Before just telling all pollsters to get another job and abandoning/ignoring everything, shouldn't we find out why the polls were so far off...again?
Were Trump voters not willing to admit they were going to vote for Trump? Was it something else? Did they just assume that college-educated people would prefer Biden 10-1 when it was instead 6-4 or something?
If you were someone who was going to vote for Trump and for the last 4 years you were continually portrayed in the media as a deplorable racist nazi would you be super excited to share this with a random stranger?
No, they are different people. "The soccer mom doesn’t want to say she’s for Trump because she doesn’t want you to think she’s one of them." Just like "there are a lot more people who like professional wrestling than admit it. There are lots of fans who don’t want you to think they’re like the other people who like professional wrestling."
Could also be that the models used were never verified for such a record turnout.
If, of people saying they’ll vote A, normally 100% will actually vote, and of those saying they’ll vote B, normally 80% will actually vote, the models that predict election results from poll result will correct for that.
So, if the poll says it’s 50/50, the model will predict a clear win for A, with about 55% of the vote.
If turnout increases, all extra voters will vote B, but the model might still predict that 55% (I kept the example simpler than reality will be)
OK, honest question: How many people here actually answer polls?
I got more poll calls than I had any interest in answering. I especially had no interest in answering a push poll, from either side. The result was that I answered none of them.
Now, if we had only one outfit doing polls so that I didn't get bombarded, and if there was no such thing as push-polling, sure, I might answer their poll questions. But as it is? No. You'll find out what I think when I vote; in the meantime; go away and leave me alone. (It's amazing how much quieter my phones are today than on Monday.)
They have been as accurate as can be. Most suggested a tight race.
By leaving the door ajar for surprises, 538, for example, used their smart algorithms to tweak their projection. I have not been disappointed by their rhetoric either.
If you scroll down to the path to victory, this is how I read it: The cusp of the election lies in whether or not the following states flip - Georgia, North Carolina, Florida, Arizona, Pennsylvania, Nevada. Sure looks like they hit the nail on the head with regard to WHERE the uncertainty was. Maybe that was obvious, maybe every outlet had that same list. But before the election started, we had 6 states to pay extra attention to, and as of right now only one of those is called for certain (Florida) and one is being debated (Arizona.) For what I looked to the forecast to learn, it delivered.
Looks like they're going to be pretty accurate with presidential electoral college and popular vote overall, but off on the margins within each state and the senate and house.
Not a disaster for forecasts though especially in light of COVID which made people vote and parties campaign very differently from previous elections
16 comments
[ 0.25 ms ] story [ 48.1 ms ] threadWere Trump voters not willing to admit they were going to vote for Trump? Was it something else? Did they just assume that college-educated people would prefer Biden 10-1 when it was instead 6-4 or something?
https://news.ycombinator.com/item?id=25011681
[0] https://www.politico.com/news/magazine/2020/10/29/2020-polls...
If, of people saying they’ll vote A, normally 100% will actually vote, and of those saying they’ll vote B, normally 80% will actually vote, the models that predict election results from poll result will correct for that.
So, if the poll says it’s 50/50, the model will predict a clear win for A, with about 55% of the vote.
If turnout increases, all extra voters will vote B, but the model might still predict that 55% (I kept the example simpler than reality will be)
I got more poll calls than I had any interest in answering. I especially had no interest in answering a push poll, from either side. The result was that I answered none of them.
Now, if we had only one outfit doing polls so that I didn't get bombarded, and if there was no such thing as push-polling, sure, I might answer their poll questions. But as it is? No. You'll find out what I think when I vote; in the meantime; go away and leave me alone. (It's amazing how much quieter my phones are today than on Monday.)
By leaving the door ajar for surprises, 538, for example, used their smart algorithms to tweak their projection. I have not been disappointed by their rhetoric either.
If you scroll down to the path to victory, this is how I read it: The cusp of the election lies in whether or not the following states flip - Georgia, North Carolina, Florida, Arizona, Pennsylvania, Nevada. Sure looks like they hit the nail on the head with regard to WHERE the uncertainty was. Maybe that was obvious, maybe every outlet had that same list. But before the election started, we had 6 states to pay extra attention to, and as of right now only one of those is called for certain (Florida) and one is being debated (Arizona.) For what I looked to the forecast to learn, it delivered.
Not a disaster for forecasts though especially in light of COVID which made people vote and parties campaign very differently from previous elections