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."

"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."

"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."

RealClearPolitics straight averages were used to compare poll accuracy across recent off-year contests. In New Jersey the RCP average showed Mikie Sherrill up by 3.3 points while she won by 13.1 points, a 9.8-point error. Post–mid-October Quinnipiac and Fox News showed Sherrill +7; AtlasIntel, Co/efficient, and Trafalgar-InsiderAdvantage showed Sherrill +1. All cited New Jersey polls surveyed likely voters. In Virginia RCP averaged Abigail Spanberger +10.2; she won by 14.6, a 4.4-point error. A late YouGov had Spanberger +15, and Washington Post–Schar, Echelon Insights, and Emerson showed about +12.

Read at Intelligencer