Comment by fabian2k
8 hours ago
FiveThirtyEight gave Trump a 30% chance. Their reporting did make clear that with margin was within the range of a normal polling error. And sometimes you get more than a normal polling error.
It doesn't help that the US has a terrible election system that often leads to small margins in some states being decisive.
> FiveThirtyEight gave Trump a 30% chance.
I know I'm being super conspiratorial here but why wouldn't all forecasters predict just between 30% - 70%? That way if they're "right" they can take the credit for it and if they're wrong they can say "well, we weren't that wrong". That's probably what I'd do anyway...
It implies a close race or a strong reason to believe there's some sort of systemic polling miss, and if it's a blowout you still look pretty bad. Especially if you don't have some kind of good explanation for the miss/you keep making those kinds of misses frequently.
Also there's more going in those forecasts besides just the "% chance to win". There's expected results in terms of %'s of the vote for the candidates, and that's what people tend to focus on for actually analyzing your performance and credibility after the fact.
You getting the outcome correct but being off by 20 points on the margin is a much worse performance than you getting the outcome wrong but being within 0.5 points of the margin. (ex: Results are 49.75/50.25, you predicted 30/70, another outlet predicted 50.25/49.75).
538 claimed that they post-checked thousands of elections and their percentages were pretty close. (E.g., 30% chances happened about 30% of the time, million to one chances happened every single time)
Of course there's more than that. Predicting higher uncertainty than warranted would be a different failure in the model. But that didn't really happen in that election.
> I know I'm being super conspiratorial here but why wouldn't all forecasters predict just between 30% - 70%? That way if they're "right" they can take the credit for it and if they're wrong they can say "well, we weren't that wrong". That's probably what I'd do anyway...
For anyone making many predictions, you can analyze the outcomes to see how accurate those percentages are.
For anyone making few predictions, you should never trust their track record even if it's technically perfect.
You're completely right, although I believe this number was not decided personally. They just happened to pick algorithms that have this "nice" property because it will lead to the same result.