Besides, if you’re going to apply game theory to the situation in a shortsighted local fashion—not taking into account others thinking similarly, and not taking into account the incentives you create for later elections based on what potential future candidates see you doing today—if, I say, you think in such a strictly local fashion and call it “rational”, then why vote at all, when your single vote is exceedingly unlikely to determine the winner?
Consider these two clever-sounding game-theoretical arguments side by side:
You should vote for the less evil of the top mainstream candidates, because your vote is unlikely to make a critical difference if you vote for a candidate that most people don’t vote for.
You should stay home, because your vote is unlikely to make a critical difference.
It’s hard to see who should accept argument #1 but refuse to accept argument #2.
It seems like the correct decision procedure would be:
Figure out the subjective utility you’d assign to each candidate’s being elected.
Figure out the probability that, for each candidate, your vote will singlehandedly make the difference between that candidate winning and losing. These probabilities will be tiny, of course, but I do think it’s the correct basis for deciding; I say this because if you had good reason to believe that your vote would literally not be used in computing the outcome (if on election day, the news channels were all saying “The government has announced that all ballot boxes will be replaced with shredders...”, or if you lived in a dictatorship where you know the elections were rigged, or if you, personally, were told that your vote would not be counted but everyone else’s would, etc.), then as far as I can tell, you would be perfectly justified in not voting. So it does seem that the minuscule probability that you will personally influence the outcome is the right basis for deciding to vote at all. That minuscule probability will at least vary between candidates, so you can compare between them.
Figure out the expected utility of voting for each candidate according to the two previous factors as well as the value of “sending a message” for good but unpopular candidates.
That’s only when it comes time to actually vote, of course. Until then, do everything you can to push those outcome utilities and probabilities in the right direction (respectively: push all the candidates in the direction of adopting better positions, however you can do that; and do things that will increase the probability of the best candidates winning, again according to expected utility per dollar, per time unit, etc.). But when it comes down to election day, does “stop voting for nincompoops” override “shut up and multiply”?
Edit: This might make an interesting web app, in the style of The Uncertain Future. You’d put in all these values and it would tell you the expected utility of voting for each candidate. It would have some subsections… one that would pull probability data from prediction markets and estimate the probability of your vote making a difference; one that would compute each outcome utility by taking a survey of the policies and personal qualities you favour, your credence that each candidate is being truthful about their stated policies and would be able to implement them, etc. In previous elections, there have been sites where you put in your preferred policies and it tells you which candidates are closest to you, but this would take it quite a bit further and would probably yield some very interesting results. I might do this come 2012.
It seems like the correct decision procedure would be:
Figure out the subjective utility you’d assign to each candidate’s being elected.
Figure out the probability that, for each candidate, your vote will singlehandedly make the difference between that candidate winning and losing. These probabilities will be tiny, of course, but I do think it’s the correct basis for deciding; I say this because if you had good reason to believe that your vote would literally not be used in computing the outcome (if on election day, the news channels were all saying “The government has announced that all ballot boxes will be replaced with shredders...”, or if you lived in a dictatorship where you know the elections were rigged, or if you, personally, were told that your vote would not be counted but everyone else’s would, etc.), then as far as I can tell, you would be perfectly justified in not voting. So it does seem that the minuscule probability that you will personally influence the outcome is the right basis for deciding to vote at all. That minuscule probability will at least vary between candidates, so you can compare between them.
Figure out the expected utility of voting for each candidate according to the two previous factors as well as the value of “sending a message” for good but unpopular candidates.
That’s only when it comes time to actually vote, of course. Until then, do everything you can to push those outcome utilities and probabilities in the right direction (respectively: push all the candidates in the direction of adopting better positions, however you can do that; and do things that will increase the probability of the best candidates winning, again according to expected utility per dollar, per time unit, etc.). But when it comes down to election day, does “stop voting for nincompoops” override “shut up and multiply”?
Edit: This might make an interesting web app, in the style of The Uncertain Future. You’d put in all these values and it would tell you the expected utility of voting for each candidate. It would have some subsections… one that would pull probability data from prediction markets and estimate the probability of your vote making a difference; one that would compute each outcome utility by taking a survey of the policies and personal qualities you favour, your credence that each candidate is being truthful about their stated policies and would be able to implement them, etc. In previous elections, there have been sites where you put in your preferred policies and it tells you which candidates are closest to you, but this would take it quite a bit further and would probably yield some very interesting results. I might do this come 2012.