Wikipedia talk:How to hold a consensus vote

I'm writing this page after seeing several unrelated attempts at holding a vote to find consensus between many options, all of which devolved into arguments about voting. I think it would help Wikipedia discussions if people have a resource they can look to on what kinds of voting work and what kinds don't, and when it is or isn't appropriate to use a complex voting system.

RSpeer 19:50, May 8, 2005 (UTC)

I think the first and only aim should be to achieve consensus. Voting can be a guide to whether there is a consensus (as can comments that come along with votes), but it can be no more. It is not a be-all-and-end-all.

This in itself means we should shy away from the more complicated voting systems - because it is not the vote that's important - it's whether consensus is achieved. How "votes" and comments are counted really shouldn't matter so far as that's concerned - if there is a consensus it will emerge under all possible ways of counting votes.

• I don't agree with this. I think that there are many ways of counting votes that will devolve into a fiasco and lead to no consensus, when one could have been achieved. And I've seen these a lot recently. That's my point in writing this article. RSpeer 20:35, May 8, 2005 (UTC)

I'm particularly concerned that the more recent ones are not only experimental, but are also supported by WPians with an outside interest in showing they work. This is not the purpose of WP, and we should not encourage the development of ever more complicated and ever more theoretical voting methods. The other problem is that the raw marks by this voting method will, in part, be dictated by who understands (including who wants to be bothered to understand) the voting method. None of this helps with consensus.

• I fully agree. Hence my quoting of WP:WIN#Wikipedia is not an experiment in democracy. RSpeer 20:35, May 8, 2005 (UTC)

So in summary - aim for consensus. Look at comments as well as "votes" - and keep it simple so everyone can understand! jguk 20:27, 8 May 2005 (UTC)

Most of this is the point I'm trying to make. I made the assumption that if you get past part 1.1, you've already given up on discovering a consensus through comments, though, so I didn't say anything about that. I'll add a section.

RSpeer 20:35, May 8, 2005 (UTC)

On your most recent edit - if you have to explain the voting method to someone it proves you haven't done your survey right! jguk 20:59, 8 May 2005 (UTC)

This edit is on the "fourth choice" situation, where you've determined that you need a Condorcet vote, because simpler methods won't do. You can argue that this should very rarely happen, of course.

It's true that you don't really need to explain Condorcet - you can just say "rank your preferences, and I'll try to find a consensus from them" - but people will not believe this, as we have seen. So you have to give them some kind of idea of how Condorcet works.

Part of how Condorcet works is that "preferences" do not just involve options you like, but can involve ranking an option you are indifferent to over one you despise. This is unintuitive, and you can't just expect people to immediately grasp this from a one-sentence description of Condorcet.

RSpeer 21:16, May 8, 2005 (UTC)

My point is that we should never have a "condorcet" method. They are too complicated to explain - and, as noted above, the aim should be to achieve consensus: we do not need complicated voting methods to do that. They are also highly experimental. Why should WP choose highly experimental, little-understood voting methods? jguk 21:41, 8 May 2005 (UTC)

Yes, I see now from the styles discussion that you are strongly opposed to Condorcet voting. That's too bad, because the Condorcet method is a proven method of finding a consensus between many competing options, recommended by such guides as Robert's Rules of Order. And it may have been "experimental" during the French Revolution, but now it's pretty well established. You may be confusing Condorcet with modern cycle-breaking methods like Cloneproof Schwartz Sequential Dropping. Such methods are experimental and difficult to understand, and that's why I oppose using CSSD on Wikipedia.

And I recognize that Condorcet voting is little-understood. That's why I put it on this page - so that when it needs to be used, it can be understood better.

You seem to be opposed to the idea that you should ever have to express any opinion about an option that is not your favorite. If you have many options, and everyone only votes for their favorite, how do you expect to find a consensus? The point is that when you can't make everyone happy, you can at least make some people happy and others indifferent. It's called "compromise", and it's an important part of consensus.

RSpeer 22:00, May 8, 2005 (UTC)

I'm opposed to condorcet voting methods as they do not help us achieve consensus. If anything they hamper it as they will mean one side declaring "victory" in a marginal position, which those on the other side will not understand. In situations such as you describe, it's best not to be counting votes but looking at comments. You cannot force the issue and claim a consensus where there is none - and this sort of voting can only hinder. As I note above, I hope never to have to see it on WP again. Kind regards, jguk 06:12, 9 May 2005 (UTC)

Emphasized surveys instead of votes

I changed section 1.1 to emphasize that informal surveys work well on Wikipedia, and that a vote should only be used if a survey has been attempted unsuccessfully. RSpeer 00:32, May 9, 2005 (UTC)

darn, they're not called surveys

It looks like a "survey", according to Wikipedia:Survey guidelines, is a fairly formal kind of vote. I don't know if there is a good name for the "discussion under the heading you support" format. I stopped calling it a "survey" in the project page. RSpeer 00:48, May 9, 2005 (UTC)

It's called a strawpoll. Zocky 17:52, 9 May 2005 (UTC)

None of the above != Status quo

"None of the above" means "better nothing than other options", not a vote for the status quo. If "none of the above" wins, the practical result will be the same as status quo, but status quo did not win the consensus. Zocky 17:52, 9 May 2005 (UTC)

Condorcet section

In a way, I'm writing that section at cross purposes with myself. I think Condorcet votes are great when applied correctly, but I've also seen how prone they are to break down on Wikipedia. So I'm suggesting that Condorcet might be a viable option, but also doing some damage control by explicitly warning people not to use it without a very good reason, and to stay away from overspecified algorithms like CSSD.

Note that I have to fend off both sides: people like User:jguk who say that Condorcet is never appropriate, and people like User:Whig who say that Condorcet is the only reasonable choice if there are more than two options.

If I didn't mention Condorcet at all, someone would come along and look at the page and conclude one of these things:

• Oh. He's an Approval fanatic. I don't trust this page if it only mentions one multi-choice voting method.
• Wikipedians don't know enough about voting methods. I'll show them what a really good voting method is by using CSSD.
• Hmm. These procedures don't fit my situation. I want to know what people prefer, so I'll use a preferential voting method like... (looks at the voting method article) IRV. That sounds good.

The idea I'm going for is this: Suppose you're in the middle of a debate that needs to be resolved with Condorcet (and therefore suppose that such debates exist). The advantage that Condorcet would bring, that simpler methods wouldn't, is to find a compromise option. But you can't find an informed compromise if most people are voting A>*, saying "I want A or nothing at all". You can't get a compromise from that information. So somehow, in this situation, you'll need to let voters know that their full preferences are appreciated.

RSpeer 18:44, May 9, 2005 (UTC)

Action recently in the Wikipedia:Manual of Style (biographies)/Survey on Style-Prefixed Honorary Titles survey suggests to me that it is, indeed, possible to "game" Condorcet given the pecularities of voting on Wikipedia. You can't make it more likely for your preferred choice to win by voting insincerely. But when you can see everyone else's vote and change your vote, and you see your preferred option is losing, you may be able to vote in such a way as to make a Condorcet cycle more likely. I've suggested on User talk:Whig that this may be a facet of Condorcet voting that's entirely unresearched. (Then again, it may just be that I haven't read enough theory to know if it's new or not.) In any case, it complicates the case for Condorcet on Wikipedia in my mind, though in general I'm a Condorcet supporter. TreyHarris 00:19, 11 May 2005 (UTC)

You know, that's a very good point. I figured that the cycles were arising in that vote just because alternative 4 was so unclear.

Along those lines, someone on the election-methods list made a model of approval voting that had people adjusting their votes based on a sequence of polls, and the result converged on the choice that maximised some sort of utility, or was the Condorcet winner, or something. I really don't remember the details, but it was a fairly nifty result, because it was a rare situation where everyone voting strategically led to a better result.

I have to wonder if constant polling (which is basically what we have here) lets people take strategic advantage of irrelevant alternatives, which would make any ranked method perform worse than usual.

This is all pointing to Approval being the best method for Wikipedia. But the problem with supporting one voting method is that nobody with any sense will believe it. So many people writing about voting methods say "method foo is the only way", and they're all hiding some downside of method foo in their explanation. What should I say on the page?

RSpeer 01:33, May 11, 2005 (UTC)

There are a number of reasons why condorcet is inappropriate for Wikipedia. First and foremost - we operate by consensus. A consensus result will shine out from any voting system - it is not useful that condorcet systems can provide a "winner" where there is no such thing.

Condorcet systems, where used, tend only to be used where the promoters of the system wish a vote to go in a particular way. That may mean that they are prepared to accept their second-best option rather than risking what they see as the worst option. For example suppose I really would like position A, and detest position B. I can select position C which is a mild compromise on A and force a vote (where I choose the questions) between A, a very well-worded C, and a poorly-worded B. The result is obvious - I might get A, but if I fail to get A, I get B. Either way, I have fixed the vote. (Unfortunately we do have a very poorly worded B in the MoS (bio) vote, where Whig is erroneously claiming that "Dear Leader" is an honorific style that we ought to consider appending to Kim Jong-il's name. That, coupled with a principle of not changing wording for a vote after it has started has made things extremely divisive on that page.

Finally - and also importantly - on WP we get comments - we can see how everyone has voted, and more often than not why. This isn't true of political elections. On 5 May I made one cross on a ballot paper. But there was no way of conveying why I had made that cross. This isn't true on WP - we have oodles of info in addition to a straightforward vote. We should use that information to see if there is a reasonable consensus - not polarise issues by selecting a "winner" by condorcet (or indeed any other method), jguk 05:30, 11 May 2005 (UTC)

Condorcet systems, where used, tend only to be used where the promoters of the system wish a vote to go in a particular way. This is a rather inflammatory claim. Can you provide concrete examples, without speculating on people's motivations? Or are you stating your opinion as fact?

You seem to be conflating two issues here, but I think part of that is our language. There are at least three common usages of consensus. One, as a synonym for unanimity, is erroneous, but I don't think anyone's under that misapprehension here—I won't refer to it again. A second usage is an accord; the alternative which optimizes the preferences of the whole. A third usage is an opinion forged out of discussion. You seem to be denying the second usage exists, and insisting that the third usage is the only one.

When days, weeks or months of discussion do not result in an opinion that "shines out", as you put it, one can either continue to discuss, despite all evidence that an impasse has been reached, or one can agree that a consensus opinion is not being forged and agree to abide by the—dare I say it (Wikipedia is not an experiment in democracy)?—will of the majority.

Once that happens, it is an admission that no amount of discussion will forge the opinion that will shine out, and that simply moving on demands action. At that point, any voting system is an admission of defeat. It so happens that—and I'm aware that you don't believe this, and I probably will not ever persuade you—the Condorcet method has been designed to find the opinion which optimizes the preferences of the whole, to reach the accord which is another kind of consensus. Condorcet does, however, require voters who a) are willing to vote on their sincere preferences; and b) are not fooled by the rhetorical fallacy of false compromise.

That second bit bears expansion. (I'm embarassed to see that there seems to exist no reference in the Wikipedia to the Fallacy of False Compromise; I'll try to fix that soon. In the meantime, here's one reference. Update: turns out somebody had created an article in the interim on false compromise. TreyHarris 07:24, 13 May 2005 (UTC)) If one bloc wants to build a building on the seashore, and another bloc wants to build it on an island, it would be an example of false compromise to declare that the whole therefore want to build it in the water halfway between. If that were given as an option in a Condorcet vote, and a large fraction of the electorate did fall for the false compromise, it could indeed emerge as the Condorcet Winner. I'm tempted to say that a great deal of the problem with the style survey is the blind faith by many that there must be a compromise to be reached, somehow, some way. (I also think one of the other problems is that one of the alternatives seems to be "none of the above" and another one seems to be "no, really, none of the above".) But this doesn't impugn the reliability of Condorcet's method; it simply highlights that, indeed, this is a thorny and divisive issue, one that simply does not have a consensus. That can happen, and no amount of railing that it can't will make it so.

(I'm reminded of a caucus I observed once at a leftist political event that was trying to decide where to go for lunch. They could not reach consensus; they would not vote, because they operated via consensus. They continued to "process"—the word they used to mean discussion—until they noticed that the lunch period was over. Ludicrous.) TreyHarris 07:19, 11 May 2005 (UTC)

You said:"If one bloc wants to build a building on the seashore, and another bloc wants to build it on an island, it would be an example of false compromise to declare that the whole therefore want to build it in the water halfway between. ". There are some other option too....

1. Built half size building on island, and half size building on seashore.
2. Built whole building on island, let it stay there for half human lifetime years (aprox 40 years), then destroy it an built again the whole building on seashore and let is stay again for 40 years.

I really doubt that the option asking to "build the building in the water halfway between" will ever be ranked in a good position, but if you insist we can try it, in real conditions and see the results. Even in case the "build the building in the water halfway between" option will be selected, this should still be implemented, as long as with this option the people are actually declaring that "we really dont want the building, but we are borred to death and we do want to have some vain job to do".

A poor workman blames 07:40, 13 May 2005 (UTC)

New Condorcet section

I've revised the Condorcet section to warn against it some more, but acknowledge that it could work under the right conditions. RSpeer 06:57, May 15, 2005 (UTC)

New Borda section

I added this section because Borda count has been successfully used to build consensus.--Fahrenheit451 21:41, 15 May 2005 (UTC)

Welcome, creators of voting fiascos

I'm happy that this page got made a guideline. This, unfortunately, seems to have gotten the attention of the people who created the counterexamples to successful votes.

I don't want to make personal attacks using this page, so I am not naming any names. I just want to prevent future voting problems, due to the way they cause snowballing conflicts that obscure the actual issues.

Let me just say that if you created a vote that became a problem, you are probably not in a position to give advice, and your point of view on what happened is unlikely to be neutral.

RSpeer 01:10, May 25, 2005 (UTC)

NPOV dispute

RSpeer is POV-pushing and unilaterally claiming that cycle-breaking methods should never be used in Condorcet polls, and further making his opposition to Condorcet methods plain. There is no problem with him having this position, but his opposition must not be given as some sort of consensus guidance which was ever adopted by anyone. This entire article, in fact, was RSpeer's personal project, and as such he may feel a certain proprietary interest in its contents, but this is not the way Wikipedia is supposed to work. Furthermore, his unilateral deletion of balance in the Condorcet section was made after someone else had declared this article to be a consensus guidance (on what basis I cannot find, either). This is just a giant mess, at the moment, and I don't see a simple solution, I don't want this dispute, but please don't expect that any one person can or should unilaterally dictate who may or may not contribute to an article, especially one as contentious as this. Whig 03:01, 25 May 2005 (UTC)

Forgive me for having a "personal project" to improve Wikipedia. Would you prefer if I delegated this issue to a committee comprised of you and jguk? You can have a CSSD vote on it. Good luck!

To be more serious, I don't think I'm pushing much of an agenda here. I have allowed changes to the article (like Fahrenheit451 adding the Borda section) that do not match my personal opinions on voting methods. Why do I revert Whig's changes and not Fahrenheit451's? Because Fahrenheit451 has demonstrated, through other edits, that he is knowledgeable about the pragmatic aspects of voting methods, and Whig has demonstrated the opposite.

I guess this makes me "one person dictating who may or may not contribute", but this is only because nobody else is taking the initiative. My goal with this page is to encourage successful votes, and it seems implausible that it can do that when Whig adds a paragraph encouraging exactly the kind of vote that failed spectacularly.

Whig, I am on your side in the styles debate, I think that CSSD is a wonderful method for non-consensus-based elections, I think the RfC against you is frivolous, and I think that jguk is slightly more unreasonable than you are. But you have not shown that you know how to organize a vote. You are clearly pushing an agenda to say that Wikipedia should use CSSD to resolve debates, and this is an opinion that very few Wikipedians agree with.

RSpeer 03:55, May 25, 2005 (UTC)

A couple of questions

This whole flurry of making things into "guidelines" without having really decided what that means is a bit disconcerting. So, I'd like to ask a couple of questions to see if I understand the intent of this page correctly.

• Is it meant to describe how votes to "officialy" establish actionable consensus are (or should be) held? If so, it would probably need a lot of discussion before it's anything more than a proposal. If not so, it should be moved to a better name.
• How binding is it meant to be? Would votes have to follow it to be considered legitimate?

Zocky 03:18, 25 May 2005 (UTC)

"Guideline" is the new name for "semi-policy". No, I don't think a vote should have to "follow" this page. I'm not even sure what "following" would mean, since this page gives a lot of possible ways to set up a vote. I was quite impressed that this page was made a guideline so soon. Perhaps this page shouldn't be a guideline yet, but if it has to wait for Whig to stop saying "Use CSSD for everything!" and jguk to stop saying "Never use Condorcet for anything!", it'll never happen.

Choosing a voting method is difficult, requiring taking a lot of things into consideration, and Wikipedia's pages on voting methods may not even help you - they discuss these methods in the context of political elections, not consensus. I want to collect some past experience from Wikipedia to make votes go smoother in the future. Thus, someone who is starting a vote can look at this page for guidance.

RSpeer 03:55, May 25, 2005 (UTC)

Please don't misrepresent my viewpoint, I am not saying "use CSSD for everything" and in fact my edit did not push that POV, but rather said to consider the drawbacks and benefits of cycle-breaking. You struck that and replaced it with text never to break cycles, which is in effect to dictate against Condorcet in any case, regardless of how well or badly I may personally have managed the survey on prefixed styles. (If cycles cannot be broken, then Condorcet can always be sabotaged by those who wish to block consensus by creating the most marginal and minority-favored of cycles.) This is not meant to be a debate, I don't ask you to agree or disagree with me. I only ask that this like any Wikipedia entry maintain balance and not represent the personal opinion of one editor, excluding neutral discussion. Whig 04:11, 25 May 2005 (UTC)

Btw, and just a personal note. I've met Nick Tideman many years ago and if I was going on a personal bias here, I'd be an advocate of Ranked Pairs. I feel he has a very comprehensive understanding of economics and social choice theory, but the method is considerably harder to explain and gives insubstantially better results than CSSD, IMHO (though the nice graphs you can generate are very pretty to look at, they do less to really clarify the result than to make some people feel more comfortable with it). I also understand that Marcus Schulze is an editor here, though I've had no personal discussions with him. While he might reasonably have a bit of bias, perhaps he might also have some insights on the limitations of his method. I think this page, if it is going to be a useful guidance, needs to be the product of a community process, I object principally to the unilateral dictation of guidelines by one person who might think that a particular survey or two went badly (which may not be due to the method used at all, but other factors entirely). I think the styles survey was going to be disrupted no matter what method we chose, and we probably got better information under CSSD than we would have under Approval or other methods, despite their greater popularity among some number of Wikipedians (largely, perhaps, due to their precedence, rather than any particular merit or demerit versus CSSD). Whig 04:50, 25 May 2005 (UTC)

If you can create a Condorcet cycle, you can strategically manipulate how that cycle will be broken. I think it's far better to get no conclusion from a cycle than to get a manipulated one.

I've encountered Markus Schulze in a few places. He's made some really helpful edits, especially when it comes to providing/correcting mathematical information about voting systems. But I think he's smart enough to stay the heck away from this kind of debate.

RSpeer 06:38, May 25, 2005 (UTC)

I think we're having a fundamental misunderstanding, and I hope to clarify what we're talking about. A Condorcet cycle can exist where there are multiple distinct majorities for differing sets of preferences, and in this case the result is highly ambiguous because there exists a majority both FOR and AGAINST a member of the Schwartz set. Breaking these cycles, even though potentially possible with CSSD, may be very undesirable because it would require disposing the votes of the "smaller" majority to achieve nothing more than the best "plurality" outcome, a very weak result, and one which the majority of ballots explicitly opposed in some fashion. This isn't so remarkable, really, in standard FPTP elections, pluralities often do prevail, as a candidate receiving 40% of the vote versus two other candidates with 30% each would "win" but would be opposed in fact by 60% of the voters. However, this goes against consensus, and I agree we should not break cycles in this case.

However, and as my prior edit made clear, a faction with less than a majority seeking only to disrupt consensus could create a "weak cycle" which prefers an option by less than a majority of total ballots to an otherwise prevailing choice. Breaking this cycle will always mean removing the smallest minority-preferred option, leaving the majority satisfied. Furthermore, nothing, and no possible arrangement, can cause the cycle to be broken in any other way. An option carrying a majority cannot be defeated by a cycle having only minority support.

With that said, the existance of a cycle, even a weak one like this, indicates a need for some sort of ratification, in my opinion, especially where the prevailing alternative was prescriptive, as a majority here cannot be taken for granted as consensus. It is important, in particular, to note that in breaking a "weak cycle" in this way, we must recognize that those who expressed no preference are deemed satisfied with either option, and opposition to the prescription may require resolution in the reverse direction, that is, to break the cycle in the opposite direction. That appears to be how the survey on prefixed-styles will resolve. A majority (absolute majority, not plurality) oppose prefixed styles. A significant minority support them in all cases, but this is unachievable and rejected by many (particularly American biographical) page editors. A minuscule minority support them selectively. And every other option defeated the option to disregard the survey. So there will be a result, and your comments that the survey "ended" in some odd and ambiguous fashion are not really correct, the survey includes the ratification which is still ongoing, and a referral to arbcom may be necessary to rule on the validity of the outcome, but it is absolutely unquestionable that there exists no consensus for prefixed styles and a substantial NPOV dispute opposed to them.

Sorry for getting a bit off-track from the subject of this page, but I feel that your comments on the project page and edit comments have brought the survey into question prematurely. In any case, I hope you appreciate my distinction between strong and weak Condorcet cycles, and that you will not prejudge that breaking cycles must inherently be a bad idea. My edits were neutrally framed, and you have given no argument against them, except to unilaterally revert them. Please respond here and/or restore my comments, but of course feel free to supplement or extend upon them, to resolve the NPOV-section issue. Whig 10:04, 25 May 2005 (UTC)

No majority voting

Wikipedia does not support majority voting. We attempt to reach consensus. We're not going to implement *any* majority voting systems, except in extremely exceptional cases. (en.wikipedia previously only used majority voting for arbcom elections. I believe this might be declared a failed experiment and such an election might not be repeated) Kim Bruning 18:50, 13 July 2005 (UTC)

To exclude majority voting, one would need to exclude majoritarian methods. In the discussions written above, condorcet methods were extensively discussed, but these are majoritarian. The only two common non-majoritarian systems I am aware of are Approval voting and Borda count.--Fahrenheit451 21:45, 30 July 2005 (UTC)

We don't vote. We use consensus. Sometimes we hold polls. Consensus is arrived at by discussion, and sometimes by negotiation. Kim Bruning 23:58, 30 July 2005 (UTC)

In my opinion, the whole discussion misses the point. Facing the fact that the larger Wikipedia gets the less probable unanimity gets, we should rather discuss which quorum should be required to replace the status quo. The consensus-seeking decision-making procedure can then be chosen quite independently from this quorum requirement. Markus Schulze 3 Aug 2005

I deleted the first sentence from the project page, because it is so obviously untrue:

This is not a current practice on Wikipedia.

We vote all the time: on vfd's, RFC's, RFA's. And the arcbom votes on RFArb's. To say that Wikipedia:voting does not exist is a either a deliberate lie or a misunderstanding of the word vote. Calling "voting" polling does not stop votes from being votes. Just look at Votes for deletion. All it is, is voting.

Saying that we vote infrequently is also incorrect. It has gradually taken over, to the point where we hardly ever have rational discussion at all.

Saying that we should not vote, or that our votes ought not to be considered binding, is another thing.

I have corrected policy pages in a few places (yes, I too wish I had saved the diffs). One said that "voting at Wikipedia is infrequent" and rhapsodized about how visitors simply marvel over how we get anything done without voting. Well, I say we DO vote, and that it's killing us.

For Pete's sake, the current title of this page is "Wikipedia:How to hold a consensus vote"!

well, how do you hold a "consensus vote" ? Isn't that like one of those newspeaky contradictios? I definately haven't seen condorcet voting used yet! ;-) Kim Bruning 21:27, 7 August 2005 (UTC)