility. The final Emerson College poll, a rare nonpartisan poll of the race, reported a much younger likely electorate than in other recent primaries in Michigan. There are other plausible challenges in handling turnout, like the possibility that Republican-leaning voters — who presumably would have backed Ms. Stevens — turned out in an open primary.

If turnout were the problem, pollsters could conceivably correct for it. Even if they couldn’t, it’s far from guaranteed they would err in the same way in other races. It’s even possible that a different set of pollsters wouldn’t have made the same mistake, as each pollster has a different approach for estimating the likely electorate.

A more serious problem would be something called “nonresponse bias,” in which one candidate’s supporters are much likelier to take a poll than others. This is the same type of problem that’s plagued polling about President Trump, whose supporters appear much less likely to respond to surveys. In Michigan, nonresponse bias could have been responsible for the polling error if Dr. El-Sayed’s supporters were more likely to respond to surveys than those who backed Ms. Stevens.

Is nonresponse bias to blame?

It’s really hard to prove whether nonresponse bias is an issue, but it’s plausible to think it could become a problem in Democratic primaries. Progressive candidates tend to fare better among highly educated and highly engaged voters, who tend to be likeliest to respond to political surveys. If progressive voters are likelier to respond, it might simply be the flip side of the better established tendency of Mr. Trump’s supporters to be less likely to respond.

Worse, pollsters increasingly rely on methods that might exacerbate this problem, such as by surveying people who sign up to take polls online or who are eager to click on a text message link to a poll. In our own experimentation, these techniques reach the most highly educated and politically engaged voters. These techniques might be likelier to reach progressives as well.

In our poll of the New York mayoral race in 2025, for instance, 70 percent of registered Democrats who took the poll by text message said they backed Zohran Mamdani in the general election, compared with 54 percent of those who took the poll on the phone. The only traditional telephone poll fielded over the last month in either Michigan or Wisconsin was also the only poll that actually underestimated the progressive candidate.

If the problem is nonresponse bias, pollsters could be in trouble. Over the last decade, they’ve relied on heavy-handed techniques to represent Mr. Trump’s supporters, such as ensuring the right number of voters who say they voted for him in the previous election. There is no equivalent way to ensure the right number of both progressives and moderate Democrats.

Worse, pollsters’ techniques to represent Mr. Trump’s supporters may have blinded them to just how bad their data really is. These techniques can be like a bludgeon, hammering even the worst samples into the ballpark of a reasonable Democrat-versus-Republican result. But their data might still be way out in left field in other contexts, like a Democratic primary.

The Michigan primary could just be a one-off polling error — the wrong pollsters in the wrong place at the wrong time. But if not, it might offer a hint of just how poorly many polls might fare without the crutches they’ve used to hobble through the Trump elections.