“I wonder if a pollster could use crowd turn out at campaign rallies as a measure other than robo calls.”
I think you have hit on a huge predictor. But the problem for pollers is they won’t get paid much for taking a picture and estimating a crowd. Biden has trouble filling three rows of tables at the VFW while Trump can fill stadiums with standing room only. (You’re the thread winner.)
It presents an interesting mathematical challenge. I suspect that if it could be “figured out” that the polling agent would be FAR more accurate.
But how to turn into a profitable enterprise? That level of analysis cant be cheap.