Biden's Electoral College win probability down a bit, to 70%, after some mediocre polls for him this morning. He had been hovering around 73% in recent days. So not a big shift, but worth a quick Twitter thread. https://projects.fivethirtyeight.com/2020-election-forecast/

First, Biden does not appear to be getting much of a convention bounce. He's at +8.8 in our national poll average (still quite good) but that's only up from +8.4 before the conventions.

Our model expected pretty small bounces this year, but still it sees this small a slight net negative for Biden, as it adjusts polls taken during what's supposed to be Biden's convention bounce slightly downward.

Second, Biden has gotten some not-so-spectacular swing state polls lately. His lead in Florida is down to +5.4 and in Pennsylvania to +5.8, which remain the most likely tipping-point states. Not bad, but the Electoral College - popular vote gap would appear to be widening a bit.

The BIG caveat is that this is not based on the world's most spectacular set of polling, especially at the state level. And almost of it is online panels, with a robopoll thrown in here and there; basically no live caller polls since the pre-convention round from ~10-14 days ago.

That matters more during the conventions than it might ordinarily, because online panel surveys tend to make assumptions that can flatten out bounces, whereas live-caller polls tend to show bigger swings after major events.

Simplifying a bit, but online polls would allege that live polls suffer from more partisan non-response bias (e.g. Democrats are more likely to pick up the phone if they're excited after their convention) and so undergo phantom swings.

Live polls would allege that online polls bake in too many assumptions about the electorate that can make them miss real change. They may also say polls are a snapshot of the electorate, so if one party's voters are more enthusiastic at any given time, they should reflect that.

I am actually slightly more on the side of the live polls argument. I think partisan non-response bias can be a Thing, but also that the cures for it may be worse than the disease, and you may be trading Type I error for Type II error. But that's neither here nor there.

The point is, ordinarily we have a pretty nice *mix* of online and live polls, and it would perhaps be especially important to have a good mix of live and online polls around the conventions, but we don't have that now. The live polls come in waves, and we're in a low ebb now.

At the same time, there was a smattering of pre-convention live-caller state polls that weren't spectacular for Biden either, such as his being "only" +4 in Pennsylvania (Muhlenberg) and +4 in Wisconsin (Marquette) which were some pretty 2016-ish type numbers.

So we're at kind of a weird place where if you're Biden, you neither have all that much reason to be worried (wouldn't lose sleep over a bunch of Change/CNBC & Trafalgar polls) *or* all that much reason to feel secure (not a lot of great high-quality state polls for you lately).

It would have been nice to have a high-quality live caller poll sandwiched in between the conventions. And for all we know, it could have shown a big bounce for Biden. But that's a somewhat academic point because even if there had been a bounce, it might fade quickly anyway.

We are likely to see a lot of post-convention polling, including high-quality live-caller state and national polls, and I'd say it's pretty important for Trump that those show demonstrably tighter results than they did in June/July/early August.

It's not like there's some magic inflection point, but being down say 3-4 points in the tipping point states (versus ~6 before) would be a lot more tenable. That's within the range you can make up at the debates and also within the range of a "normal-sized polling error".