What the Polls Got Right and Wrong in the 2025 Elections
At a time of maximum popular mistrust of major political institutions, the polling industry hasn’t been immune from battering. Indeed, now, as ever, we can periodically read angry screeds suggesting we get rid of polling — or at least campaign-specific “horse race” polling — altogether. So with another election in the books (albeit a limited off-year election in which some of the results were in scattered locations with no public polling), it’s worth taking a look at the polls’ relative accuracy.
For purposes of comparisons across elections, I’m going to use the RealClearPolitics polling averages since we have them for every election and they use the simplest methodology (straight, unweighted averages). According to RCP, the averages were pretty far off for the three marquee contests in New Jersey, Virginia, and New York City.
In the New Jersey gubernatorial race, the averages placed Mikie Sherrill up by 3.3 percent; she won by 13.1 percent. That’s a 9.8 percent error. Individual pollsters got closer or further away, as it happens. Of the polls taken after mid-October, Quinnipiac and Fox News were closest, showing Sherrill leading by 7 percent. AtlasIntel, Co/efficient, and Trafalgar-InsiderAdvantage (all reputed or proud to be Republican-leaning) were all far off, showing Sherrill a single point ahead of Jack Ciattarelli. All of the above polls were of likely voters.
In the Virginia gubernatorial race, the polls were closer to the results, though, again, most of them underestimated the Democrat’s performance. The RCP averages gave Abigail Spanberger a 10.2 percent lead; she actually won by 14.6 percent of the vote. That’s a 4.4 percent error, which is pretty typical for a non-presidential election. The only poll to nail the results was a late-October YouGov survey showing Spanberger leading by 15 percent. But the Washington Post–Schar, Echelon Insights, and Emerson were close, giving Spanberger a 12-point lead. The most inaccurate Virginia polls were an October 23 survey from Christopher Newport University and, more egregiously, an Election Eve poll from Trafalgar, both showing Spanberger winning by 7 percent. Again, all of these polls were of likely voters.
And in the New York City mayoral race, the Democratic winner, Zohran Mamdani, actually undershot the RCP polling averages, which had him leading by 14.3 percent (all the late polls were of likely voters); he won by 8.8 percent. The closest polls to the final result were from Quinnipiac and Suffolk, which both showed Mamdani leading by 10 percent. There was also an October 30 poll from AtlasIntel giving Mamdani a seven-point lead, but it was superseded by a last-minute AtlasIntel poll cutting his lead to 5 percent. The clearest outlier was from Emerson, which showed Mamdani leading by 25 points. All the October and November polls shared a common problem that affected their accuracy: They did not foresee the collapse of Curtis Sliwa’s support, which clearly cut Mamdani’s margin over Andrew Cuomo. The averages pegged Sliwa’s vote at 16.3 percent, and he actually got 7.1 percent.
So any narrative that the polls skewed Republican in 2025 has to take the New York numbers into account.
RCP didn’t publish averages for California’s Proposition 50 ballot initiative, the other big contest on November 4. But several pollsters were reasonably accurate in predicting the results, even though ballot initiatives are typically difficult to poll. An October 21 CBS survey showed the retaliatory gerrymandering measure ahead 62 percent to 38 percent. It actually passed by 63.8 percent to 36.2 percent. Two well-reputed California pollsters underestimated the popularity of the initiative a bit: PPIC showed it ahead 56 percent to 43 percent as of October 30. At about the same time, the Los Angeles Times–IGS survey was closer to the results, showing it winning by 60 percent to 38 percent.
For all the abuse pollsters take these days, the polls were relatively accurate in both 2022 and 2024, however they clearly underestimated Donald Trump’s vote in the latter year (though by less than in 2020). Because of the peculiarities of the New York City mayoral race (the principal challenger, after all, was a self-identified Democrat, albeit a renegade Democrat), it’s safer to say the polls and the pundits underestimated a national Democratic wave. But it’s 2026 and 2028 that may provide the acid test for the public-opinion research field when it comes to elections: Two straight elections without Trump on the ballot may create a return to polling normalcy.