All pollsters only speak to a sample of people in the US population, and those samples generally aren’t enormous — often around 1,000 voters. That sample should reflect the state’s population, with a proportionate mix of genders, races, and economic backgrounds. When one of those proportions is off, pollsters “weight” the sample by giving more statistical value to the responses of voters with that trait.

As it happens, voters with a college education are more likely to take a survey in the first place. The obvious solution is to weight those respondents less, and this year, some pollsters are doing just that. But many are not. The New York Times reported that this year, about 20% more pollsters are weighting by education, but that’s still fewer than half of the total.

Some of these changes got a test-run in the 2018 midterm elections. When the results came in, CNN’s Harry Enton wrote, “public polling passed this test with flying colors.” FiveThirtyEight’s Nate Silver unenthusiastically agreed, writing that “they’re about as accurate as they’ve always been.” In his view, polls have always been stronger than people give them credit for.

Yet some still worry about polling bias toward Democrats. Real Clear Politics election analyst Sean Trende wrote that “pollsters have consistently underestimated Republican strength.”