The New York Times is, once again, worried about the polls. In a piece last week, Reid Epstein brought up a number of concerns facing the polling industry, full of hand-wringing about whether or not the polls are blowing it this cycle. Epstein cites several issues in the piece, among them that campaigns are using unreliable polling data to try to shape their races, that prediction markets are skewing how people understand polls (and perhaps leading to an increase in shoddy or fake data), and that pollsters are increasingly relying on modes like text-to-web rather than traditional phone polling. And at The Washington Post, Henry Olsen asserted flatly last week that “Polls are getting worse.”
In another Times piece last week responding to news of a fake pollster, Median Strategies, whose “polls” had been reported on by a very small number of news outlets, Ruth Igielnik and Caroline Soler claim that the number of public pollsters is “skyrocketing,” noting that there are around 30 new pollsters so far this year. (Our numbers differ somewhat, because we have different aggregation standards and track different questions.) They say the increase in new, untested pollsters can make it difficult to judge survey quality.
But there’s a question amidst all this panic that the Times and Post seem not to have asked: Is any of this different from the past, or is this totally normal? If you want to claim, as Epstein does, that “misleading polling has shaped campaign narratives, donor behavior and news coverage like never before” — or, as Olsen does, that polls are getting worse — it might be worth looking at the data from past election cycles.
Paid subscribers to 50+1 get access to premium analysis, plus sortable tables and complete data access on our polling website. If you want to follow the 2026 cycle with the best data at your fingertips, become a paid subscriber.
Campaigns love sharing questionable polls
It is worth stating up front that using polls to try to shape media coverage is not a new campaign tactic, despite the impression one gets from the Times and WaPo stories. Campaigns have long been sharing unreliable or biased polls to gin up support and try to control the narrative of a race. It’s their whole jam! After news broke in Maine earlier this year of Democratic candidate Graham Platner’s Nazi-linked tattoo, for example, groups allied with the oyster farmer quickly released polling showing Platner still ahead, in an attempt to reassure supporters and stem the bleeding.
Campaigns have been using internal polling to spread information they want public for decades. For example, in 2024, Leah Askarinam outlined how Bernie Moreno’s U.S. Senate campaign in Ohio leveraged its internal polling, selectively releasing good news and holding back bad news, to win Trump’s endorsement. In that case, he laundered the polls through Politico, but the ultimate outcome is the same.
And it goes beyond just horse race polling and congressional campaigns: President Donald Trump has incorrectly cited inflated approval rating data from Rasmussen Reports (whose polls we have banned from our database). In past election cycles, he has also shared unscientific social media “polls” on his Twitter and Truth Social accounts that vastly overstated his support.
Shoddy polling being spread by politicians with an agenda is part of politics. What may be different in the modern era is the way these polls are spread — on social media, rather than in inside-the-beltway news coverage — but the fact remains that this is nothing new.
Fake data from prediction markets goes back almost a decade
Prediction markets, and fake polls trying to influence them, have also been around for a long time — like, since the 1980s. We know how to deal with them. Sure, the success of Kalshi and Polymarket in recent years has propelled the prediction market industry to greater prominence than ever before. But even before the rise of those two giants, there were fake surveys released specifically to swing the markets, and with much wider distribution than the Median Strategies affair.
In 2017, a series of “polls” from a group calling themselves “Delphi Analytica” were released. The most ludicrous of these surveys showed then-Sen. Debbie Stabenow losing a U.S. Senate race to Kid Rock, and was widely shared in right-of-center media outlets including the Daily Caller and Gateway Pundit, tweeted by sitting Texas Gov. Greg Abbott, and posted by Kid Rock himself. The “polls” were actually produced by bettors from the most popular prediction market at the time, PredictIt, and temporarily influenced the market on Stabenow’s reelection.
But professionals, like our team here at FiftyPlusOne, the Times polling team, and various other aggregators, know about these sorts of problems and have mechanisms in place to screen out bad actors. When we make mistakes or learn new information that causes us to doubt previous decisions, we respond quickly and publicly to explain what happened and why. It takes more than a random amateur cooking up fake data in a college dorm to fool us (though we can’t control how other media outlets cover such polls).
Is text-to-web a worse way to conduct surveys?
The Epstein article includes a quote from Rep. Haley Stevens saying that part of the problem with Michigan Senate primary polling was that pollsters are no longer using live caller phone surveys. We do not find a lot of evidence supporting this claim.
But the concern about a drift away from live caller polls is one worth pausing on for a moment. Pew Research has documented how quickly the industry has shifted away from live phone polls, with nearly all pollsters using the mode in 2000 — by 2022, that number had shrunk to six.
Many of the moves away from live phone polls were actually encouraged by the polling industry writ large. Fewer people answer their phone than in previous decades, and using texts to reach respondents may help with recruiting Americans who are difficult to reach, like younger voters (as the Times notes). And in their 2024 autopsy of the polling industry, the American Association for Public Opinion Research (AAPOR) found that “mode of data collection … showed weak associations with error,” but text and live phone polls didn’t differ very much in their average error.
To gut-check this, we looked at all the 2026 Senate primary polls in our database for races that have already happened where we know the mode of data collection (some pollsters do not include this information in their poll releases) that were conducted within three months of the election. There are no surveys in that dataset that were exclusively fielded with text-to-web (TTW), though many included TTW as part of a multimodal methodology. Among the 84 surveys that included TTW, the average absolute error on the margin between the top two candidates in the poll and the actual election outcome was 15.4 percentage points, not too far off from the average 13-point error we expect to see in Senate primary polls.
Among the 63 polls that included live callers, either as the only mode of reaching respondents or as part of a multimodal approach, the average absolute error was 14.9 percentage points – almost the same. Of the 40 polls that included an online opt-in panel as part of their method to reach respondents, the average absolute error was 15.5 percentage points.
There were 23 polls that were conducted exclusively with live callers, and these performed a touch better than the multimodal polls, with an average absolute error of 11.7 percentage points. We saw similar results in 2024, though there were far fewer Senate primary polls that cycle — just 50 polls were conducted within 3 months of a Senate primary and disclosed their methodology. Of those, we see around 15 percentage points in absolute error on the margin between the top two candidates in polls that include text-to-web (14.7), live phone (15.9), and online opt-in panel (15.0), but just 7.5 percentage points for polls that are exclusively live phone (take that with a grain of salt: there are only 4 such surveys).
All this may be a sign that including more live phone responses in primary polling could be helpful for pollsters that can afford it. Primary voters tend to be older and more engaged — exactly the kind of voters that answer phone polls more often. But the AAPOR results from last year suggest this may not replicate in the general election.
Are there really a lot of new pollsters this year?
There’s also no evidence that new pollsters are flooding polling aggregates at a higher-than-normal rate. So far this cycle, we have added 65 brand new pollsters to our database. That may seem like a lot, but it’s only three more than during the 2024 cycle, and less than the 75+ new pollsters we saw in 2020 and 2016. While it’s more than we’ve seen in a midterm cycle going back to at least 2010, the differences aren’t all that large — there were 57 new pollsters in 2010, for example. In addition to totally new groups, we’ve seen 12 new collaborations between existing firms (these are often bipartisan teams working together in an attempt to produce more neutral results), a number that’s also in line with historical precedent.
And not every one of those are actually “new” pollsters — this analysis counts every pollster whose first poll in our database was released this cycle. But many of these pollsters are well-known and established firms who just hadn’t yet released survey data that included a question we track (like presidential approval or horse race polling). BSP Research, for example, has been around for 20 years producing surveys mostly focused on Hispanic and Latino voters, but their first poll asking a question we track in our database was released earlier this month. Same with the University of Maryland Program for Public Consultation, whose issue polling we’ve cited many times over the years (including in last week’s Trendline!), but whose congressional approval question in March was the first to make it into our database.
We still have two months to go before the election, which means more new pollsters could (and probably will) show up. But as of now, we don’t have any evidence that the number of new pollsters this cycle is alarming or unusual in any meaningful way. While that might not align with many people’s perceptions of the state of the polling industry, this sort of churn isn’t really out of the ordinary.
There’s no need to panic
The upshot: All of these problems have been around for a long time. And the Times should know that, because they seem to write at least one article every cycle about how the polling industry is blowing it again. Here’s one from 2024, 2022, and one of several from 2020. And they aren’t the only ones — Politico publicly frets at least once per election, too.
In fact, a lot of the concern going around in the last week seems to have sprung up from what appears to be a total coincidence. In a very short span of time, we saw two high-profile (but only slightly larger-than-average) polling misses (in Michigan’s Democratic Senate primary and Wisconsin’s Democratic gubernatorial primary), a completely fake pollster, and some shenanigans from existing pollsters. While, as we’ve documented here, all of these issues are pretty par for the course, their confluence during the dog days of August seems to have catapulted some normal issues into the spotlight. And despite all the hand-wringing we see each cycle about the slow death of the polling industry, pollsters in 2022 had one of the most accurate midterm elections in the history of scientific polling.
You don’t have to trust any individual poll, or any individual pollster, to give you the best read on a given race — and actually you shouldn’t! We maintain polling averages precisely to avoid that problem. And while we’ve been banging the drum about high average error in Senate and gubernatorial primary polls for months, general election polling is typically much more accurate. Of course, we can’t know yet just how accurate the polls will be in November, but there’s nothing especially troubling about the polls this cycle. Polling is getting a lot harder, and much more expensive, but there is no evidence it is getting worse.
Paid subscribers to 50+1 get access to premium analysis, plus sortable tables and complete data access on our polling website. If you want to follow the 2026 cycle with the best data at your fingertips, become a paid subscriber.
So here is my issue with polling. The problem isn’t with methodology, the problem is we are so inundated with spam and junk texts/emails/phone calls that I now consider anything not from a known contact to be potential phishing or even worse like malicious code insertion. As such, not even the most reliable pollster has any chance of reaching me.
This is more of an opinion take on what should be an easily determined data point. I could not click on the WAPO hyperlink that said "Polls are getting worse" because of paywall but perhaps they showed how accurate the polls have been to the actual result over the years and that would give an answer more than theoretical arguments about type of polling.
I have my own election data and for example if you look at the swing state results in the last three presidentials that Trump was in, the average bias (poll margin vs actual vote) in the final RCP aggregates were Dem 1.9, Dem 1.4, and Dem 1.8. So really no change. And of note in the two Obama elections the bias was R 0.87 and R 2.4.
Again, before parsing out the different reasons for inaccuracy, what is the actual trend. I dont think it has changed much.
btw I have favored RCP as much as I dont care for their right bias reporting, their numbers and/or predictions have had only a slight Dem bias since 2016 but less so than Silver or Sabato. They got two states wrong in 2024, MI and WI where the others got more eg Sabato also called NV and PA for Harris and I dont have Silvers breakdown but he had Harris favored by a tiny margin.