But consider this: The one pollster that consistently has Trump doing better than the national average—well enough to win—is The Trafalgar Group. They are often discounted as being to the right of the average, which is true. The Atlanta-based firm also happens to be the only polling firm that accurately predicted Trump’s margin in Pennsylvania and Michigan in 2016. So why are we so quick to discount the only pollster that got it right last time? (Right now, Trafalgar has Trump down 2% in Pennsylvania but up +1% in Michigan, and predicts to the Rich Lowry at National Review that the president will be re-elected).

Some of the pollsters who got those state races wrong in 2016 have adjusted their methodology to account for a lower black turnout and so on—but some, apparently, amazingly—have not. “Since then, some of the most prolific state polls, including Monmouth University’s, have started to weight their samples by educational attainment to make sure they are not falling prey to survey bias based on voters with more degrees being more likely to answer their phones,” the Washington Post reported. “Yet several firms have not, including Marist College, Mason-Dixon, and EPIC-MRA, which conducts Michigan polls for the Detroit Free Press.”

Then there’s the mysterious matter of voter registrations. Tom Edsall in the New York Times quoted a Democratic consultant about what seems to be a spike in the voter registration of the demographic group toughest on Democrats– non-college aged whites: “Since last week, the share of white non-college over 30 [years of age] registrations in the battleground states has increased by 10 points compared to September 2016, and the Democratic margin dropped 10 points to just 6 points. And there are serious signs of political engagement by white non-college voters who had not cast ballots in previous elections.”