With Election Day less than two weeks away, University of Michigan public opinion researchers Michael Traugott and Josh Pasek provided an overview of what the data currently tell us about the upcoming midterms.
The talk was presented as part of the ISR Insights series held at the Institute for Social Research, where Center for Political Studies experts present analyses in panels before and after US elections. Three takeaways:
1. Every indicator political scientists watch is pointing the same direction
Political scientists typically treat midterms as a referendum on the sitting administration, and they measure that referendum several ways. Right now, those measures agree:
- Presidential approval: Trump’s approval sits at roughly 39% against 59% disapproval — about 20 points “under water,” and the lowest reading for an incumbent president at this point in the era of modern public polling.
- The generic congressional ballot: Aggregated polling shows roughly 51% preferring a Democratic candidate to 42% preferring a Republican — about an eight-point gap that has widened over the course of the term.
- Historical base rates: Since the 1930s, the president’s party has almost always lost House seats at the midterm. Only three times has it gained seats. In 2018, Republicans lost 40.
- Issue concerns: Recent Fox News polling found three in four Americans or more are extremely or very concerned about corruption in government (the top issue), inflation and prices, gas prices, health care, and government leadership.
When the fundamentals and the polling converge, Pasek noted, that agreement itself suggests the environment is tilted meaningfully in one direction, rather than reflecting noise in a single data source.
2. The Senate is the real contest
Republicans currently hold narrow majorities: roughly 219–214 in the House (with two vacancies) and 53–45 in the Senate, with two independents caucusing with Democrats. A four-seat shift would flip either chamber — not a big lift in the House, but a heavy one for the Senate.
House: Seat-translation models built on the generic ballot suggest that an eight-to-nine-point Democratic popular vote advantage would produce a Democratic pickup in the neighborhood of 26 to 28 seats. Alan Abramowitz’s model for Sabato’s Crystal Ball produces a comparable estimate (about 192 Republican seats at a minus-eight Republican margin). Aggregators are broadly confident Democrats take the chamber.
Senate: Senate forecasting works differently — it’s done state by state, with heavy weight on candidate quality and well-funded statewide advertising. The Cook Political Report identifies seven toss-ups: Alaska, Iowa, Kansas, Maine, Ohio, Texas (all currently Republican-held), and Michigan (an open seat). Nate Silver’s simulation model puts Democratic control at roughly a three-in-four chance. A caveat: The 2026 Senate map was long considered unfavorable for Democrats, because competing at all requires winning in territory with a partisan voter index of R+5 or greater. The current polling suggests they may be there — but it’s a harder map than 2028.
3. Election polling is genuinely hard, but this margin is larger than typical polling error
Pasek, who led the AAPOR Task Force on 2024 Pre-Election Pollin, spoke on the challenges of election polls that face, including low response rates, participation bias, and the burden of predicting future behavior. “Polls are excellent at telling you which races are competitive,” said Pasek. “They are not precise enough to tell you who wins a race decided by a point.”
That said, larger margins make contests easier to call, and recent polls have a decent track record of predicting Senate races within a 5-point margin of error, Pasek said. An open question is whether something specific was going wrong in ballot years when Trump was on the ballot — distrust of pollsters among his supporters, for example, leading them to hang up. If so, it’s unclear whether that carries into a midterm where he isn’t a candidate but is working hard to make the election about himself. “Not obvious,” Pasek said. No one really knows.
A referendum on democracy
Both presenters argued that how people want to vote may be a less decisive factor this year than usual, because of pressures on the process itself.
Pasek flagged claims about placing federal officers at polls, attempted changes to how the Postal Service handles ballots, and a growing number of officials who denied the 2020 results now administering elections. The presence of these issues in the media alone can impact voter turnout, Traugott noted, since uncertainty about federal officers at polling places, registration rules, or mail handling can depress participation on its own. Most concerning to Pasek: Apparent degradation of losers’ willingness to concede.
But U.S. elections are run locally, across nearly 200,000 polling locations. That decentralization is a real check against any single national intervention.
On Election Day, we may see an enthusiasm gap that drives more Democrats to the polls, and a potential shift away from Trump among Hispanic and younger voters.
Questions about how the process unfolds, how accurately polls predicted behavior, and who shows up to vote will only be knowable on election day. Several toss-ups sit in the Eastern time zone, so early returns from Maine, Michigan, and Ohio will offer an unusually good early read for East Coast viewers on the night.
Michael Traugott is a political scientist who studies campaigns and elections, voting behavior, and the impact of mass media on American politics, and is the author of A Voter’s Guide to the Election Polls, now in its 6th edition.
Josh Pasek is Professor of Communication & Media and Political Science, Research Professor at the Center for Political Studies, Institute for Social Research, and Core Faculty of the Michigan Institute for Data Science at the University of Michigan.
This post was developed by Tevah Platt, communications manager for the Center for Political Studies. Recordings of past and upcoming talks are available through ISR.
Watch the full talk:
