Nearly 70% of national election polls in 2024 adjusted their results using political party affiliation or how respondents said they voted in a prior election, a once-controversial practice that Pew Research Center says can modestly improve accuracy.
The change reflects a polling environment in which response rates have collapsed and Republicans have become slightly less likely than Democrats to participate, creating risks that ordinary demographic weighting cannot fully correct.
Pew's report examined its own surveys and outside research to explain when political weighting helps, how pollsters apply it and why the technique cannot eliminate every source of error.
Why demographic weighting is no longer always enough
Pollsters have long adjusted samples to match reliable population estimates for age, race, ethnicity, education, gender and geography. Those targets help correct for groups that are underrepresented because they are harder to reach or less likely to participate.
Telephone survey response rates at Pew fell from 36% in 1997 to the low single digits in recent years. The New York Times/Siena telephone poll reported a 0.4% response rate in 2022, while Pew estimated an overall response rate of about 3% for its American Trends Panel after recruitment, enrollment and later nonresponse were considered.
Low participation creates a problem when the people who respond differ politically from those who do not. Pew cited evidence that Republicans have become less likely than Democrats to participate in some polls, especially around presidential elections.
That gap can remain after a poll is weighted for conventional demographics. If older White men who answer a survey are more Democratic than older White men who refuse, adjusting the sample only for age, race and gender will not remove the partisan imbalance.
Pew said that imbalance may have contributed to polls understating support for Donald Trump and other Republican candidates in 2016, 2020 and 2024. A post-2020 review by the American Association for Public Opinion Research found unusually large errors in national and state election polling.
Errors were generally smaller in 2024, when political weighting was more common, although most polls again underestimated Trump's support. Pew noted that average errors are often less than 5 percentage points on the margin between candidates but can still determine perceptions in close races.
In Pew's own 2020 and 2024 postelection polls, weighting by party affiliation improved candidate estimates by about 1 to 2 percentage points, or roughly 2 to 3 points on the margin between candidates.
Political weighting can also reduce short-term swings caused by who agrees to take a poll after a debate, convention or major controversy. Panel data have shown that campaign events can temporarily change partisans' willingness to respond even when their underlying views have not permanently shifted.
Three methods address different problems
One approach asks respondents whether they identify as Republican, Democratic or neither, including whether independents lean toward a party. Pollsters then align the sample with an outside estimate of the national partisan balance.
That approach lacks the kind of federal benchmark available for age or race. Pollsters may instead use Pew's National Public Opinion Reference Survey or their own address-based surveys to estimate party affiliation.
A second method asks prior voters which candidate they supported and then aligns the survey with official turnout and election results. Pollsters must decide how to handle people who were ineligible, did not vote or decline to disclose their choice; research suggests estimating a "synthetic past vote" can work better than excluding them from the adjustment.
The third method uses voter files. Those records can show registration, party enrollment in states that record it, voting method and participation in partisan primaries, although they do not reveal a person's chosen candidate.
Voter-file vendors and pollsters may also model party affiliation or past vote using residence, primary participation, age, gender, race, ethnicity and other characteristics. Pew said some organizations used modeled party measures in 2024.
Pollsters sometimes combine these techniques. The best choice depends on whether the target population is all adults or registered voters, whether the survey is about politics, whether it is national or local and whether a credible benchmark exists.
Survey panels can separate old and current views
Party affiliation can change faster than demographic characteristics, creating a danger that a poll will be forced to match an outdated partisan benchmark. Pew said survey panels offer a way to limit that problem because the same people can be asked about affiliation more than once.
In 2025, Pew measured party affiliation for members of its American Trends Panel around the same time as its national benchmark survey. That benchmark found 46% of Americans identified with or leaned Republican and 45% identified with or leaned Democratic.
Pew asked the same panelists again from May 26 to June 1, 2026. It weighted the new survey using each participant's 2025 affiliation, matching the measurement to the time of the benchmark, but reported political results using participants' current 2026 affiliation.
The later poll found 45% Republican or Republican-leaning and 49% Democratic or Democratic-leaning. A separate 2026 national reference survey conducted around the same time found a similar four-point Democratic advantage, at 43% Republican or leaning Republican and 47% Democratic or leaning Democratic.
The process allowed the survey to remain anchored to a reliable benchmark without suppressing genuine changes in how people currently identified. One-time surveys cannot make the same past-and-present comparison because they have only a current measure from each respondent.
Weighting helps on average but does not perfect a poll
Pew emphasized that political weighting improves average accuracy, not every estimate. It can make little difference on some questions and can slightly reduce accuracy on subjects such as personal finances.
Accurately matching the Republican share of the population also does not help if Republicans who answer a poll are unrepresentative of Republicans overall. The 2020 AAPOR review found that weighting by partisanship or 2016 vote did not make polls more accurate on average than standard demographic weighting, although it may have reduced some errors.
Political weighting makes the most sense for low-response surveys measuring attitudes or behavior connected to politics. It may be unnecessary for subjects with no meaningful political relationship or for high-response government surveys such as the American Community Survey, which Pew said has compulsory participation and a response rate around 83%.
Some topics that appear nonpolitical still carry partisan differences. Pew pointed to 2021 health surveys that overestimated COVID-19 vaccination rates without weighting for party affiliation, when Democrats were more likely than Republicans to be vaccinated.
Past-vote weighting can also correct the balance between voters and nonvoters, because politically engaged voters are more likely to participate in surveys than people who do not vote.
The need for political weighting could change if partisan response differences fade. For now, Pew said declining response rates and evidence of differential partisan participation make careful adjustments useful, while representative recruitment and accessible survey participation remain essential.
FingerLakes1.com is the region’s leading all-digital news publication. The company was founded in 1998 and has been keeping residents informed for more than two decades. Have a lead? Send it to news@fingerlakes1.com.