Being in California, Gelman et al. put my probability of a decisive vote around 1/(5 million).
As the paper says:
[W]e consider how the results would change as better information is added so as to increase the accuracy of the forecasts. In most states this will have the effect of reducing the chance of an exact tie; that is, adding information will bring the probability that one vote will be decisive even closer to 0.
And as it turns out, conditional on polls and other information from right before the election, one would have to assign a very low probability that California will (almost) vote Republican. Also, conditional on California (almost) voting Republican, one would have to assign a very high probability that enough other states will vote Republican to make California’s outcome not matter.
It seems to me that a reasonable probability estimate here would be multiple orders of magnitude lower than the cited estimate; and it seems to me that together with the optimal philanthropy point made by user:theduffman and user:dankane and user:JohnMaxwellIV elsewhere in the thread, this makes voting in states like California not worthwhile based on the calculation presented in the original post.
Similarly, California’s Senate race isn’t significantly likely to shift. But at the House of Representatives level, the probabilities could be more significant—depending where you live. State representatives, mayors, plebiscites, etc.: there are many opportunities.
Irrespective of California, many people even in swing states think voting is silly, so I would hope that they read this post… but regarding California,
And as it turns out, conditional on polls and other information from right before the election, one would have to assign a very low probability that California will (almost) vote Republican. Also, conditional on California (almost) voting Republican, one would have to assign a very high probability that enough other states will vote Republican to make California’s outcome not matter.
Thumbs up, except that conclusion here is not to not vote… it’s to either
1) watch the polls and vote based on proximity to a tie at both the state and federal level,
or if the time watching the polls is more of a sacrifice to you than the time spent on late-stage voting (can’t vote by mail),
2) just vote without the poll information.
Reason: supposing that (a) without the poll information, the EV of voting is high, and (b) finding out the poll results can change your decision, (a+b) implies that the poll results have high VOI. More precisely,
Since Pr(decisive | not-close poll) is many orders of magnitude closer to 0 than 1/(5 million), and Pr(close poll) is quite small, say 1/N, Pr(decisive | close poll) must be on the order of N 1/(5 million), so the payoff would be N whatever is reported in the post, which would be huge.
So the conclusion here is that “Voting without poll results is like donating to charity, but adopting the policy of watching the polls and deciding to vote based on proximity to a tie is like donating almost as much to charity, and saves you time, unless you spend more time watching the polls that you would spend to vote.”
As the paper says:
And as it turns out, conditional on polls and other information from right before the election, one would have to assign a very low probability that California will (almost) vote Republican. Also, conditional on California (almost) voting Republican, one would have to assign a very high probability that enough other states will vote Republican to make California’s outcome not matter.
It seems to me that a reasonable probability estimate here would be multiple orders of magnitude lower than the cited estimate; and it seems to me that together with the optimal philanthropy point made by user:theduffman and user:dankane and user:JohnMaxwellIV elsewhere in the thread, this makes voting in states like California not worthwhile based on the calculation presented in the original post.
Similarly, California’s Senate race isn’t significantly likely to shift. But at the House of Representatives level, the probabilities could be more significant—depending where you live. State representatives, mayors, plebiscites, etc.: there are many opportunities.
Irrespective of California, many people even in swing states think voting is silly, so I would hope that they read this post… but regarding California,
Thumbs up, except that conclusion here is not to not vote… it’s to either
1) watch the polls and vote based on proximity to a tie at both the state and federal level,
or if the time watching the polls is more of a sacrifice to you than the time spent on late-stage voting (can’t vote by mail),
2) just vote without the poll information.
Reason: supposing that (a) without the poll information, the EV of voting is high, and (b) finding out the poll results can change your decision, (a+b) implies that the poll results have high VOI. More precisely,
1/ (5 million) = Pr(decisive | no info) = Pr(decisive | close poll) Pr(close poll) + Pr(decisive | not-close poll) Pr(not-close poll)
Since Pr(decisive | not-close poll) is many orders of magnitude closer to 0 than 1/(5 million), and Pr(close poll) is quite small, say 1/N, Pr(decisive | close poll) must be on the order of N 1/(5 million), so the payoff would be N whatever is reported in the post, which would be huge.
So the conclusion here is that “Voting without poll results is like donating to charity, but adopting the policy of watching the polls and deciding to vote based on proximity to a tie is like donating almost as much to charity, and saves you time, unless you spend more time watching the polls that you would spend to vote.”
The problem is that letting polls influence voting decisions is subject to Goodhart’s law.