Francesca Hong was riding high. The last five polls in Wisconsin’s gubernatorial contest had her winning by as many as 22 percentage points. She lost narrowly, and so did the pollsters.
The week before, the polls had shown Abdul el Sayed leading by a big margin in the Michigan Senate Democratic primary. He won narrowly, and both results have reminded people of polling’s problems.
Recommended Stories
We’ve heard all the explanations for problems before. Primaries are notoriously difficult to poll. An analysis by G. Elliott Morse and Mary Radcliffe of Senate and governor primary elections in 2014, 2016, and 2020 tells us that primary polls are off by 13 points on average.
There are a few potential reasons for these inaccuracies. Democrats on the Left are showing record enthusiasm this year, and too many of them may be ending up in pollsters’ samples. Estimating turnout in primaries and the composition of the electorate adds to the difficulty. Closely contested states with open primaries, where people from either party can vote, are another challenge. Pollsters also have trouble predicting late movement if they aren’t in the field.
Most pollsters are trying to address the problem. After the pre-presidential election poll problems in 2020 and 2024, the American Association for Public Opinion Research convened experts and practitioners to figure out what went wrong. They issued this comprehensive report.
Despite these impressive efforts, the new failures are what many people will remember — and, according to experts, we can expect more of them. Others also argue that the overall track record of presidential polling is good once you get reasonably close to an election.
Let’s take a look at how hard it is to poll today from two very different perspectives. The most recent highly regarded New York Times/Sienna poll surveyed 3,659 likely voters between June 15-29, reaching around 600 voters in six states expected to be competitive this fall. The pollsters were examining voter views at this early stage of the 2026 campaign in Senate battlegrounds of Alaska, Iowa, Maine, North Carolina, Ohio, and Texas.
How many interviews did the pollsters have to conduct for their poll? Here’s a portion of the New York Times’ detailed description of how it did the poll: “Voters were selected for the survey from a list of registered voters. The list contains information on the demographic characteristics of every registered voter, allowing us to make sure we reach the right number of voters of each party, race and region. For these polls, interviewers placed more than 467,000 calls to more than 174,000 voters.”
You heard that right, 467,000 calls to more than 174,000 voters using live interviewers! Pollsters in primary contests are often polling on a shoestring, in much shorter time frames, using smaller samples.
Here’s another example from a very different kind of survey. Government officials are deeply worried about declining response rates (in general, they are below 5%), the growing expense of good surveys, and budget cuts to government surveys.
AEI’s Michael Strain and Brookings economist David S. Johnson have both discussed the technical and political challenges to the nation’s vital federal statistical system. To take just one such survey, the Federal Reserve is one independent government agency that uses its Survey of Consumer Finances to understand how consumers are faring. NORC has conducted this survey since 1992. The survey is conducted every three years and is the gold standard in terms of in-depth information on Americans’ financial situation. The questions the survey asks require respondents to provide detailed information on their financial situations. I’d have to spend some time looking at my own finances to figure out how to answer some of the questions.
Getting people to respond to all the questions is a challenge. NORC surveys about 6,000 people every three years. Noting that researchers face declining survey response rates, increasing costs, and competition, NORC has been exploring different incentive strategies to try to address these problems.
The good news is NORC found that increasing paid incentives did not decrease survey quality — an important finding given how much the nation relies on these statistics and wants to get people to take the surveys.
I can’t help thinking that the kinds of election polling failures we saw in Michigan and Wisconsin contribute to people’s willingness to take any survey. Pollsters conducting surveys in hotly contested primaries have special challenges, as these recent elections show, and their failures may have spillover effects to other invaluable surveys the nation relies on.