Let's talk about what the GOP, and more specifically David Perdue, needs to do on Election Day in order to prevail. It's nothing impossible, in my view. But it's worth thinking through where the GOP burden stands after the strong Democratic early vote in Georgia

Let's stipulate for this whole discussion that absolutely no one has changed their vote since the election.
If so, then our estimate is that Ossoff will amass a lead of around 350k out of the advance vote, including what he'll net out of ~100k absentee votes still to arrive

There's some fuzziness on any estimate like this, of course. But the fundamentals here are pretty straightforward, and clear to anyone with a voter file, based on the partisan makeup of the vote and the vigorous Black turnout. It's confirmed in our NYT/Siena poll data as well

One rough way to see: in the gen, Ossoff won 51.6^ of the major party advance vote, and Dem primary voters represented 49.9 percent of the advance vote. In runoff, it was D53.1 yesterday, projected to 53.4 after mail done. That 3.5 pt shift + 51.6 = 55.1%; our estimate is 55.5.

So now Perdue would need 350k votes to fight back to a tie. What would that look like?
Well it would entail something more than Nov., when he won by 218k votes on Election Day with nearly 62% of the mjr vote. Of course, the strong Dem turnout means there are more GOPers left now

How much does that help? Well, let's take a model of Election Day vote in November, applied to the current set of votes who remain, and calibrate overall turnout at 81% of Nov. levels (early voting pace, adj. for holidays). That would mean 917k votes, Perdue at 68% and +323k

Perdue wins by more this time with the same turnout model since there are a bunch of extra highly reliable GOP voters hanging around who weren't out there heading into election day in November. But even this wouldn't be enough--would need something more than 917k eday voters

If you use the 2020 general election day turnout model, you wind up getting Perdue to a win somewhere over a million Election Day votes--more than November--as total turnout approaches 4.2 million (which doesn't seem crazy)

The challenge for Perdue, of course, is that this is a 2020 general election turnout model. So far, Democrats and esp. Black voters have plainly outperformed the general election in the early vote. If they outperform on Election Day too, then Perdue needs a greater turnout

One could make this argument the other way, though. If the GOP was particularly depressed in early voting in the runoff v. the general, there may be more upside for them to outperform the general election turnout model on Election Day

Anyway, I don't like using the early vote to predict election results and I will make no exception here. I do think it can show changes in the composition of the electorate, and directionally those changes seem favorable to Dems. Favorable enough? We won't know until it's over