Pollsters badly missed the mark in two closely watched Democratic primaries in Michigan and Wisconsin, where surveys suggested progressive candidates were headed for comfortable victories. Instead, both races came down to the wire.
In Michigan, Abdul El-Sayed was repeatedly shown holding a commanding lead over Rep. Haley Stevens in the Democratic Senate primary. Some polls had El-Sayed ahead by more than 20 points. When the votes were actually counted, however, he won by just one percentage point, 48.5% to 47.5%.
A similar story played out in Wisconsin. Progressive gubernatorial candidate Francesca Hong had been shown leading David Crowley by more than 20 points in several surveys. Crowley ultimately defeated her by less than 1%.
The size of those misses immediately raised questions about what went wrong. Pollsters say the answer is not as simple as faulty methodology. Primary elections, particularly open primaries, are much more difficult to measure than general elections because determining who will actually vote is considerably harder.
Michigan and Wisconsin both allow voters to participate in primaries without registering with a political party. That creates an additional layer of uncertainty for pollsters trying to build an accurate model of the electorate.
Eli McKown-Dawson, a senior elections analyst at Silver Bulletin, told the Daily Caller News Foundation that large polling errors are far more common in primaries than they are in general elections.
“The polls in Wisconsin and Michigan did miss in a big way,” McKown-Dawson said. “But primary polls in general are more prone to error than polls of general elections.”
He noted that polling errors in presidential primaries historically average about nine points and can climb into the double digits in congressional primaries.
The problem is that the usual tools pollsters use to build representative samples become less reliable when the electorate itself is unpredictable. In a general election, researchers have a much clearer idea of what turnout will look like. In an open primary, Democrats, Republicans and independents can all potentially participate, making the final voter pool much more difficult to forecast.
Another major factor appears to have been age.
Polls in both Michigan and Wisconsin included more young voters than ultimately showed up at the ballot box. That mattered because younger and more enthusiastic voters tended to favor candidates such as El-Sayed and Hong.
“It also looks like polls of both races contained too many young voters, and too many voters who were very enthusiastic about the election,” McKown-Dawson said.
Online and text-based polling may have contributed to the imbalance because older voters can be more difficult to reach through those methods.
Democratic pollster Adam Carlson reached a similar conclusion after examining the Michigan results. Carlson, a founding partner at Zenith Research, said El-Sayed’s supporters appeared far more eager to participate in surveys than Stevens’s supporters.
“El-Sayed’s supporters were more fired up and willing to take polls down the stretch, but more enthusiasm does not necessarily equal more votes,” Carlson wrote.
That difference between enthusiasm and turnout proved important.
Stevens performed particularly well among older voters, people without college degrees and lower-to-middle-income voters. El-Sayed was stronger in counties with larger populations of white, college-educated and higher-income voters.
The candidates also represented different ideological wings of the Democratic Party.
Stevens positioned herself as the more moderate choice, supporting reform of Immigration and Customs Enforcement rather than abolishing it, favoring a public health insurance option rather than Medicare for All and maintaining strong support for Israel.
El-Sayed attracted prominent endorsements from the party’s progressive wing, including Sen. Bernie Sanders, Rep. Alexandria Ocasio-Cortez and Rep. Rashida Tlaib. Senate Minority Leader Chuck Schumer, meanwhile, reportedly backed Stevens and encouraged donors to support her campaign.
Wisconsin pollsters faced many of the same problems.
G. Elliot Morris wrote that the average polling error in Senate and gubernatorial primaries is roughly 13 points. In Wisconsin, one of the biggest mistakes was predicting an electorate that was much younger than the one that actually voted.
Among voters whose ages were known, roughly 52% to 53% of the primary electorate were seniors. Only about 2% to 3% were younger than 30.
A July poll, by comparison, modeled an electorate in which just 36% were seniors and 17% were under 30.
“It looks like both of these processes failed to anticipate an older, more moderate primary electorate,” Morris wrote.
Late movement also complicated the picture. Some Wisconsin surveys were conducted before voters had fully settled on a candidate, and Crowley appeared to gain substantial support in the final stretch.
A Marquette Law School poll conducted in late July found that 34% of respondents were still undecided. Many of those voters appear to have eventually moved toward Crowley.
McKown-Dawson emphasized that primaries can shift dramatically in their final days, making polls especially vulnerable to becoming outdated.
“In Wisconsin, for example, you had candidates drop out quite late, and the eventual winner got back into the race just a few weeks before Election Day,” he said.
Crowley and Wisconsin Gov. Tony Evers also emphasized electability late in the campaign, while Hong faced renewed scrutiny over previous social media posts, including a 2020 statement calling for Thanksgiving to be canceled because of its association with colonialism. Hong said on Election Day that she would not cancel the holiday.
Dan Pfeiffer, co-host of “Pod Save America,” argued that those controversies may have pushed undecided voters toward Crowley. He also suggested that the absence of endorsements from Sanders and Ocasio-Cortez could have raised additional questions among progressive voters.
The results in Michigan and Wisconsin offer a reminder of just how uncertain primary polling can be.
The problem was not necessarily that voters suddenly changed their minds en masse. In many cases, pollsters were simply looking at the wrong electorate. Younger, highly engaged voters were easier to reach and more eager to respond, while older and more moderate voters were underrepresented.
Some of those problems can be corrected with better weighting and more accurate voter-file data. Others are harder to eliminate.
As Morris put it, primary polls are best understood as snapshots with very short shelf lives.
The lesson from Michigan and Wisconsin is straightforward: a 20-point primary polling lead can look impressive, but when turnout is unpredictable and a large share of voters remain undecided, it may tell far less about Election Day than it appears to.