3 ms·
I want to push back against this, and specifically fair vote. - What they are referring to as RCV is called Instant Runoff Voting (IRV). IRV is a RCV, but RCV
by godelski 6y ago
I want to push back against this, and specifically fair vote.
- What they are referring to as RCV is called Instant Runoff Voting (IRV). IRV is a RCV, but RCV isn't IRV. Think "Rock and Roll is The Beatles" vs "The Beatles are a Rock and Roll group."
- IRV isn't that great when it comes to Voter Satisfaction Efficiency (VSE)[0]. There's even other RCV methods like Ranked Pair[1] that are exactly the same for the voter. Essentially IRV is a 6% improvement upon plurality (what we currently use) but RP is a 15% increase from plurality. So I ask, why go to IRV when we can do twice as good? (or better! Keep reading)
- Fair Vote claims that IRV has a high resistance to spoiler candidates (a candidate that takes away votes from another similar candidate). This is actually why the distinction from RCV and IRV is important and why Fair Vote is misleading. While RCV is better than plurality in this respect, some RCVs are better than others. IRV doesn't adequately solve this issue (but there are other RCVs that do!). IRV fails when two candidates are similar in positions and similar in popularity. [2] This is specifically the kind of spoiling that we are trying to prevent. We're trying to prevent a candidate like Sanders from spoiling Biden (or vise versa) not Jill Stein spoiling Biden. IRV prevents the latter, but not the former. I am specifically upset with Fair Vote because they are misleading in this context. Because they claim IRV := RCV, they take all the things different RCV methods solve and claim IRV solves them. This is simply not true (and why I make the first bullet).
- There are just better methods besides ranking (also called "ordinal" methods). There's a class called Cardinal[3], that is actually simpler. The simplest is called approval. An example of approval is Netflix's rating system (binary). It does pretty well. Better yet is range/score voting. This would be the old Netflix system where you rated out of 5 stars. The difference in ranked vs cardinal is that you rate each candidate or policy independent of the others. This allows you to be more precise/honest (you may actually like two candidates equally!). It also reduces complexity because a common accident in ranking is rating two candidates the same. In ordinal systems this is a bug, in cardinal this is a feature.
- I want to specifically mention STAR voting[4]. While STAR isn't perfect, it much better handles various types of spoiler candidates, has a high VSE (similar to that of RP), and handles strategic voting well. The last is the major reason you should choose STAR over RP (and there's no reason you should choose IRV over RP, because RP is strictly better). While all three systems encourage honest voting (in a single agent scenario) strategic voting is always an issue. Under no circumstance is STAR worse than IRV because STAR handles strategic voting. Specifically the most important kind is the "1-sided strategy"[5] which is arguably very common.
- Lastly, I want to appeal to authority. Kennith Arrow (who you may be familiar with from Arrow's Impossibility theorem and Gibbard's extension) is a Nobel Laureate and said cardinal systems are "probably right."[6] And I think many of us know science speak into the confidence that conveys. That link is a full interview with him from Election Science. It is a good read/listen.
TLDR: We want a good voting system, use STAR, not IRV.
[0] https://electionscience.github.io/vse-sim/vse.html https://electionscience.github.io/vse-sim/vse.html
[1] https://en.wikipedia.org/wiki/Ranked_pairs https://en.wikipedia.org/wiki/Ranked_pairs
[2] https://www.electionscience.org/library/the-spoiler-effect/ https://www.electionscience.org/library/the-spoiler-effect/
[3] https://en.wikipedia.org/wiki/Cardinal_voting https://en.wikipedia.org/wiki/Cardinal_voting
[4] https://en.wikipedia.org/wiki/STAR_voting https://en.wikipedia.org/wiki/STAR_voting
[5] https://electionscience.github.io/vse-sim/VSE/ https://electionscience.github.io/vse-sim/VSE/
[6] https://www.electionscience.org/commentary-analysis/voting-theory-podcast-2012-10-06-interview-with-nobel-laureate-dr-kenneth-arrow/ https://www.electionscience.org/commentary-analysis/voting-t...
- dragonwriter 6y agoMost of what you write is good, but I want to point out that non-ordinal methods suffer from a problem that is at a level more fundamental than most analyses address, to wit, that there is no objective question about preferences that they ask, and thus no clear “correct” honest ballot marking given a set of voter preferences. There are conditions which resolve this (e.g., when ballot markings in a method like score voting represent the amount of some valuable resource the voter commits to paying if the given choice is elected, or where markings in approval represent a binding commitment to not participate if an non-approved option is chosen and/or to participate if an approved options is chosen.) But these resolutions are typically not available or desirable in most public elections.
- godelski 6y agoThank you. I'd like to respond in kind > that there is no objective question about preferences that they ask I don't see this as an issue. I also don't see it as a problem to specifically non-ordinal, but to every system. I like to try to explain a STAR05 ballot simply to people with a forum they are very familiar with. A ballot might look like below How do you feel about ___x___ candidate? Very Unfavorably, unfavorably, neutral, favorably, very favorably Those are your 5 options. Yes, this isn't exact, but we can't have an infinite score. But as we understand distributions, we know that the fluctuations won't matter much. If a group of people have an "honest" opinion that a candidate is rated "favorably" (or a 4) we'll see a distribution around this (a fairly tight distribution). So you'll get some 3's and 5's, but most people will give 4's and the average will be 4. So it isn't an issue. Similarly, if you are ranking people and view two candidates equally 50% of people will put candidate A in front of B and vise versa (well real world there'd be a bias for things like: who is listed first, "nicer" sounding name, familiarity, etc. But still, that's not too big of an issue because at the end we get pretty close to the optimal efficiency that the system can provide). You have to remember that there is no perfect system, Arrow and Gibbard showed that. But the question is if we statistically come close to the optimal (that the system can provide). I want to point out that simple approval voting is VERY effective, even though it is low precision. You're literally choosing "like" and "dislike." Analysts aren't that concerned with the lack of objectivity in the difference between a 4 and 5 or a 3 and 4 because it is much more refined. If this is a major concern, then extend the expressiveness by increasing the range, e.g. 0-10. I would also encourage you to listen to that interview that I linked. Arrow briefly discusses this topic. He makes some criticisms of score voting but also says it is the option he would push (question is which he'd advocate for). Remember, there is no such thing as a perfect voting system. There's always a trade-off. We're just trying to minimize the trade-offs. And personally, if someone made RP popular I would gladly advocate for it. But because we're trying to trade one shitty system (plurality) for something only slightly less shitty (IRV) I am going to be vocal about the issues and lying that is happening. I don't really have an issue with RCVs, but specifically IRV. This is why I wanted to make a clear distinction between the two. Because when the discussions come up we end up using different languages, using the same words to describe different things. This may have been a smart move by Fair Vote, but we're on Hacker News and there's an expectation of more nuance than we'd have in other places like Reddit or in person.