2 ms·

Voter Satisfaction Efficiency (http://electionscience.github.io/vse-sim/VSE/ http://electionscience.github.io/vse-sim/VSE/) is another way of looking at differe

by jefftk 7y ago

Voter Satisfaction Efficiency (http://electionscience.github.io/vse-sim/VSE/ http://electionscience.github.io/vse-sim/VSE/) is another way of looking at different options, and IRV compares very poorly by it

joveian 7y ago
That is very interesting and I hadn't seen it before. IRV still does better than plurality by quite a bit, although it doesn't do well compared to the rest. As the chart shows, one of the biggest outliers is 100% honest voting with plurality. I hadn't heard of 3-2-1 voting before but that is interesting. Maybe it would be one of the best. I generally prefer ranking to scoring (I'm not sure how much better someone prefers a candidate should really be part of the process) but the simpler tallying and presentation of results is nice. I also like that in the "scenereo type" breakdown the easy case gets the best results, condorcet cycles get the worst results, and the rest are in the middle (all with little effect of strategy). It seems like the right distribution to me, but none of the other methods measured get that distribution. IMO a ranked preference is usually much easier to determine how to vote (although it can still be an issue when some candidates are much closer than others), although I guess the limited effect of strategic voting should hopefully mean that it usualy doesn't make much difference exactly how you vote. OTOH, at least the names of the options should be changed; it is the rare race where I could describe any of the candidates as anything other than "bad". https://electowiki.org/wiki/3-2-1_voting https://electowiki.org/wiki/3-2-1_voting IMO, proportional representation would be a much better change of voting system rather than just changing the method. Voting for one person is almost always going to leave lots of people unrepresented.
thaumasiotes 7y ago
I hate the way they've defined their metric: > VSE is expressed as a percentage. A voting method which could read voters minds and always pick the candidate that would lead to the highest average happiness would have a VSE of 100%. A method which picked a candidate completely at random would have a VSE of 0%. In theory, VSEs down to negative 100% would be possible if a voting method did worse than a random pick If VSE is going to be a percentage, it should be a percentage of the maximum happiness -- that is, the total happiness if everyone got their first choice, not the total happiness that is the highest any candidate can actually achieve. I say this mostly because it's much easier to calculate. But more fundamentally, there is no reason to expect the worst-case candidate to be equally as bad as the best-case candidate is good. It's a huge mistake to rank systems on a scale from +100% to -100% where the units on the positive and negative "halves" of the scale are different! And it's also a mistake to define 0% as whatever sortition achieves. It's trivial to construct a pathological system in which 0% VSE is equal to 100% VSE. (Everyone is equally un/happy with all candidates.) This should be impossible.
- thaumasiotes 7y ago
  Let's consider an example: You have 100 candidates. For 99 of those candidates, the entire electorate agrees that any one of these interchangeable guys would do a fine job. Nobody prefers any of them to any other. The last candidate is Mao Zedong. Under his leadership, the country's industrial and agricultural bases will be systematically destroyed, anyone who owns a rental property will be executed, and tens of millions of people will starve to death for no reason. The electorate agrees that Mao is an undesirable choice. So: any method that can't pick Mao achieves a VSE of 100%. Sortition, by definition, achieves a VSE of 0%. But it's still pretty good -- there's a 99% chance of selecting one of the good guys, and a 1% chance of epic, country-shattering disaster. A method that is guaranteed to select Mao has a VSE of, I assume, -100%. But it is much, much worse than sortition (the notional midpoint, 0%), while the VSE of 100% is only moderately better than sortition. These numbers don't make sense.

by jefftk 7y ago

joveian 7y ago

That is very interesting and I hadn't seen it before. IRV still does better than plurality by quite a bit, although it doesn't do well compared to the rest. As the chart shows, one of the biggest outliers is 100% honest voting with plurality. I hadn't heard of 3-2-1 voting before but that is interesting. Maybe it would be one of the best. I generally prefer ranking to scoring (I'm not sure how much better someone prefers a candidate should really be part of the process) but the simpler tallying and presentation of results is nice. I also like that in the "scenereo type" breakdown the easy case gets the best results, condorcet cycles get the worst results, and the rest are in the middle (all with little effect of strategy). It seems like the right distribution to me, but none of the other methods measured get that distribution. IMO a ranked preference is usually much easier to determine how to vote (although it can still be an issue when some candidates are much closer than others), although I guess the limited effect of strategic voting should hopefully mean that it usualy doesn't make much difference exactly how you vote. OTOH, at least the names of the options should be changed; it is the rare race where I could describe any of the candidates as anything other than "bad". https://electowiki.org/wiki/3-2-1_voting https://electowiki.org/wiki/3-2-1_voting IMO, proportional representation would be a much better change of voting system rather than just changing the method. Voting for one person is almost always going to leave lots of people unrepresented.

thaumasiotes 7y ago

I hate the way they've defined their metric: > VSE is expressed as a percentage. A voting method which could read voters minds and always pick the candidate that would lead to the highest average happiness would have a VSE of 100%. A method which picked a candidate completely at random would have a VSE of 0%. In theory, VSEs down to negative 100% would be possible if a voting method did worse than a random pick If VSE is going to be a percentage, it should be a percentage of the maximum happiness -- that is, the total happiness if everyone got their first choice, not the total happiness that is the highest any candidate can actually achieve. I say this mostly because it's much easier to calculate. But more fundamentally, there is no reason to expect the worst-case candidate to be equally as bad as the best-case candidate is good. It's a huge mistake to rank systems on a scale from +100% to -100% where the units on the positive and negative "halves" of the scale are different! And it's also a mistake to define 0% as whatever sortition achieves. It's trivial to construct a pathological system in which 0% VSE is equal to 100% VSE. (Everyone is equally un/happy with all candidates.) This should be impossible.

thaumasiotes 7y ago
Let's consider an example: You have 100 candidates. For 99 of those candidates, the entire electorate agrees that any one of these interchangeable guys would do a fine job. Nobody prefers any of them to any other. The last candidate is Mao Zedong. Under his leadership, the country's industrial and agricultural bases will be systematically destroyed, anyone who owns a rental property will be executed, and tens of millions of people will starve to death for no reason. The electorate agrees that Mao is an undesirable choice. So: any method that can't pick Mao achieves a VSE of 100%. Sortition, by definition, achieves a VSE of 0%. But it's still pretty good -- there's a 99% chance of selecting one of the good guys, and a 1% chance of epic, country-shattering disaster. A method that is guaranteed to select Mao has a VSE of, I assume, -100%. But it is much, much worse than sortition (the notional midpoint, 0%), while the VSE of 100% is only moderately better than sortition. These numbers don't make sense.