Comment by godelski | Hacker News Reader

Comment by godelski | Hacker News Reader

Parent Full thread

For a more interesting experiment Fargo had an election with Approval Voting[0].

Unfortunately RCV (IRV) doesn't solve many of the problems that it claims to do. One of the claims is that IRV prevents spoilers, which is objectively false[1][2]. We're pretty interested in the Favorite Betrayal Criterion to prevent this[3]. We also see that IRV has a fairly poor voter satisfaction efficiency (VSE)[4][5].

It really feels like people just watched the CGP Grey videos on election science and dug no further. Even Arrow himself said that cardinal voting (Approval, score, STAR) is probably the likely answer. They also tend to be much simpler as they only are single rounds and resistant to people accidentally "ranking" two candidates equally (which you might actually favor two candidates equally and these methods let you express that).

[0] https://electionscience.org/commentary-analysis/fargos-first...

[1] https://www.youtube.com/watch?v=JtKAScORevQ

[2] https://electionscience.org/library/the-spoiler-effect/

[3] https://electowiki.org/wiki/Favorite_betrayal_criterion

[4] https://electionscience.github.io/vse-sim/VSE/

[5] https://www.youtube.com/watch?v=-4FXLQoLDBA

I knew about Fargo using approval voting (and I think that's awesome), but it surprised me that they used it in a multi-winner election. Using approval voting for multi-winner can work, but it'll tend to produce winners who are ideological clones of each other, rather than something more like proportional representation.

To give an example of what I mean, suppose 51% of the city is solidly for party A, and 49% is solidly for party B. If you run an election with two candidates of party A and two candidates of party B, and everyone votes according to their partisan preferences, then both the party A candidates win rather than, say, one A and one B.

There are some proposed methods that extend approval voting to do something closer to proportional representation by weighting the votes of voters who have gotten one of their candidates elected less than ones who haven't, but I don't remember all the specific details.

That said, I think this makes sense if having multiple winners is unusual (like there were two seats up this election rather than one because someone left office before their term was up), or if there's a bunch of commissioners who are replaced in stages like U.S. Senators, and it would be weird to make results proportional within each "class" but not proportional overall.

Anyone know the full story here?

> it'll tend to produce winners who are ideological clones of each other, rather than something more like proportional representation.

I suspect this will be beneficial. We'll see. They could eventually move to a proportional form of approval voting.

> Even Arrow himself said that cardinal voting (Approval, score, STAR) is probably the likely answer.

Cardinal systems are mathematically better by some criteria but have the extreme problem, that those criteria overlook, that their is no clear mapping from real preferences to honest ballot markings (and, similarly, no clear interpretation of what any given set of ballot markings mathematical analysis which assume that there is some real underlying thing which maps neatly to ballot markings on cardinal ballots may find that cardinal voting systems are good at aggregating that mythical thing, but we know from experience with rating systems that markings in cardinal systems aren't consistent for similar preferences but instead vary, for similar relative preferences and absolute opinions on the rated subjects, by culture and subculture. And you can't really argue that that would be mitigates in practice by people learning voting techniques that are efficient in achieving goals under such systems, because that is too complex of a function of other voter’s expected markings to be tractable. (There are situations where the markings on cardinal ballots can be concretized in terms of prices people are will to pay for or to avoid certain outcomes, and where those become concrete binding mutual commitments among voters, those systems have clear utility. But in general public elections they aren't.

Where you have multiple winners naturally (which includes elections of Presidential electors, which are always at least 3 per state), proportional systems (whether party-list or candidate-centered like STV) are the natural solution, as multiwinner elections are fundamentally easier than single-winner.

> Even Arrow himself said that cardinal voting (Approval, score, STAR) is probably the likely answer.

Unfortunately, as I understand it, it turns out the problem is basically unsolvable: https://politics.stackexchange.com/a/14245

Voting itself, yes. Which is why Arrow (who is being referenced in that post) is hedging. But just because there is no global maxima doesn't mean there aren't better solutions than others. It depends on the criteria that you're looking for. While Cardinal systems don't maximize VSE like Schulze or RP do (which also don't get 100%!) they much better counter strategic voting as well as are dead simple. If you read a few of the links I posted or watch a few of the videos they explain some of these concepts.

But yes, voting math (social choice theory) is hard. No one is going to combat you on that.

Cardinal systems generally do BETTER than Schulze or ranked pairs. Especially STAR voting.

http://electionscience.github.io/vse-sim/vse.html

Hey, it's that guy from the video I sourced!

And yes, but people tend to pay attention to only the maximal value. Just trying not to complicate things.

Reply on news.ycombinator.com