Support score voting

[wolftune]

The best system for elections is scoring by a whole range of measures, not even subtle. We need more support for this sort of approach. See, e.g. http://electology.org

FWIW, this would be the difference for whether the helios system is usable by some organizations I'm involved with.

The simplest initial version would just be approval voting where voters can vote for any number of candidates as simply an "I approve" vote and the one(s) with the highest score win. The full score voting version lets people use a range (5 stars or 1-10) to score their judgment of each candidate, and include the ability to abstain from scoring someone; then the winner(s) are those with the highest average score.

[kfogel]

At one organization that uses Helios, we are discussing the Single Transferable Vote (STV) method, a form of ranked-choice voting that is quite appropriate for many elections (e.g., open seats on boards of directors), but which is not supported by Helios. There's a fork of Helios called Zeus that does support STV, but Zeus appears to be less-actively maintained that Helios, so it's tough to choose it even though it supports STV.

Has anyone in the Helios project looked at supporting STV, either through copying code from Zeus or implementing it independently? The question of STV appears on slide 19 of Ben Adida's 2010 presentation about Helios.

@wolftune, FWIW, I wasn't sure whether to create a new issue about STV or just tag along in this issue. I guess I'll put this here for now; it's part of the general question of supporting more (and better!) voting methods in Helios. Yes, we all know Arrow's Impossibility Theorem, but that doesn't mean some voting systems aren't better than others :-). The current choose-only-1-out-of-N system is pretty much the worst, I think.

[wolftune]

@kfogel be super wary of all forced-ranking systems. They are pathalogical and can result in endless complex debates about the various ramifications of the ranking algorithms. You can do endless discussion about this, and you'll never get to something without really weird problems and people skeptical about it all.

Here's some concrete examples: http://www.rangevoting.org/STVPRunger.html and http://www.rangevoting.org/PRcond.html

There's no sense in which you'll benefit from STV over range/score voting. Score voting is far simpler, less pathological, and everyone can immediately understand that it's just the average score, end of story. Whenever strange ranked-vote results happen, people can endlessly evaluate and screw around with all the strange ramifications.

It's not hypothetical, see this rank-vote real-world election: http://www.rangevoting.org/Burlington.html where this is literally true: "Kiss won, but if 753 Wright-voters had switched their vote to Kiss, that would have made Kiss lose!"

I urge you to support my request for score voting and use it. It's just like giving stars to movies or books or whatever, very familiar, and zero ambiguity, and unambiguously the best voting system, regardless of not being "perfect".

One last addendum: the only main reasons anyone rejects score voting is because it violates "later-no-harm" and "majority-rule" which are actually an undesirable features that people wrongly think are desirable. Later-no-harm says that if A is my first choice, then where I put B should never cause A to lose the election. It sounds nice, but actually, if B is everyone's second-choice and A is top choice of a plurality and hated by everyone else, then we should elect B as everyone supports B. B is the consensus candidate. Score voting says that we do want the candidate who everyone thinks is 4/5 stars over the candidate even a majority thinks is 5/5 but a large minority thinks is 1/5. In other words, score voting is for electing the best representatives that everyone agrees on, regardless of whether it's the majority's top choice. Majority-rule is actually not a healthy principle compared to consensus. So, anyone who says "sure, ranked systems are problematic but so is score voting" is generally using bad criteria for evaluating why they think score voting is a problem.

[kfogel]

@wolftune:

Er? Isn't it possible to come up with pathological results for any voting system, including score voting? I'm too lazy to actually construct one now, but if I just wave my hands and say "Arrow's Theorem", is that enough? :-)

A presentation of pathological cases that occurred, or could occur, with STV or other ranked-choice systems is not, by itself, evidence that such systems are inferior to score voting. One would have to also examine the pathological cases of score voting. I haven't done so. If you have, then it would be useful here to hear more about that.

None of the systems proposed result in ambiguity per se: the algorithms are deterministic. The results may sometimes be counterintuitive or feel wrong, but that's different from being ambiguous.

Meta-note, of the epistemological variety: I'm not against score voting, or necessarily even more in favor of STV than of score voting. What we have here is a classic problem: although there really are some experts who know much more about the relative strengths and weaknesses of various voting systems than non-experts do, we have no procedure for recognizing such experts (unlike with, say, physics, where we have an academic credentialing system that works reasonably well, etc). So I'm essentially asking "How can I (and others) recognize you as an expert?" I trust that question will be taken in the spirit of genuine inquiry with which I ask it; it's not a statement about you, but about how to evaluate expertise in a relatively new field.

[wolftune]

@kfogel please take my word for it: there is not some total-relativism here. Ranked voting systems are fundamentally far more pathological and there are not comparable pathologies with score voting.

Score voting actually achieves all of Arrow's Impossibilities! http://rangevoting.org/ArrowThm.html The entire concept of Arrow's Theorem is that you can never have a good rank voting system. It doesn't address score voting at all.

The point is: people have done the research and the work, and I've read it, but it's all thorough on http://rangevoting.org — you're completely right that there are strengths and weakness that relate more to values. I.e. I think consensus candidates everyone supports are better than polarizing candidates loved by some and hated by others. That value is consistent with score voting.

Let me be as clear as possible: There are people who have spent years of expertise on these things, and I've read their work pretty thoroughly. All the critics of score voting I've ever encountered are either ignorant or actually disagree with the values I mentioned (i.e. they think strongly that a polarizing candidate supported by the majority should win over a consensus candidate).

Is it enough for you that I can show you that score voting actually solves Arrow's Theorem?

P.S. Besides my research, the wide range of folks involved with Snowdrift.coop discussed this issue in forming our bylaws, and everyone came to consensus on this, so if you respect the judgments of the many other folks, you can respect that we did our homework thoroughly here.

[kfogel]

Thanks; yes, that's enough for me. I was looking at http://electology.org/ for this comparison, and they didn't have it. Meanwhile, I went to http://rangevoting.org/ but was so put off (I'm admitting this openly) by the secondary signals sent by the site's look-and-feel and terrible visual presentation of what it's trying to say that I didn't spend much time there.

Whether consensus candidates are better than polarizing candidates is indeed a complex question, and one about values. A group could vote on which of those values they prefer, but then... Well, I'm sure you've been down that road before in your thoughts too :-).

FWIW: https://identi.ca/kfogel/note/Fic-llKzRmiWhl3vIwg3Gw and https://twitter.com/kfogel/status/705857077768376320

[wolftune]

@kfogel :) for reference, the electology folks have specifically taken the rangevoting.org stuff with the goal of making it more presentable, but need to do a far better job. All of this could use professional curation and editing. Electology focuses, as a practical matter, on approval voting which is very similar to the lowest-resolution score voting and has mainly all the same advantages with just lower resolution.

[kfogel]

Electology just needs a section explaining to voting geeks why approval-voting and score-voting are better than ${SOME_OTHER_SYSTEM_VOTING_GEEKS_KNOW_ABOUT}, that's all.

[kfogel]

@wolftune BTW, I think the way in which score voting avoids Arrow's Theorem is simply by refusing voters a way to express a preference they might actually have (e.g., the voter may be forced to give the same score to two candidates even though the voter actually has a finer-grained preference between them). This isn't a showstopper, but I do think http://rangevoting.org/ArrowThm.html is a bit unfair on this point.

[wolftune]

@kfogel don't want to continue long discussion here, but score voting can have a range of anything, even 1-100 (or 0-99) so if you want higher resolution, you can get it. The other criticism of score is a case with someone who thinks they can get their top choice by refusing to consent to someone they otherwise still like, so they strategically devalue their second choice just to favor their first. That approach is risky (could end up with neither!), evidence shows people wanting to express their honest judgments (they feel bad devaluing someone they like, people like expressing their judgments), and the most extreme strategic score voting is still better end result than strategic voting in any other system. In the end, strategic voting can't be eliminated, but no voting system is immune to it, and score voting retains its advantage regardless (the strategic version of score is better than the strategic version of other systems). Anyway, if people really care about expressing their subtle differences (the opposite of strategic voting, super-honest-detail voting), you can give them resolution to do so if that matters.

[kfogel]

Good point, thanks @wolftune.

[dessalines]

Score voting would be very easy to implement, and questions should absolutely have it as an option. Just do 0-100, and take the averages. I've implemented it in two different websites, if anyone needs help with it.

[wolftune]

For reference, I think it's far better to make the range a setting the user can adjust. There's evidence that 1-10 is a nice enough resolution and more cognitively comfortable than a 100-point resolution, and 0 holds a special significance in people's minds so 1-10 comes off more neutrally (it feels like less of a jump to go from 1 to 2 in your judgment than from 0 to 1). But none of this is really fundamental to score voting, so it just makes sense to offer a few options of ranges (including the option for a 2-point range, i.e. 0 or 1, which basically is approval voting).

Options I suggest considering:

• range options: start with either 0 or 1, end with either 1, 3, 4, 5, 9, 10, 99, or 100
• option to leave a candidate unscored
• option to include an explicit "no opinion" vote either instead or in addition to option of leaving blank
• option to require a percentage of voters entering a score in order for a candidate to be viable (so that an unknown candidate with a few supporters and 95% "no opinion" can't win)

Of course, starting with any single version of score voting would be great. These extra options would just make the system most robust and adaptable to everyone's needs.

[dessalines]

I agree.

I could see 3 default options as useful:

• 0-1 (IE, approval)
• 1-10
• 1-100

No opinion as a default is a must IMO, people shouldn't have to score absolutely everything unless they want to.

I've implemented a voting threshold percentage in another app, the easiest way to do it(if you want) is to

• Find the candidate with the most # of votes, name as maxVotes.
• If your threshold is 20%, then make sure only candidates with votes > 0.2 * maxVotes are included in the results.

[wolftune]

Note that for true approval voting, the 0-1 should actually be displayed as blank vs checked, and no-opinion wouldn't be an option (it would be the same as blank). Whereas a score vote of 0-1 with a no-opinion default is just low-resolution score voting.

I agree that it should be fine to leave things blank and that's counted as no-opinion. It's just that having an explicit no-opinion spot to mark may change behavior because it feels different than leaving blank, so that's a nice option to offer.

[wolftune]

Btw, I should have mentioned that the multi-winner election situation should offer Reweighted Range Voting which is basically STV-style of multi-winner applied to score voting.

Incidentally, Score Runoff per http://www.equal.vote makes a lot of sense to support. Both of these would be trivial enough to support once plain score voting is available.

[wolftune]

In recent discussions among voting reform people, I've come to the thinking that STV is not as flawed as IRV (and furthermore that despite Arrow's theorem applying, Ranked Pairs works out to be acceptably better than IRV to be okay). I've heard that STV with a none-of-the-above (NOTA) option could be pretty good.

I still think approval and score are good systems and want to see them used more (and we'll get more empirical real-world info from their use). Score Runoff is among the best too.

Finally, it should be emphasized that generally-aligned members of an organization are always going to be a different matter than the concerns of elections from the general public to typical public-servant representative seats…

[posting this on this thread because I want to emphasize that my initial push-back on STV was from it's being basically the same as IRV, and I'm now convinced that the flaws are less serious in STV circumstance than in IRV (and here's an improved STV idea)]

[eternaleye]

I'll note that while Arrow's theorem (and Gibbard-Satterthwaite) is specific to ranked voting systems, the more general Gibbard's theorem applies to any deterministic process of collective decision.

Further extensions (Gibbard's 1978 theorem, Hylland's theorem) also cover nondeterministic processes.

[BenSmyth]

I wrote a brief note on using Helios for ranked voting. Although my approach is possible, it is inefficient. By comparison, the mixnet variant of Helios (USENIX Security'08) is efficient. That variant has been implemented as Zeus (mentioned above) by @grnet and helios-server-mixnet by @RunasSudo.