I have just joined a reading group related to prediction markets and crowd wisdom. One topic that has come up is range voting. Our current election system in the US is known as plurality voting: the person with the most votes wins. This is fine in elections with two candidates, but when there are more candidates it is very bad. More on that in a second. Range voting takes a different approach and lets you assign a value on a scale to each candidate. Let’s suppose the scale is 1 to 10, with 10 being your favorite choice. The person with the most points at the end of the day wins. This is much, much better. So much better.
Bayesian regret is very simply the difference between the best case scenario and what actually happened measured in terms of utility. The obvious question is “what does utility mean in a voting system?” It’s not exactly something that’s easy to quantify. Will I gain financially if Obama is elected? Will I gain satisfaction if Nader is elected? Will I lose my mind if McCain is elected?
Warren Smith has this solution: generate a multitude of random simulations and see which method comes out on top. It’s impossible to generate every possible permutation of voters in an election of any significant size, so he uses Monte Carlo methods to randomly sample the space. This has the added advantage of alleviating the need to quantify utility for real people. Just assign them a random value according to some model of voter mentality. Smith points out that’s not perfect since these simplified models cannot exactly match real human behavior. But if you run the simulation enough, it all comes out in the wash.
The result is that range voting beats everything every time in his simulations. Even when considering strategic voters, range voting performs better against other voting styles with strategic voters. Strategic voters are those who choose the lesser of the evils that they think might win. Plurality voting is actually one of the worst methods when it comes to elections with more than two candidates. So our voting system reinforces the two-party system we have in America. Go figure. The high numbers our method gets on Bayesian regret plays out in everyday life. There are all these rallies to get people out to vote. Most won’t, since they are disillusioned with the whole process. Why shouldn’t they be? It blows. Read the paper or check out RangeVoting.org for details.
Instead of rocking the vote, let’s change it altogether. Less Bayesian regret. I should sell T-shirts.