I understand how you came to that conclusion, since those graphs are extremely confusingly designed*, but the 10% change is the predicted change in margin of vi

I understand how you came to that conclusion, since those graphs are extremely confusingly designed*, but the 10% change is the predicted change in margin of victory (which the article is examining because said prediction was wrong), and the actual change is more or less negligible, trending slightly towards more Republican. Behind the paywall, the article text confirms this: >However, there was scant evidence that white voters without degrees preferred Mr Biden to Mrs Clinton. Mr Trump’s margins of victory in white working-class counties this year were just as large on average, and in some places even bigger, than in 2016. * Honestly, they're astoundingly bad, to the point of being actively misleading if you're not looking carefully. I highly suspect they were thrown together by someone without statistics training using bad autofit at the last minute.