Pollers brace for surprise as fake surveys and nonresponse skew 2026 midterms

Democratic candidates appear poised to win control of the U.S. House and potentially the Senate, according to current polling data. However, veterans in the industry caution that these numbers may not hold on November 3. The upcoming midterm election arrives with a heightened risk of polling errors due to unverified surveys and persistent gaps in respondent participation.

The reliability of recent polls has been shaken by two major incidents during the 2026 cycle. In Michigan and Wisconsin, statewide primary results diverged sharply from survey projections. Additionally, two firms, Median Strategies and Wedgewood Polls, released fake data that gained widespread traction. The New York Times and the aggregator FiftyPlusOne both initially included Wedgewood’s numbers in their averages. Both fake pollsters later admitted their actions were a social experiment to test media vetting standards.

Charles Franklin, director of the Marquette Law School Poll, identifies nonrandom nonresponse as a core challenge. He notes that voters who distrust pollers are often the least likely to answer surveys. This dynamic became evident in 2016, when a segment of the electorate mobilized by Donald Trump declined to participate in phone interviews. Franklin estimates this group represented 4% to 6% of the electorate. Their silence created a systematic bias that favored Republican candidates in key states.

The polling industry has adjusted its methods to reach harder-to-contact voters. Franklin’s team now relies primarily on online surveys sent via email and text to registered voters. They reserve telephone contact for those who do not respond to digital outreach. This approach leaves only 5% to 10% of the sample reached by phone, a shift from the 100% telephone methodology used before 2022. Despite these changes, Franklin admits that pollsters cannot fully account for the preferences of those who never respond.

The Wisconsin Democratic primary for governor illustrated how campaign chaos affects polling accuracy. Francesca Hong, a democratic socialist, led in surveys for months. She ultimately finished with 39.3% of the vote, narrowly losing to moderate David Crowley. Franklin points to a volatile final four weeks, which saw multiple candidates drop out and return. The incumbent governor also endorsed Crowley late in the race. Franklin noted that undecided voters over 50, a group typically firm in their choices, lacked a positive view of democratic socialism.

Fake polls have also influenced prediction markets and campaign finances. Democratic state Sen. Mallory McMorrow dropped out of the Michigan U.S. Senate primary after a fake Wedgewood poll showed her in single digits. She stated that her fundraising dried up following the release of that data. Franklin observes that while single polls rarely move betting markets like Kalshi or Polymarket, a single result in a low-polling race can shift odds significantly.

Social media amplifies both real and fabricated data, creating incentives for unverified firms to seek clicks. Franklin predicts that aggregation sites will tighten their vetting processes ahead of the 2028 presidential cycle. The rise of prediction markets adds another layer of complexity, as political betting now responds to survey outputs. Franklin’s team currently faces growing skepticism from respondents who question the legitimacy of digital survey links.

The Marquette Law School Poll continues to draw its state sample from public registered voter lists. The team sends four attempts to contact each participant via email and text before placing a call. Franklin acknowledges the trust deficit, often telling inquirers that he is glad they paid attention to their IT training. The next major test for these methodologies arrives with the November 3 general election.