> However, Republican Donald Trump pulled off one of the greatest upsets in history—rendering the polls useless.
This is dumb. Going into election day, 538 gave Trump a roughly one-in-three chance of winning, based on polls. (Anyone who's played X-COM will know better than most what a 70/30% split actually means.) The end result in each state wound up very close to what the polling data indicated; it was only a small number of votes in a small number of swing states that had to vary. With a margin of error of 2-3%, a close race always has the potential for surprises - especially if the reasons for a mild shift in one direction are correlated between states.
This is a great breakdown of prediction markets and how it would be far more accurate than a poll due to loss aversion which cancels out bias. This is classic “Put your money where your mouth is”.
Edit: In other words, if you believe in modern portfolio theory and market efficiency, aka boglehead index fund beats all; then you must conclude that if this was actually an efficient betting market than the outcome would be near 100% certainty.
This article seems pretty light on evidence. Where is the data comparing election outcomes to both polls and betting odds over time?
The argument seems to be based mainly on this:
> Political betting sites won’t give one presidential candidate great odds of winning just to influence voter perception. They’d stand to lose serious money in such an event.
Of course, political pollsters also stand to lose money if their polls are wrong, because if they have a bad enough track record they will no longer have customers. When the polls are wrong it really pisses off the pollsters' two main customers: political campaigns who want to know how popular their candidate really is, and the news media which wants their election predictions to be correct.
"Various news outlets conduct polling on presidential elections. The public often relies on such polls to stay informed on each candidate’s chances of winning."
Perhaps what they tell the public is different than internal analysis, but national popularity polls are completely irrelevant in a national POTUS election.
We have the electoral college. It's about states, and most would say specific counties.
5 comments
[ 2.9 ms ] story [ 30.1 ms ] threadThis is dumb. Going into election day, 538 gave Trump a roughly one-in-three chance of winning, based on polls. (Anyone who's played X-COM will know better than most what a 70/30% split actually means.) The end result in each state wound up very close to what the polling data indicated; it was only a small number of votes in a small number of swing states that had to vary. With a margin of error of 2-3%, a close race always has the potential for surprises - especially if the reasons for a mild shift in one direction are correlated between states.
https://fivethirtyeight.com/features/the-polls-are-all-right...
https://fivethirtyeight.com/features/election-update-why-our...
Edit: In other words, if you believe in modern portfolio theory and market efficiency, aka boglehead index fund beats all; then you must conclude that if this was actually an efficient betting market than the outcome would be near 100% certainty.
The argument seems to be based mainly on this:
> Political betting sites won’t give one presidential candidate great odds of winning just to influence voter perception. They’d stand to lose serious money in such an event.
Of course, political pollsters also stand to lose money if their polls are wrong, because if they have a bad enough track record they will no longer have customers. When the polls are wrong it really pisses off the pollsters' two main customers: political campaigns who want to know how popular their candidate really is, and the news media which wants their election predictions to be correct.
Perhaps what they tell the public is different than internal analysis, but national popularity polls are completely irrelevant in a national POTUS election.
We have the electoral college. It's about states, and most would say specific counties.