Here's the thing: if my math is right, early voting for November 8 starts as soon as September 30*. This varies widely by state, with some not allowing early voting at all, so it's not something you want to generalize too much about.
And most early voting begins less than 3 weeks before November 8.
But if you remember back to 2012, you'll remember the polls in August pretty much predicted the outcome in November. Yes, there were shifts and concerns about debate performance and the "47%" video, etc. But the outcome didn't change all that much between the end of the conventions and election day.
And we always forget how many people have already decided, and voted, as early as a month before election day.
The current models at 538 have been moving toward Clinton since the conventions ended. When Trump started railing about Khizr Khan, the polls took off like a rocket. As I write, the Nowcast says Clinton has a 91.5% chance of winning, and would be expected to take 366 electoral votes. The polls-only forecast gives her an 80% chance of winning, taking 341 electoral votes. The polls-plus forecast, which Nate Silver considers the best prediction, gives Clinton a 74% chance of winning, with 313 electoral votes.**
Josh Marshall sees it this way:
First, Clinton got a sizable bounce out of her convention. That bounce appears not only to be persisting but actually growing. That's a big deal. It's still too soon to say we're in the post-convention period. I'd say we need about a week more to be there. But historically speaking where the polls are a week or two out of the conventions tends to remain fairly stable. There aren't a lot of opportunities to really change the game. The debates are the big exception. But for all frenzy, there have been few cases where the presidential debates have really reset the race. People say a lot that three months is a long time, that polls are only a snap shot of the race as it stands today. All true. But we're coming up on the phase of the campaign where polls really start to matter and become much more predictive of the outcome.
As I said, if you recall 2012, even Obama's "loss" in one of the debates wasn't really all that significant in November, by which time a number of people had already voted (as they will this year). And nobody thinks Trump is going to do well in the debates. Nobody.
Second, it's not just a Clinton bounce. Trump's support appears to have eroded significantly. What counts as 'significantly' or 'a lot' is relative of course. We're only talking a few percentage points. But national elections, especially in this hyper-partisan era, play out in a highly constrained band. And Trump has fallen below what I and I suspect most other observers consider a key benchmark, 40%.It's only fair to point out 538.com doesn't have Trump below 40% of the popular vote in any of their models. But the trend is very, very clear, which ever poll numbers you aggregate and analyze.
I think the early voting is the key. Campaigns still operate as if debates in October make a significant difference in November, or as if any other event will change the outcome (there's still speculation a terrorist attack or another Wikileaks release will change things. Anybody remember the last Wikileaks document dump now? Anybody care? Anybody notice a change in Clinton's numbers since that release?). I don't think it's too bold to say, while it ain't over 'til the fat lady sings, she's warming up in the wings.
*close enough, as they say, for government work.
**This was actually as of last night. As I post this morning, the models have been updated, and favor Clinton even more.