[erdevs]

I see many comments here asserting that RCV is "fundamentally sound" and that there is proof that it satisfies at least strategyproofness (in the sense of Cooperative Game Theory and Social Choice Theory, s.t. revealing/voting for true preferences is at least weakly dominant).

But can anyone provide backup for these claims?

I'm far from an expert, but I'm curious to reconcile these claims vs Arrow's Theorem and Gibbard's Theorem. My (potentially flawed) understanding as well is that RCV leads to a greater propensity for "extreme" (in the strict sense of being the top choice of a small minority of voters) candidates to win.

[rrrrrrrrrrrryan]

FPTP makes the mistake of assuming that the person with the most votes is the best.

Likewise, the biggest flaw of RCV (to me) is that it makes the opposite mistake: it assumes the person with the least votes in each round is the worst. This isn't necessarily the case.

It's still much better than our current system because it collects more data-points from voters about their preferences, and uses that data to make more comparisons between different batches of candidates.

But there are simpler systems out there with even greater voter satisfaction efficiency, because they don't make the same flawed assumptions about voter preferences. In approval voting, you either approve of a candidate or you don't, and you vote as such.

Election science wonks like to talk about theorems and satisfaction of obscure criteria, but I think they often undervalue the two things that matter most: voter satisfaction and simplicity. "Simplicity" meaning: an optimal strategy that's simple-to-explain, an easy to understand outcome, low cost of implementation, etc.

Most of us here are software folks, and we understand that designing a great system is the easy part; the hard parts are implementation and change management. Voting systems should be thought of the same way.

[godelski]

I'm glad you brought this up, because ~RCV~IRV does not actually satisfy these conditions (it is a personal point of frustration how Fair Vote is misleading, even by using the term RCV). IRV doesn't pass the favorite betrayal criteria. IRV also causes a weaker definition of what it means to be a spoiler[1]. Specifically you should look at monotonicity[2].

I'd encourage you to dig a bit into electionscience.org (they have some good YouTube videos as well) as they do a deeper dive on some of these topics. There's a reason Arrow himself was a fan or cardinal systems. The other part is that we need to consider factors other than VSE and spoilers. A lot of people get hyper focused on VSE (typically Condorcet supporters) and while it is an important factor 0.5% isn't a big deal (really a few percent isn't). Also you might want to check out our (HN) resident voting expert's[3] comments, because he links to a lot more information. I'd call myself a hobbyist where Clay is an expert.

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

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

[2] https://electionscience.org/library/monotonicity/

[3] https://news.ycombinator.com/user?id=ClayShentrup

[nickff]

RCV is basically just a very quick runoff election; it has the same propensity for extremeness that a runoff does.[1] That said, I think it is very likely that more people will vote for third-party candidates (TPC) under ranked-choice.

You could make the case that TPC will allow for the representation of a wider spectrum of views, or alternatively that TPC are likely to be extremists. Both may be true.

[1] https://en.wikipedia.org/wiki/Ranked_voting#Ranked_Choice_Vo...

[kibwen]

I would be sensitive to the "TPC under RCV are more likely to be extremists" argument were it not for the fact that FPTP has already allowed profound extremism in one of our two parties, and since FPTP is so anti-competitive it means that it's extremely difficult to combat. I'll take a system that makes parties more competitive, which non-FPTP systems all do.

[nickff]

I'm not sure what your definition of 'extremist' is. If either major party is getting close to 50% of the votes, it must be quite mainstream. I don't have to like it, and I may in fact find it abhorrent, but that has nothing to do with extremism.

[snewman]

A common scenario in many locations in the US:

A district tilts heavily toward one of the two main parties -- it is a "safe" district for that party. The candidate nominated by that party in the primary election is nearly certain to win the general election. If the candidate is highly extremist / flawed, they might lose, but party affiliation is sufficiently strong nowadays (aka the electorate is sufficiently polarized) that a candidate can be pretty far to one extreme -- farther out than the bulk of the electorate -- and still win.

Meanwhile, there is a tendency for the more centrist voters in both parties to skip the primary. Thus, the candidate who survives the primary is often relatively extreme.

The result is that the victor of the general election is often to the extreme side of not only the electorate as a whole, but the membership of their party.

To oversimplify, imagine that political views fall on a one-dimensional spectrum ranging from 0 to 1, and the electorate consists of:

  - 40% at 0.4 (center-left)
  - 30% at 0.6 (center-right)
  - 30% at 0.8 (heavy right)

In the primary, center-right voters are under-represented, and a candidate at or beyond 0.8 has an excellent chance of being nominated. Then in the general election, at least 2/3 of the center-right voters are likely to swing toward that candidate (because of polarization / strong party affiliation).

[nickff]

Looking at your spectrum, I'd see the same distribution as:

- 40% left

- 30% center

- 30% right

Any rating of 'how far' someone is along the spectrum is arbitrary, as the spectrum itself is arbitrary. I wouldn't see any extremism either.

[snewman]

Agreed, the spectrum ultimately is arbitrary. I was thinking in terms of a spectrum normalized to the country as a whole. In a particular district, you often have a breakdown that is off-center relative to the overall nation (of course that can be in either direction).

And in any case, the key point is that the primary process can lead to a candidate taking office who is well off-center even within their district.

[kibwen]

Amusingly, upthread you and I are talking about whether encouraging more people to vote leads to better outcomes, and this is an example where the answer is yes: a candidate can be both an extremist and receive 50% of the vote if, as in is true in the US, less than half the eligible electorate actually votes. The current president was elected on the backs of 25% of eligible voters; does that make him an extremist candidate even by the standards of the US electorate? If we had more data on account of more Americans asserting their preferences, then we could actually determine that.

[nickff]

I think that's a sign that each of us has nuanced views!

'Extremist' is a bit of a slippery fish; I can't imagine calling something with even 10% support 'extremist'. If I had to draw a line, it would probably be that the sum of all 'extreme position support' would have to be less than 5% for a given subject. That is not to say that positions with >5% support are necessarily correct or conscionable.

[pessimizer]

For "extremism" to be anything but a meaningless slur, it has to at least include in its meaning "not common."

[1980phipsi]

Well it's similar to a traditional runoff when there are two major candidates, but it could lead to different results when support is split more evenly among three or more candidates.

[nickff]

RCV gives you the same results as a runoff if the voters have ordinal preferences that they would have exercised in a runoff. There are a number of reasons why either of these may be wrong, but it seems to be a fairly reasonable set of assumptions.

[RangerScience]

IMO it also lets you just count the interest, regardless of outcome. Now you could see Nader (or Bernie, or the libertarian candidate) get 5, 10% of 1st choices and know it's legit, even if it doesn't change the outcome of the election.

[jessaustin]

This is why it's opposed by both faces of the status quo party. Once a third party gets to 10% for a given office, there won't be much stopping it from getting to 40% in the next election for that office.

[al132]

I can't speak for any theorems so take this with a grain of salt, but if the candidate who was the top choice of a small minority manages to win that would mean that they are someone that a majority would settle for, which I would think is a more optimal outcome than having a highly polarized winner who half love and half hate when the goal of an elected official should be to represent as many of their constituents as possible.

[1980phipsi]

There are properties that different voting systems have. Not all voting systems can fulfill all of these properties. In many cases, it is a trade-off of some having some properties but not others and people deciding what they care about more. For instance, RCV satisfies the majority criterion, meaning that if a candidate receives more than 50% of the vote, then they are the winner. However, it does not satisfy the Condorcet criterion, which is that if there is a person who wins in a pairwise matchup against every other candidate, then that person is the winner.

RCV is also sensitive to different tactical strategies. You can rank a weak candidate higher in the hopes that they do well in one part of the run-off and then worse later. You may also have a preference for voting a less preferred candidate first for other reasons.

I wouldn't say it's perfect. It has some nice properties. Other systems may have nice properties too. The useful question to me is if you compare the current system vs. RCV, would it elect of those who more closely match the preferences of people they represent. I would say yes to that, but I also think Condorcet methods would do an even better job.

[nickff]

Have you seen any research on which parties or groups might benefit most from Condorcet methods? It would seem to me that it would benefit smaller parties who share something in common with a couple of major parties (such as the Libertarian party in the USA).

[kibwen]

The more typical definition of an extreme candidate is not "this person is not my top pick", but rather "I would not vote for this person at all". Candidates under RCV win when they get 50% of the vote, so if you're so extreme that only a small minority bothers to rank you then you're not going to find yourself elected on behalf of a minority. This is in contrast to FPTP, where any extremist can win with any miniscule percent of the vote if there's enough vote-splitting going on.