According to Gallop the percentage of people who identify as republican, democrat, and independent are:
D- 27% R- 28% I- 42%
Yet many polls have samples which don't represent this at all.
Taking 3 polls (that have info on party affiliation) from the last week:
Reuters/IPSOS, Oct 20th-22nd, national poll:
D-44% (17% oversample)
R-37% (9% oversample)
I-12% (30% undersample)
IBD/TIPP, Oct 21st, national poll:
D-38.0% (10% oversample)
R-32.8% (5.8% oversample)
I-28.8% (13.2% undersample)
The economist/YouGov, Oct 18th-20th, national poll:
D-38.1% (11.1% oversample)
R-28.1% (0.1% oversample)
I-33.1% (8.9% undersample)
If you take the YouGov poll, which has Biden at 47% and Trump at 40% and break it down by party you get:
Democrats: Biden - 85%, Trump - 4%, not sure -3%
Republicans: Biden - 5%, Trump - 86%, not sure - 2%
Independents: Biden - 39%, Trump - 42%, not sure 8%
Taking this party breakdown and then applying the Gallup party affiliations to it generates this result:
Trump = (0.04 * 0.27) + (0.86 * 0.28) + (0.42 * 0.42) = 0.428 = 42.8%
Biden = (0.85 * 0.27) + (0.05 * 0.28) + (0.39 * 0.42) = 0.4073 = 40.7%
The poll goes from Biden - 47%, Trump - 40% to Biden - 40.7%, Trump - 42.8%, and that's nation wide.
I know that pollsters use 'likely voters' as a metric rather than a direct proportion of each party from the population, but when new voter registrations are up substantially for republicans over democrats (see below) it seems that democrats are being oversampled in these (and many other) polls.
Math is not my strong suite, but given the numbers at rallies, the percentage of difference between signs, my money is on Trump winning, holding the Senate and maybe some gains in the House.