I put this up on twitter, but I might as well write a post about it: I feel that there's pretty good information out there about the rules and extent of early voting (see Michael McDonald's excellent work tracking voting-so-far, plus follow links to his other work on this), but a lot less about how the various public opinion polls handle early voting and how the various predictor models out there that are primarily based on polls handle it.
Or maybe it's out there and I just haven't seen it. But for example Nate Silver today, in a very nice state-of-the-race post, noted in a short paragraph that there are lots of people voting over the next few weeks...but as far as I know, he hasn't detailed how his model handles it. Nate Cohn's similar post didn't mention it at all.
I vaguely remember that pollsters deal with it, but I don't really remember what I'm supposed to know about it. Hasn't mattered yet -- there are just a handful of people who have voted so far -- but in two or three weeks, it's going to start becoming a huge deal. When I brought this up on twitter, Josh Putnam pointed out that part of the problem is that there's limited reporting on all of this by some of the states. So there's that, too.
I don't know; I guess if there is something out there about it, I'd appreciate a pointer from someone; if not, I hope that one of the polling aggregator/polling-based prediction folks can give us a thorough discussion of the issue. Presumably, early voting makes the election a whole lot easier to "predict" from this point on than if there was no early voting, right? But I don't have any idea how much easier, or what kinds of error we should be looking out for.