This comes in an extremely detailed article by Bruce Gyory on The Bulwark. Gyory's novel approach is to take AP/NORC VoteCast for 2020, on the grounds the 2020 exit polls aren't so good (true), and compare 2020 VoteCast to.......the 2016 exit polls! What Gyory doesn't seem to realize is that the exit polls have been providing suspect data for many elections (not just this one) so it makes no sense to compare a good poll from 2020 to the 2016 exits.
That said, the article has a lot of interesting data from the VoteCast survey that is worth a look. Just don't take the 2016 comparisons very seriously.