Quote:
Originally Posted by wheatrich
Graph showing it was a few days before instead of literally the day before isn't exactly pwning a lib here.
The ongoing graph indicates that it was more of a slow trend showing an increasing likelihood of Trump winning (getting up over 30% just prior to November before lurching back under it in the last couple of days) rather than this odd story you told of Nate Silver just waking up one day and putting his thumb on the scale.
During the last few weeks before that election, Access Hollywood wore off, Comey happened, and somebody finally got Trump to shut up for a prolonged period of days, which I was under the impression caused polls to tighten from where they were.
Admittedly I'm not inclined to conduct a research project into all of the late state polling, etc., but I do take Silver at his word that his model spits out what it spits out and that he's not just arbitrarily shifting numbers to project sudden gut feelings or whatever it is you think happened here.
Obviously probabilities by their very nature don't outright promise results of any individual one-time outcome. That doesn't mean that people should stop building predictive models.