You'll notice that the great bulk of election commentary that uses voter group survey results (as opposed to county data) relies on the exit polls. That's unfortunate because the exit polls have many problems (especially on sample composition) and this year have produced some particularly fanciful results. Of course, the exit poll folks are now merrily reweighting their data to put it in somewhat closer touch with reality but the whole exercise does not inspire confidence.
Are there alternatives? Maybe in a couple of months we'll see some publicly released Catalist data and months after that I think we'll be able to generate States of Change estimates for 2020. Both data sources will be able to provide high quality estimates that can be compared across elections to see how different voter groups shifted 2016-2020.
I have been doing some informal comparisons using the AP VoteCast data which I believe to be of higher quality. but VoteCast just stared in 2018 so there is no 2016 data available. I compare to 2016 States of Change estimates, but that is data that is not publicly available.
The Financial Times has done something useful that I can recommend at this point. They have generated 2016 and 2020 voter group estimates by essentially averaging across different available post-election data sources. This is not perfect but it beats looking at the exit polls alone. Their results are worth checking out in detail, but I reproduce their key chart below. The story told by their data is generally pretty consistent with the analysis I have presented here.
Note: for those interested in exit poll data in all their glory, I provide links to an excellent compilation by Karly Bowman of historical exit poll data and a very detailed piece by Bill Frey comparing 2016 and 2020 exit poll results with some very nice graphics and tables. Don't say you haven't been warned though!