I was going to make a blog or video about this, but I've decided to just make it a twitter thread (b/c lazy). This map is my best guess as to what is going to happen. The reasoning behind it is in the rest of this thread. 
Generally polling for the presidency is pretty accurate. 2016 saw better than average national polls and worse than average state polls and those state errors clustered in swing states causing a large upset. The fact that the national polls were better suggests that...
...a significant proportion of the error in state polls was not due to "Shy trump voters", as that would distort national polls as well. Many pollsters plausibly argue that they under-estimated the number of white low-education voters last time and have adjusted for that...
...in their current weighting system. This gives us good reason to think that the polls will be somewhat more accurate this time compared to 2016. There is also a larger enthusiasm gap favoring Biden than there was Clinton (who, remember, won the popular vote). This gives us...
...some reason to think that there will be more surprise Biden voters than there were surprise Clinton voters. For Trump to win this time, you'd need to increase the state level errors by 40% relative to last time, or shift every poll by six points. Thus, to think that...
...Trump will win we need justification for the view that the polls are worse than last time. As I've said, I think we have justification for the opposite view.
Polls suggest Biden is winning among those who already voted by 39 points while trump is winning among...
Polls suggest Biden is winning among those who already voted by 39 points while trump is winning among...
...those who will vote on election day by 37 points. If, as is predicted, roughly 150 million people vote then early voters will account for roughly 2/3 of voters meaning trump has to win election day voters by twice the margin that Biden won early voters to break even. The...
...above polling suggests this will not happen. In terms of actual registration of voters who voted early, in the swing states this election looks like an outlier favoring democrats. Many people are trying to predict election outcomes w/target smarts predicted early vote count...
...TS tells people not to do this for good reason, in 2016 there was hardly any relation between how republican they thought the early voting of swing states was with how republican the final vote was. Their modeled data should not be used to make forecasts. That being said...
...in FL the registration based early voting looks better for Trump now than in 2016, and in general we'd expect it to look worse even given the same outcome become more dem votes will have shifted to early votes due to covid fears, so I think Trump will win FL. In GA,
...polling aggregates are being distorted by historically inaccurate polls, and most of the good and very recent polls show Trump winning, so I think he will win GA. Registration based early voting in NC is way better this time than in 2016, so I think Trump will win NC....
...all of these races are close according to the polls, and so while my predictions contradict what 538 type forecasters are predicting they don't require abnormally large polling errors to occur. Other than FL, GA, and NC, my predictions are inline with what more forecasts...
...say and are just based on the assumption that a huge polling error wont occur. I think this is what the data says will most likely happen, but there is a ton of uncertainty in this prediction, and if the data is going to be massively contradicted by reality then, well, I...
...obviously wont be able to see that by looking at the data. But I hope that is what happens. We'll see shortly, but I hope I am wrong and Trump wins re-election.
Also: some people think stuff like incumbent advantage or people saying their lives are going well favors trump, but I'm not convinced those things are strong predictors beyond what is already predicting by direct polling and early voting.
This map contains at least two errors (b/c tired): Iowa should go to Trump and NV should go to Biden. They are each worth 6 votes so this doesn't change the outcome.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
