Quote:
Originally Posted by NickMPK
Were you taking into account the historically inevitable decay of convention bounce when you estimated Hillary at ~99% to win in August?
No, we didn't do any such corrections. I don't think that the post-Convention boost doesn't reflect an actual temporary boost in voter preferences. And we didn't want to make the assumption that it will necessarily be followed by a decay, just because such an effect has been observed in some previous elections. This effect has been getting smaller in the past few elections and I don't think there is very strong evidence that the effect is that strong or pervasive to begin with.
Quote:
Really, if -at any point- your model estimates one candidate to be at 99%, it should take some sort of once-in-a-century black swan event for than candidate to ever drop down to 60% at a later point. Hillary's drop between her convention and the first debate was more than average, but far from unprecedented.
Quote:
Originally Posted by TheNewT50
Another way of looking at it is that it would take a really historically horrible thing to happen to have a candidate go from 30% equity to nearly drawing dead in the span of two weeks eight weeks removed from the election.
You can look at win expectancy graphs in sports to see how rare and extreme an event it takes for that sort of shift to occur in anything but the final minutes of a game.
I don't want to repeat what heehaww already said, so I'll just add an example. If Real Madrid is leading Barcelona 2:0 in the 70th minute, you would objectively get a Barcelona win probability very close to 0%, which would be an accurate probability. If, however, Barcelona scores 2 quick goals between 71' and 74' (which happens rarely in football, but not that rarely either), their win probability would jump to at least 30%.
Given the events that happened starting (and including) the first debate and how they have been presented in the media/perceived by the public, this type of a sharp drop for Hillary isn't that surprising.
I guess it's fair to say that compared to other models (like 538's) our model is less conservative and I don't exclude the possibility that we're slightly overestimating Hillary's win probability. But the truth is that now she is leading quite significantly in many of the key states' polls and it's hard to see how Trump is going to recover from that in 3 weeks. The few national polls that show Trump leading have almost no implication for the final win probabilities, given how the US presidential election process works.
