This is a very cool new interactive from
@FiveThirtyEight
. It lets you see how the odds change depending on who wins specific states.
So if Trump wins PA, he has a 73% probability of winning the election, but just 3% if he loses it.
@FiveThirtyEight
What makes this really powerful is that it incorporates correlated errors. Biden is ahead in the polls in PA, so if Trump wins it, that means the polls were off (or that something else changed). That shifts probabilities elsewhere as well, esp. nearby states like Ohio.
@bencasselman
@FiveThirtyEight
If I understand this correctly, the outcomes are correlated. So choosing Trump to win PA increases his odds of winning other states. It’s the correct model IMO but changes the interpretation.
@danielmangrum
@FiveThirtyEight
Yes, that's right. The polls favor Biden in PA, so if Trump wins it, that indicates the polls were off there (or something changed in the real world), so probabilities in other states shift as well. (With some states correlated more strongly with PA than others.)
@bencasselman
@FiveThirtyEight
You've not understood the model - by giving Trump or Biden PA the other states' results are affected, as you can see on the screenshots
@AScribbledEagle
@FiveThirtyEight
Right. The model (correctly) recognizes that results are correlated. If Trump wins PA, that means the polls there were off (or something changed), so that means other states will probably swing in his direction as well.
@dumpchrump
@FiveThirtyEight
Well, but the example I showed was assuming you make no other selection, and it updates its probabilities based on how you assign PA.
@bencasselman
@FiveThirtyEight
Their biggest problem is still that common folks don't trust the polls to be right. Sure Nate got it 2 prez elections in a row, but basically he saying "our models say 9 out of the 10 invitees to your dinner party aren't axe murderer of democracy" but if it happens... He still $
"I wasn't coming to Erie ... I have to be honest, there was no way I was coming" -- Trump's pitch to Erie is basically telling people how much he doesn't want to be there. Bold strategy.
@bencasselman
@BrandenHarvey
@FiveThirtyEight
Surprising that tossup states have a large effect on model predictions for other states. E.g., calling Ohio (predicted 50/50) should change outcome distribution (locks in EVs), but absent data on margin of victory, outcome of Ohio doesn't provide much information abt other states