Nonresponding Voters Distort 2026 Midterm Forecasts

Charles Franklin, director of the Marquette Law School Poll, says polling errors stem from voters who refuse to answer surveys. Franklin has conducted national polls since 1990. He notes that the 2016 election exposed a flaw in how surveyors track nonrandom nonresponse.

Franklin explains that Donald Trump mobilized 4% to 6% of the electorate in 2016. These voters often disliked government and media, so they avoided pollsters. Their absence skewed results toward a tighter race than actually occurred. Trump won Wisconsin, Michigan, and Pennsylvania by less than a point despite polls suggesting a loss.

The industry adjusted methods after 2016, but uncertainty remains for the 2026 midterms. Franklin reports that polling was accurate in the 2018 and 2022 midterm cycles. He attributes this accuracy to the absence of Trump on the ballot. Without his influence, the group of nonresponding voters shrank, allowing traditional methods to capture results within two points of the actual outcome.

Challenges reemerged in recent statewide primaries. In the Wisconsin Democratic primary for governor, polls showed Francesca Hong leading with 38% support. She received 39.3% of the final vote. The race featured chaotic turnover, including a candidate dropout, a financial scandal, and a late endorsement by the incumbent governor for moderate David Crowley. Franklin notes that undecided older voters favored Crowley over a democratic socialist candidate.

Fake polls added confusion this cycle. Median Strategies released a false survey for the Los Angeles mayoral race. The city mayor retweeted the numbers before the Los Angeles Times identified the fraud. The New York Times and FiftyPlusOne also included a fake survey from Wedgewood in their averages. Both outlets later discovered the group lied about its methodology. Franklin says this issue will require new vetting standards for the 2028 election cycle.

Social media drives the creation of fake polls. Franklin states that pollsters face high demand for data on minor races. Outlets and creators seek numbers to generate clicks, even when major organizations do not prioritize those contests. Prediction markets also react to poll data. In the Michigan Senate primary, a fake Wedgewood poll showed Mallory McMorrow in single digits. This false data contributed to her decision to drop out of the race.

Marquette now conducts most polling online. The team draws samples from state registered voter lists. They send emails and text messages with links to surveys. If respondents do not answer four attempts, staff call them by phone. This method results in 5% to 10% of samples being completed via telephone. Franklin says this is a change from the 100% telephone method used before 2022.

Trust issues affect response rates. Franklin receives inquiries from sampled individuals who question the legitimacy of survey links. He jokes that he is glad they completed IT training. The next major test for these methods occurs on November 3.