Ruffini said the data were drawn from the FiveThirtyEight polling archive, adjusted for partisan house effects, and compared with certified results from the MIT Election Lab.
The analysis does not establish that the same error will occur in 2026. But it does provide historical context for interpreting polls released during the final weeks of the campaign.
And the individual races illustrate why the question remains relevant.
In Iowa in 2020, polling at roughly this point showed Democrat Theresa Greenfield ahead of Republican Sen. Joni Ernst by 2.4 percentage points. Ernst ultimately won by 6.8 points.
Two years later, Iowa Republican Sen. Chuck Grassley was already ahead of Democrat Mike Franken by 7.7 points at the comparable stage. Grassley’s final margin reached 12.2 points.
Michigan produced another substantial polling discrepancy in 2018. Democrat Debbie Stabenow was shown with a 14-point advantage over Republican John James. Stabenow ultimately won by 6.6 points.
The 2020 Michigan Senate race was closer still. Polls showed Democratic incumbent Gary Peters ahead of James by 6.5 points, but Peters finished with a 1.7-point victory.
More recently, the 2024 Michigan Senate race showed Democrat Elissa Slotkin ahead of Republican Mike Rogers by approximately 5.5 points at the same point in the cycle. Slotkin ultimately prevailed by only three-tenths of a percentage point.
Ohio offers another example. In 2018, polls showed Democratic Sen. Sherrod Brown ahead of Republican Jim Renacci by 16.3 points. Brown won by 6.8 points.
In 2022, Democrat Tim Ryan was approximately tied with Republican J.D. Vance in polling at the comparable stage. Vance ultimately won by 6.1 points.
And in 2024, surveys had Brown narrowly ahead of Republican Bernie Moreno, by approximately 0.4 points. Moreno won the race by 3.8 points.
The pattern is important, but it needs to be interpreted carefully. Polling errors are not necessarily systematic, and a historical tendency does not guarantee that the same miss will happen in a future election.
Current polling also shows why the 2026 contests remain unsettled.
A Fox News poll conducted September 24-28 found Democrat Josh Turek leading Republican Ashley Hinson by 49% to 47% among likely Iowa voters, a margin within the survey’s ±3-point margin of error. The same survey found Republicans had an advantage in partisan identification, while Democrats were more unified behind their candidate.
Other recent polling has shown competitive races in states that Republicans carried comfortably in the 2024 presidential election. CNN’s Harry Enten has highlighted Democratic gains in several Senate contests, including races in states such as Ohio, Iowa and Kansas.
That makes the historical Ruffini analysis particularly relevant to the broader polling debate: the question is not whether polls should be ignored, but how much weight voters, campaigns and political analysts should place on surveys conducted more than a month before ballots are counted.
Ruffini had previously examined a much larger collection of Senate surveys. His August review covered 3,312 polls from the 2018, 2020, 2022 and 2024 elections. That analysis found an average Democratic overstatement of 3.7 points, with 73 percent of individual polls overstating the Democratic candidate.
The issue has also surfaced in Michigan’s 2026 Senate contest.
During the Democratic primary, several surveys showed Abdul El-Sayed with a substantial lead over Rep. Haley Stevens. One Emerson College poll had El-Sayed ahead 54% to 39%. Yet the August 4 result was dramatically closer: El-Sayed received 48.5% while Stevens received 47.5%.
That result became another example of how difficult it can be for pollsters to model turnout, particularly in lower-turnout primary elections.
The general-election contest between El-Sayed and Republican Mike Rogers is now attracting national attention. Recent surveys have produced differing results, underscoring the uncertainty surrounding the race rather than establishing a single definitive trajectory. The Washington Post, for example, recently reported that former President Barack Obama plans to campaign with El-Sayed later in October.
Republican officials have nevertheless seized on the historical polling discrepancies.
NRSC National Press Secretary Bernadette Breslin said:
“Democrats should think twice before relying too heavily on the same rosy polling that has historically written Republicans off cycle after cycle, only to be proven wrong come Election Day,” Breslin said.
The debate extends beyond Senate races. In 2024, pollster John McLaughlin publicly questioned whether some surveys were adequately representing Republican voters. Speaking with Just the News, he argued that certain polls were sampling fewer Republicans and more voters who had supported Joe Biden in 2020. Those comments reflected his own criticism of specific survey methodologies, rather than a consensus among pollsters.
The current cycle has also produced conflicting evidence regarding turnout enthusiasm.
CNN’s polling analyst Harry Enten has recently pointed to strong Democratic enthusiasm and changing party identification numbers. In September, he noted that Democrats held an eight-point advantage in party identification in one analysis, a significant change from the previous year.
At the same time, other surveys continue to show extremely competitive races, demonstrating why national narratives can obscure substantial differences from state to state.
President Donald Trump has repeatedly argued that polling has underestimated his political movement. At a recent Ohio rally, he again criticized polling and urged supporters not to become complacent.
On Sunday, Trump told supporters that Republicans are going to “win big” against Democrats in November and “shove it up their ass.”
Trump also wrote:
“The Polls have ALWAYS underestimated MAGA. They try to suppress the Vote, but I won all 7 Swing States, the Popular Vote, 86% of Counties, and 99% of Primaries this year. I AM ON THE BALLOT! President DONALD J. TRUMP
From Dick Morris: “Since November of 2025 in monthly polls by McLaughlin, your job approval rating has always been between 47 and 50 percent. I have never seen anything like that in fifty years of polling. Almost every politician sees his numbers rise and fall each month but not you.
You won the election in 2024 with 49.8 percent of the popular vote and the latest McLaughlin Poll, taken this month, has your job approval at 49% – the exact same as your vote share. No other political figure has ever had this level of consistency in his ratings.
Through the Iran War and gas price rises and falls, your job approval is the same. It never changes. And 49% is enough to win – look at how that level of vote won in 2024! In the same Mclaughlin Poll, the generic ballot is 50-48 Democrat. That is also enough to keep the House Republican. Past history indicates that anything less than a four point margin for Democrats likely points to a Republican House victory. And almost all polls are showing Republican gains in the generic ballot.”
The larger lesson from the historical data is straightforward: early and late-cycle polling can provide useful information, but it is not the same thing as an election result.
The 2018-2024 record in Iowa, Michigan and Ohio shows that significant polling errors have occurred, particularly when surveys attempted to model an electorate more than a month before Election Day. The 2026 evidence, meanwhile, is mixed enough that neither party can reasonably treat every survey as definitive.
With 33 days remaining, the numbers will continue to change. Turnout, undecided voters, candidate movement and late-breaking events can all reshape these contests.
For voters and campaigns alike, the most useful approach may be to watch the underlying data—sample composition, likely-voter models, margins of error and trends over time—rather than treating any single poll as a forecast of what will happen on Election Day.
