this almost seems like it was written in reaction to me.
i don't have any particular gripe. i don't think people vote based on their education level, and it's only useful in any analysis as a proxy for ideology. so, i ignore it because i think it's causally irrelevant. remember, kids: correlation does not imply causality. but, i appreciate the deconstruction.
and, i also think there's been a measurable movement of both blacks and latinos out of the democratic party over the last 20 years, and trying to balance it out by pointing to averages is just going to obscure it. i'm generally critical of approaching polling with averages; in the end, you get a snapshot on election day that may or may not reflect trends and averages.
i just want to point out that we saw a weird phenomenon in 2016 where all of the polls suggested white women would support clinton, and then they didn't. that would be a bizarre bradley effect, huh? i may have presented the answer in the above paragraph; they may have polled "educated whites" in california and tried to project the results into georgia, which would be daft; i don't remember figuring that out, four years ago. but, the consensus is that some kind of weird patriarchal effect set in; that, in the end, it was wives, especially, that voted with their husbands.
there was a measurable disconnect in 2016, at least, between polling and exit polls. so, careful.
https://fivethirtyeight.com/features/a-big-chunk-of-white-americans-with-degrees-and-people-of-color-are-behind-trump/