Good job, Nate
-
It makes some sense that sweeps would be a local maximum in the probabilities. As the popularity bar slides in a direction, you pass over the last swing state at some point, then the popularity gap is very wide till the next state, not a swing state, falls.
-
Our most confident prediction going forward is that election predictions are garbage in, garbage out and there is nothing anybody can do about the garbage input, because enough people are gaming polls, or the self selection bias of those who don’t, is uncontrollable.
-
I think the bashing of the pollsters is a little unfair, no?
Due to the winner takes it all system, very few votes can make a huge difference. This makes it almost impossible to predict outcomes when the popular vote in a state is close.
@Klaus said in Good job, Nate:
I think the bashing of the pollsters is a little unfair, no?
Due to the winner takes it all system, very few votes can make a huge difference. This makes it almost impossible to predict outcomes when the popular vote in a state is close.
One of the most respected pollsters in the nation, Ann Selzer, had Iowa going to Harris in a shocking upset, which she released just before the election. Her prediction was Harris by 3, and it turned out Trump by 13.
-
Ok they can of course still be very wrong, but they should be measured by how well they predict votes, not electoral college seats. The latter is basically impossible because it's not a continuous process.
You can only make predictions and simulations for situations where small changes to the variables lead to small changes in the outcome.
-
Seltzer’s state of the art method was off by 16, from -3 to +13. She was predicting the vote in a particular state. That error would be impossible if she was capable of polling a sufficiently large, random sample of voters. She got the sufficiently large part, at least, but she was incapable of getting the random sample part. She has been in the polling business for like 30 years and her methods are state of the art. Accurate polling is not a solved problem and IMO does not look to be solved any time soon.
-
I may have mentioned that I find the incessant over-analysis of polls which are clearly inaccurate in predicting the actual outcome to be enormously irritating, but not quite as irritating as the long-winded and inevitable debate about why the pollsters were inaccurate.
I also find people phoning me up to ask how I'm going to vote to be enormously irritating.
Who the fuck goes to college thinking "One day, I'm going to be an opinion pollster"?
