hokiewolf said:
I've got news for you Civ, polls are also forward projecting.
No they're not.
They're a snapshot of sentiment at the time of the poll.
By definition they don't project anything.
Modelers, of course, incorporate them into their model, and modelers and everyone else extrapolate from them.
But a poll by itself doesn't project anything.
Polls are like taking a temperature reading. If I tell you it's 68 outside right now, that doesn't tell you what temperature it's going to be five minutes from now or at 2 PM or tomorrow. Snapshots aren't projections.
If your broader point is that polling trends matter and that from polling trends, you can derive some sense of how a candidate's electoral chances change over time, that's obviously true but it's so obvious as to not even be worth mentioning.
The distinction to be made between polls and models/projections Is that very small differences in swing state polling can lead to more outsize model projection differences that fluctuate significantly over time.
Silver's analogy regarding win probability for basketball teams is a great one. Team A scoring a bucket at the buzzer right before halftime may swing the win probability for that team from 47% before the bucket was made to 53% after the bucket, without changing the game's fundamentals or the observation that the game is still effectively a coin flip.
That's Kamala and PA. Her polling there and in all the other Rust Belt states didn't broadly get "much worse" like Bas is hoping/cheerleading for. The change in model prediction over the last week was almost entirely from a small shift in polling in Pennsylvania.