All that tells me is that 538 is more reflective of the polls at face value than those other outlets. Those are p. tight right now so 56% seems normal?
I guess it's up to the individual to decide if that's for them or they are holding out hope on turnout and ground game.
This is the concern, yes. Obviously you shouldn't be doing an "if the election were held today" type chance of victory calculation as an election forecast. Because the polls are going to drift. You know that the polls right before the election are pretty likely to have moved from the polls in July, and so even if one candidate is so ahead in July that the outcome would be really certain if the election had been held then, there's a lot more uncertainty as to who's going to win the actual election months later. Because you know that the polls move, and because random movement in the polls is more helpful to the person who's losing, earlier predictions from the same polling position should be closer to 50%.
What's concerning about 538's model is that it's 80% Clinton at the start of July, then 50% Clinton a month later, then 90% Clinton a few weeks later, and now back to 57% Clinton a month later. This is saying that lots of really important and weird things are happening in the race. It's
not, or at least it shouldn't be, saying that Clinton was a bit ahead in the polls, then tied, then way ahead, and now she only has a slight lead. It's saying that,
whatever was going to happen between July and November, Clinton was very likely to win. But then a month later it's saying no that was wrong, something happened to give us more information on the race and now it's anybody's game. But then almost immediately after that Trump's chances of winning were reduced to basically nothing. No, wait, now Clinton is only a slight favorite and Trump is back in this. I know they like to use sports analogies, so what 538's model is saying has happened in this race is that one team pulled out to a big early lead, then the game was tied up, then the first team
again built up a nearly-insurmountable lead, and then the second team almost caught up again. This doesn't happen very often. A team that gets a big early lead is likely to win the game just because it's hard for the other team to catch up. It's true that sometimes you do get a really exciting game with reversals and comebacks left and right, but those are rare, and the issue here is that we're theorizing about what the score really is and how hard it is for people to come back without having direct access to that information.
Their polls-plus looks a lot less like a roller coaster, though of course what's going on here is just that the stuff they're particularly known for is being weighted less.
But their polls-only actually just looks an awful lot like their now-cast. You can see that this would look a whole lot nicer if they'd just added in a huge amount of uncertainty early on (biasing towards 50%), slowing decaying until today. That doesn't mean that it really was this uncertain the whole time, but that seems a lot more likely than all the reversals implied by their model.
