Comment by afavour
9 hours ago
I think this gets to the core of why a lot of this election prediction stuff doesn't work. People just don't parse the numbers the way the authors intend.
FiveThirtyEight had Trump at a 30% chance of winning, and he won. The model wasn't wrong. The less likely of two outcomes occurred. Even if they'd had him at 1% they still wouldn't technically have been wrong though I think complaints might be more warranted.
If they had Trump at 49% would you have still been angry? What about at 51%? Would it have been okay then?
Technically this is right. But if that is the case (and it seems to be), then a coin flip is better than their models. Because we only care about the current election, not a sequence of 1000 elections (which will not happen, by the way).
So use a coin flip to predict the weather then
We already do that in Denver.
> But if that is the case (and it seems to be), then a coin flip is better than their models.
If a coin flip is the necessary mental model to remind you both things can happen, then sure.
People just love horse race coverage. Silver gave us the most accurate horse race coverage. Maybe the lesson is stop following horse race coverage.
But most people went back to the tea leave readers. That way when the election was over, it can justifiably be the charlatan's fault that viewers got over-invested in their predictive capabilities.
I don’t disagree, I think the tea leaf reading is ultimately pretty futile.
But at the same time I do think it’s valid to say it’s more than a coin flip. The polling data over the election cycle showed that Trump had a smaller but still legitimate chance of winning. The data was different in 2020, when he lost.