A few percentage points can separate victory from defeat.
For election pollsters, capturing that difference has been a challenge for decades, and as midterms in Iowa approach, voters face a familiar question: How much can the numbers actually tell them?
The answer depends partly on whose opinions the polls capture and who gets left out.
RELATED: Iowa campaigns adapt advertising strategies for the digital age
Joshua Clinton, political science professor at Vanderbilt University, chaired a task force evaluating the 2020 pre-election polls and serves as a senior election analyst at NBC News.
In an interview with The Daily Iowan, Clinton said the underlying issue in modern day polling is simple — who is responding.
A September report from the Pew Research Center supports this concern, finding growing evidence Republicans are less likely to participate in polls. Clinton said those differences can skew results, making it difficult to distinguish changes in public opinion from changes in who answers surveys.
That uncertainty follows Iowa voters into the final weeks before the midterms, when interpreting polling results requires looking beyond who is ahead to how the numbers are produced.
“Polling is a social science, not clairvoyance, and no responsible pollster can guarantee an election outcome,” longtime Iowa pollster J. Ann Selzer said in a September article published by the Foundation for Individual Rights and Expression.
As Iowa voters weigh polling results just weeks ahead of the midterms, understanding how those numbers are produced is crucial to interpreting them.
Polls offer a measure of public opinion, but differences in who responds, how results are adjusted, and ultimately who votes can shape their accuracy — and limit what they reveal about Election Day.
Uncertainties when polling the Iowa midterms
Recent Iowa Senate polls can illustrate how results can vary.
A Rasmussen Reports survey conducted Sept. 28-30 showed Democratic U.S. Senate candidate Josh Turek ahead of his Republican opponent Ashley Hinson, 47 percent to 43 percent. On the other hand, an Emerson College Polling/Nexstar Media survey conducted Oct. 5-6 showed Hinson ahead 47 percent to 45 percent.
Polls have also offered different pictures of Iowa’s gubernatorial race.
A Suffolk University survey conducted Oct. 1-4 showed Democrat Rob Sand leading Republican Zach Lahn by 12 percentage points, 55 percent to 43 percent. An Emerson College Polling/Nexstar Media survey conducted Oct. 5-6 put Sand’s advantage at four points, 49 percent to 45 percent.
Frederick Boehmke, a professor of political science at the University of Iowa, said differences in how polling organizations collect their responses, adjust their samples, and identify likely voters can produce different results, a phenomenon known as “house effects.”
“By making different decisions, that can make some polls overrepresent certain types of voters relative to other polls when they release their final results,” he said.
The uncertainty also extends beyond a poll’s reported margin of error. Boehmke said the figure does not capture every challenge involved in producing a representative sample, including differences between people who respond to surveys and those who do not.
Another uncertainty is who will actually cast the ballot. Clinton said national attention and competitive races could increase participation in Iowa, making past midterm voting patterns less useful for identifying likely voters.
“They’re trying to figure out who’s likely to vote, so maybe past voting in midterm elections may be less predictive this time around,” he said.
Clinton said those decisions make it important to consider multiple polls and place too much confidence in any single result.
“Don’t get too despondent or don’t get too overjoyed based on what the polls are showing,” he said. “There’s lots of ways those numbers could change, and pollsters are making a lot of decisions behind the scenes that could really move those numbers around.”
Campaigns also use polling results to rally supporters and raise money. In a June 27 Substack post, Turek highlighted a poll showing him ahead of Hinson.
“The big takeaway is, of course, that I’m LEADING Joni Ernst’s handpicked successor, Ashley Hinson, by two points,” he wrote before asking readers to contribute to his campaign.
Lahn’s campaign similarly promoted his victory in the Iowa State Fair straw poll. In an August statement, his campaign spokesperson Dan Capodilupo said Sand “got crushed in the straw poll” and predicted the same outcome in November.
Beyond individual surveys, election analysts assess the broader competitive landscape.
The Cook Political Report, an independent, nonpartisan newsletter analyzing U.S. federal campaigns, elections, and political trends, moved Iowa’s U.S. Senate race from “lean Republican” to “toss up” and the governor’s race from “toss up” to “lean Democrat” in August. In October, Cook pollsters also shifted the 1st Congressional District from “toss up” to “lean Democrat.”
Who is behind a poll — and whether it aims to measure or influence opinion — matters when evaluating its results.
Boehmke urged readers to examine who conducted and funded the poll, how respondents were selected, and how results were adjusted. He also distinguished surveys intended for public opinion from “push polls,” which use leading questions to influence respondents.
“Those kinds of surveys have biases either because they’re associated with the party or a group that’s associated with the party, and they want to make it look good for their party, or the whole purpose of the survey is actually to try and influence opinion,” he said.
Behind the numbers
Before a poll produces a percentage, researchers must decide who to survey, how to reach them, and how to make their responses reflect a larger population.
Boehmke, who taught Hawkeye Poll, a public opinion survey conducted by the UI’s department of political science, said the experiential learning class gives students experience with the process. Students develop questions, work with the university’s Center for Social Science Innovation and a vendor to collect online responses and analyze the results.
“There’s the challenge of hoping you’ve made good choices to produce results that are reflective of whatever the population or the target group is that you’re trying to talk about, Iowans or people in the U.S.,” he said.
But reaching respondents is only part of the challenge. Declining participation makes it harder to determine whether those who answer share the views of those who do not, Boehmke said.
Pew Research Center reported its telephone survey response rates fell from 36 percent in 1997 to the low single digits in recent years, increasing the risk of bias when respondents differ from people who do not participate.
Pollsters use a process called weighting to adjust how much different respondents count toward the results. Clinton said those adjustments can help a sample better reflect the population researchers want to study.
He said pollsters adjust survey results to better reflect the people they expect to vote on Election Day, accounting for groups overrepresented or underrepresented among respondents.
Clinton said election polls face another complication: identifying likely voters. Researchers can consider voting history and ask respondents whether they intend to vote, but the electorate remains uncertain until ballots are cast.
“We don’t know what Iowa is going to look like in terms of who actually votes,” he said.
A history of polling errors
Arguments over polling’s inaccuracies stretch back decades, including to the aftermath of the 1948 election where polls incorrectly predicted Republican Thomas Dewey would defeat Democratic President Harry S. Truman.
“My hope is that next time polls are on the wrong side in a close election there will be less shouting and name calling,” George Gallup said in an address to the World Association for Public Opinion Research around 1949, according to the Gallup institution.
Today, pollsters face a similar challenge: explaining how surveys can offer useful insight into public opinion while still missing the final result.
Most polls underestimated support for President Donald Trump in 2024, marking the third consecutive presidential election in which pollsters underestimated his support, according to the Pew Research Center.
In Iowa, the final Des Moines Register/Mediacom Iowa poll showed Democrat Kamala Harris leading Trump 47 percent to 44 percent. Conducted Oct. 28-31, 2024 among 808 likely voters, the survey had a margin of error of 3.4 percentage points. Trump ultimately carried the state by 13 percentage points.
Polling errors can also matter even when surveys identify the eventual winner. In 2020, polls correctly pointed toward a former President Joe Biden victory but overstated his advantage over Trump, according to Pew Research Center.
Clinton said a big part of the challenge is accounting for voters who do not respond to surveys.
“If some people just won’t answer the phone, then you can’t get them, and you can’t account for their views,” he said.
The methods pollsters use to collect opinions have also changed. Boehmke said surveys have evolved from face-to-face interviews and mailed questionnaires to telephone polling and online surveys.
Some polling companies now recruit panels of people willing to answer surveys online, then draw samples from those participants. That shift creates new challenges in ensuring respondents reflect the broader population, he said.
As polling methods evolve, Clinton said voters should recognize their limits without letting the numbers determine whether they participate.
“You should act and vote in ways to kind of make sure that your voice is heard,” he said.
Other Articles
