Regarding election day coverage, 538 did report that some districts in Florida were doing very well for Clinton, compared with 2012. But they quickly reminded people that this could just mean that the votes were distributed differently.
Regarding election day coverage, 538 did report that some districts in Florida were doing very well for Clinton, compared with 2012. But they quickly reminded people that this could just mean that the votes were distributed differently.
It's also a fact that we now know that the polls themselves had some kind of measurement or sample error, and that whatever this error was, it or something like it also happened in the Brexit and Canadian election.
I'll take facts over punditry any day, even if they have errors, because errors can be corrected more easily than opinions.
Some people are going to say we need better models, but perhaps the best thing to do is to quit relying so much on polls and counting. Predicting the future is a dangerous business even if you're "right" most of the time.
A more relevant point the article made is that the continuous daily updates of the probability didn't mean much to real people. They may have represented real facts about the world in some sense, but they still weren't that useful to people in terms of keeping them updated. It's not very useful for a polling site to predict <10% chances right up until the very end, where it raises to 29%, and the 29% chance comes through.
Again, please very carefully note my distinction between useful and true. Noise wiggles and twitches months before the main event being continuously interpreted in the press aren't facts; they're just that "factiness" that feels like you're being informed, but you really aren't.
I did it too; it's not like I'm immune. I can't even promise I won't do it again. But it's not useful to the vast bulk of people.
538's numbers were very bouncy between 10% and 50%, and they took some heat for that.
http://projects.fivethirtyeight.com/2016-election-forecast/#...