Quote:
Originally Posted by Nepeeme2008
FWIW, I believe the polls in general, but I don't believe the poll that just came out that puts Biden up by 17 in Michigan. No way that's accurate either
You mean Wisconsin? To give you an idea of the absurdity, that poll isn't saying Dems will gain Wisconsin. It is saying that Dems will
hold Wisconsin
Explanation:
Unusually, they didn't weight for 2016 vote but did ask about it. They found that 4% fewer people than reality said they voted for Trump, and 4% more than reality said they voted for Biden. Based on this poll, Clinton won Wisconsin 50-43 - the real result is she lost it 46-47.
Either:
1) This is sampling error - you should correct (not exactly but as a first order approximation) 4 percent up for Biden, 4 percent down for Trump. Adjusted result Biden +9
2) This is a "shy Trump" factor, people not admitting they are Trump supporters. Again the best approach is to adjust +4/-4 as in 1) Biden +9
3) This is former Trump supporters in denial that they were ever Trump supporters. Polling result stands Biden +17.
Obviously I tend to be in camp 1)/2) so I ignore polling that doesn't weight by previous vote. Reality may well include a bit of camp 3) though too so my approach may underestimate big swings.
I'm more confident in my approach in an election like this where the previous election was incredibly finely balanced so to predict the result this time around we just need to know the direction of the swing relative to last time - we don't need to know the magnitude of the swing. The UK 2019 election was similar (following on from the knife edge 2017 election) and the same approach worked pretty well in that.