



Pollsters head into the midterms carrying a decade of high-profile misses, from underestimating Trump's 2016 support to the worst collective 2020 performance in 40 years to a third straight underestimate of Trump in 2024. August 2026 brought fresh embarrassment: double-digit polling leads in Michigan and Wisconsin primaries evaporated on election night, and a 21-year-old college graduate was found to have fabricated poll results distributed through a website, some of which got picked up by a mayoral campaign and news outlets before being exposed.
Prediction markets like Polymarket and Kalshi have moved to fill the credibility gap, correctly calling the 2024 presidential race while polls favored Harris. But those markets stumbled too, giving a Wisconsin gubernatorial primary candidate a 95-96% win probability shortly before she lost. Low survey response rates and inconsistent statistical weighting compound pollsters' troubles, while AI-simulated ‘silicon sampling’ adds a new layer of uncertainty to an already shaky trade.
The full dispatch is available from the source below.