The central problem, the authors argue, is that forecasters have the least historical evidence about the factor they most need to understand: how strongly national mood influences each individual race. National forces include presidential approval, partisanship and economic conditions, which can influence Senate contests across multiple states at once. Because all races on the ballot in a given year face the same national climate, the data offer surprisingly few independent observations of that climate’s effect.

The study draws on eight Senate election years from 2010 to 2024, covering 269 races. That may sound like plenty, but all races in a given year face the same national political climate. Because that climate varies from year to year, researchers have only eight distinct election cycles to help them estimate how strongly national forces affect Senate races.

To see how much this limitation matters, the researchers tested 120 plausible versions of their model. Each version made different calls about how much incumbency helps, how much a state’s political lean matters and how strongly national mood moves races. All but three of the versions fit past elections similarly well, yet they disagreed about who would control the Senate.

In the version analyzed in the paper, which reflects data as of Aug. 19, the chance of Democratic control ranged from 43.2% to 57.6%. About 89% of that uncertainty came down to a single modeling choice: how strongly national mood carries into Senate races. The model gives heavy weight to national conditions, but the data cannot tell researchers how much weight to give them.

The researchers’ own forecast, which updates daily, is built on a new model that reflects the growing nationalization of congressional elections. It gives Democrats a 61.2% chance of winning control of the Senate in November and a 90% chance that Democrats will win between 46 and 56 seats. Democrats need 51 seats to take control because the vice president breaks ties.

Christopher Hare, an associate professor of political science at UC Davis and co-author of the study, said races move partly on their own and partly with national tides. “Because of that, we need to widen out our uncertainty intervals that much more,” he said. “We need to say we’re that much less confident about these races.”

Hare said the study is not a critique of pollsters. As fewer people answer their phones, forecasters have less data to work with and must account for the possibility that polls do not represent the full electorate. “This is an ongoing and perhaps worsening source of uncertainty that we also need to account for in the models,” Hare said. “Those 95% credible intervals assume random samples, and do we really have random samples? If not, then you need to expand that out.”

The authors recommend that forecasters show the range of possible odds, alongside the usual margin of error, rather than presenting one number as if the data had settled every modeling choice. Hare said that matters beyond statistics because forecasts shape how people think about whether their vote counts. “There’s something maybe a little bit disenfranchising about models that are too confident,” he said, while stressing that the answer is not to declare every race a toss-up. “But it’s really dangerous if we have bad models that are overstating their confidence and that’s depressing turnout when the race is potentially a lot closer.”

Hare compared election forecasts to weather forecasts: acknowledging uncertainty is not the same as giving up. The point, he said, is to avoid being the forecaster who insists it will never rain right before a thunderstorm hits. That does not mean every race is a toss-up; the model puts Wyoming at 99% Republican. Jack Rametta co-authored the paper while a postdoctoral researcher at UC Davis; he is now a political scientist and research fellow at the Mercatus Center at George Mason University. The goal, Hare said, was a more complete accounting of where the question marks are, especially in polarized times when people look to these forecasts.

Subject of Research: Statistical uncertainty in forecasting the influence of national political mood on U.S. Senate election outcomes

Article Title: How confident can election forecasters be about who controls the Senate?

Article References: How confident can election forecasters be about who controls the Senate?. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: election forecasting, U.S. Senate, political science, nationalization, model uncertainty, presidential approval, polling, PS: Political Science & Politics, UC Davis, Rashomon effect, credible intervals, voter turnout