I prepared a NC state voting model based on county by county Early vote (which NC stratifies by Party Registration). Total Early Vote is 96 % of the 2016 Final Vote. NC records early vote by the voter’s party registration, which may be used with polling cross tabs to generate a reasonably confident prediction of the voting pattern.
We should note that the “banked” early Biden vote (base on party registration factoring) is over-running **total** 2016 Clinton votes by at least 186,000. That is to say, Biden has already pocketed 186 K votes more than the total final 2016 Dem tally. The model breaks down in Durham, Guilford, Mecklenberg and Wake counties, indicating the “Independent” lean towards Biden in those counties is likely greater than the default 51-42 split informed by the statewide NYT cross tab for independent voters (representing an undercounted 18,000 votes to Biden, and debiting the Trump sub-tally).
Trump’s Early vote is “underperforming” his 2016 final tally by 370,000 votes — but since they have explicitly discouraged early voting this is to be expected.
Final Voting turnout is unknown, of course, and cross tabs indicate “day of Election” voting is heavily skewed to Republicans. Final Voting participation may be estimated by NYT polling cross tabs which query on per party preference liklihood of day of election voting. These per party likelihood were use to expand the future day of election pools.
Expansion by statewide polling cross tabs continued to underperform for Trump in many counties (indicating that these are discouraged or switched votes), and validated by estimated turnout numbers coming in below 2016 tally in those county. I gave the Cheeto the benefit of the doubt and “tweaked” the voting back up to the total 2016 tally (ie no shrinkage at all). He still loses under that scenario, but it does come closer.
I put my spreadsheet up at: drive.google.com/...