On a mix of voting methods being used across states
I think this mixed approach is good. I wouldn’t worry about IRV not getting a foothold here though. My worry is IRV taking over the map. Also, there are other places that use IRV like Australia. To my knowledge, there’s been limited research on international and US IRV data. That could change once we have our director of research position.
Approval voting in presidential elections.
I’d point to the FAQ on presidential elections here. If you’re using approval voting, it should be part of the national popular vote. Otherwise, you get strange tactics coming up.
IRV variants & variants in general
One way to evaluate the performance of a voting method in electing a good winner is to use simulations. You have a computer take different conditions like candidate scenarios or different kinds of voters (you might call these dials) and then run these elections millions of times. Then you can see how much of the maximum utility was captured by the voting method.
We’re dealing with a model here, so there are assumptions that will vary from simulation to simulation. But in the models I’ve seen, approval voting fares well even with tactical voters, particularly against IRV and way better than the choose-one method. Also, there is a “magical best” in these simulations. That is, the “magical best” voting method magically picks the winner in each election that maximizes voter utility. “Magical best” is the unattainable ceiling of voting method performance in choosing a good winner.
That “magical best” mark is not terribly far away from approval voting’s mark. And you could get some small but likely meaningful increases by going from approval to score/range voting. Once you’re at score voting though, there’s not a lot to be gained.
And it’s because there’s little utility to be gained beyond score voting (and not much even beyond approval) that I don’t get particularly excited about the fanciest idea of a new voting method. And believe me, I hear all kinds of those ideas. I’ve actually gotten phone calls from random people on the weekend concerning this before.
The point is, we have some voting methods like approval that do really well in electing high-utility winners that are so easy. And not only is it easy but it has perks like giving an accurate reflection of support for losing candidates. And it is precinct summable and easily auditable. These factors are important. It’s not just about maximizing utility from the winner. A voting method has other jobs, too. (See the FAQ on how to evaluate a voting method.)
In terms of CES mission strategy, score voting is really the only other single-winner voting method that makes sense for us to try because it has so much simplicity going for it as well. It just has some small implementation hurdles and slightly more complexity that approval voting doesn’t have. But it’ll likely be a little bit before we consider anything with score voting, and it’ll have to be a strategic target. One step at a time, as they say.
On a mix of voting methods being used across states
I think this mixed approach is good. I wouldn’t worry about IRV not getting a foothold here though. My worry is IRV taking over the map. Also, there are other places that use IRV like Australia. To my knowledge, there’s been limited research on international and US IRV data. That could change once we have our director of research position.
Approval voting in presidential elections.
I’d point to the FAQ on presidential elections here. If you’re using approval voting, it should be part of the national popular vote. Otherwise, you get strange tactics coming up.
IRV variants & variants in general
One way to evaluate the performance of a voting method in electing a good winner is to use simulations. You have a computer take different conditions like candidate scenarios or different kinds of voters (you might call these dials) and then run these elections millions of times. Then you can see how much of the maximum utility was captured by the voting method.
We’re dealing with a model here, so there are assumptions that will vary from simulation to simulation. But in the models I’ve seen, approval voting fares well even with tactical voters, particularly against IRV and way better than the choose-one method. Also, there is a “magical best” in these simulations. That is, the “magical best” voting method magically picks the winner in each election that maximizes voter utility. “Magical best” is the unattainable ceiling of voting method performance in choosing a good winner.
That “magical best” mark is not terribly far away from approval voting’s mark. And you could get some small but likely meaningful increases by going from approval to score/range voting. Once you’re at score voting though, there’s not a lot to be gained.
And it’s because there’s little utility to be gained beyond score voting (and not much even beyond approval) that I don’t get particularly excited about the fanciest idea of a new voting method. And believe me, I hear all kinds of those ideas. I’ve actually gotten phone calls from random people on the weekend concerning this before.
The point is, we have some voting methods like approval that do really well in electing high-utility winners that are so easy. And not only is it easy but it has perks like giving an accurate reflection of support for losing candidates. And it is precinct summable and easily auditable. These factors are important. It’s not just about maximizing utility from the winner. A voting method has other jobs, too. (See the FAQ on how to evaluate a voting method.)
In terms of CES mission strategy, score voting is really the only other single-winner voting method that makes sense for us to try because it has so much simplicity going for it as well. It just has some small implementation hurdles and slightly more complexity that approval voting doesn’t have. But it’ll likely be a little bit before we consider anything with score voting, and it’ll have to be a strategic target. One step at a time, as they say.
(Bayesian Regret example for reference)