Jump to content

Wikipedia talk:WikiProject Voting systems/Included methods and criteria

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Purpose of this page

[edit]

I'd like to use this talk sub-page to propose adding or deleting methods or criteria. Methods and criteria that are included on the voting system page - and only such methods and criteria - should appear in the table and should be cross-linked between their pages. Including or excluding a method or criterion thus requires many edits, so it should be discussed first.

I propose that you add or remove something from this page by adding a section below, and achieving consensus after 5 days. (If there's no discussion at all except for the original proposal, consider that to be consensus via apathy.)

RSpeer 05:13, Jun 13, 2005 (UTC)

I request something related to what you have proposed: A consensus for consistent use of voting method criteria in voting method articles. For example, the methods used are the major ones listed on the voting methods page, but any related criteria is used only if the primary criteria is met: a condorcet method may be evaluated against Smith and Schwartz. A majority method may be evaluated for mutual majority criterion. Any methods not meeting the primary criteria, would not be evaluated against the related or generalized criteria. This would be absurd, but has occured on some edits.--Fahrenheit451 01:05, 15 Jun 2005 (UTC)
I think that ideally, the pages for all voting methods should have an explicit passed/failed criteria page, and that all of the core criteria found on the voting system page should be listed separately there. Imagine that a method passes Smith. It is possible to save a tiny bit of space by neglecting to mention that the method passes Condorcet, majority, mutual majority, and Condorcet loser, but not all wikipedia readers realize that Smith implies all of these, and so the more terse edit has a cost in terms of clarity and accessibility. Hermitage 02:01, 16 Jun 2005 (UTC)
I agree with your example, but I don't think it is necessary e.g. to say that a method fails Condorcet and Smith, or Majority and Mutual Majority. KVenzke 15:33, Jun 16, 2005 (UTC)
I can (grudgingly) accept that if a method fails a given criterion (say majority), then we do not absolutely need to also say that it fails a criterion that is very specifically an extension of that criterion (say mutual majority). This also applies to Condorcet and Smith. By the way, is it possible to fail participation but pass consistency?... Hermitage 21:56, 16 Jun 2005 (UTC)
Maybe. To quote from Woodall 1994: Convexity is one of the best-understood of all properties. Young proved that a symmetric preferential election rule for single-seat elections satisfies convexity if and only if it is equivalent to a point-scoring rule ... All serious point-scoring rules satisfy Participation. (Not all methods satisfying Participation are point-scoring rules; note DSC.) So perhaps consistency implies Participation. I can't see a way around that conclusion. KVenzke 22:31, Jun 16, 2005 (UTC)


(1) main article, (2) other method pages, (3) table

[edit]

Based on comments, I'm seeing that there should perhaps be a distinction between

  1. adding something to the list of methods/criteria
  2. adding cross-links to it from methods or criteria.
  3. adding it to the table

I was taking "add" to mean all three.

I think that 1 and 2 should always go together. I suppose there are cases where 1 and 2 but not 3 would be justified; for example, a criterion that is similar to other criteria that are already represented in the table. I can't see the argument for methods as well; the horizontal space for a new column is hard to find, but vertical space for a new row is easy. Even if the criteria satisfied by a method are the same as for another method, people may not know that, and find the repeated row of the table informative. RSpeer 23:14, Jun 16, 2005 (UTC)

I was mostly concerned with (1) & (2). So if you're opposed to (3) for a criterion/method, but in support of (1) and (2), please clarify that in your response. Hermitage 02:08, 17 Jun 2005 (UTC)
Ok, I will. KVenzke 07:08, Jun 17, 2005 (UTC)
Well, it seems to me that every criterion will be uncontroversial for 1 and 2 except for unpublished criteria. Correct? KVenzke 23:37, Jun 16, 2005 (UTC)
I can't think of any published criteria/methods that I'd like to exclude from the main article and method pages, but I'm not sure that everyone agrees on that. Condorcet loser is surely published. Smith is published, at least as a set, which implies the criterion. Minimax is surely published. I'm not aware of summability being published. Mutual majority hasn't been published under that name, but I believe that it has been published (e.g. by Woodall) under other names. Hermitage 02:08, 17 Jun 2005 (UTC) (Er, actually my article published in Voting Matters does define the mutual majority criterion... I don't know if that counts.) Hermitage 02:12, 17 Jun 2005 (UTC)
Smith is published as a criterion in Woodall, "Monotonicity of single-seat preferential election rules", Discrete Applied Mathematics 77 (1997) pp. 81–98. I'd be interested to hear who has used Condorcet Loser in a publication. Was that name used in the book you mentioned? KVenzke 07:08, Jun 17, 2005 (UTC)
"In modern terms the plurality winner A is a Condorcet loser, i.e. an alternative that would be defeated by all other alternatives in pairwise comparisons... Borda's criticism against eh plurality voting boils down to the argument that the Condorcet loser should not be elected." (Nurmi 1999, p12) I'm not going to look it up now, but my impression is that the Condorcet loser criterion has been used in a fairly large number of published articles, over a long period of time. Hermitage 08:00, 17 Jun 2005 (UTC)
Just because a method is published doesn't mean it's any good or that anyone else accepts it. For example, I'd consider someone's Ph.D. dissertation to be published. Back when I was a math major and researched this stuff, I ended up reading such a dissertation. It defined a Condorcet voting method based on taking the eigenvalues of the defeat matrix. A mathematically clever trick, but it turned out to be totally crap as a voting method; it hardly satisfied any useful criteria, not even monotonicity. RSpeer 03:29, Jun 17, 2005 (UTC)
Of course I agree that not all published criteria are worth mentioning on wikipedia, and that not all unpublished criteria should be dismissed out of hand. This is just one aspect of the criteria that we're looking at, I think. Hermitage 08:01, 17 Jun 2005 (UTC)
I think material that is not published and peer reviewed should receive special scrutiny and discussion before being used for evaluation purposes.--Fahrenheit451 17:38, 19 Jun 2005 (UTC)
I would say that being unpublished is an argument against something, but not the last word. Likewise, being published is an argument for something, but not the last word. Hermitage 21:20, 20 Jun 2005 (UTC)
For example, the various Woodall monotonicity criteria have been published to some extent, but I don't think that they are worthy of mention on wikipedia. (e.g. mono-raise-random, mono-sub-top, mono-remove-bottom, etc.) --Hermitage 23:36, 21 Jun 2005 (UTC)

Proposals

[edit]

I added info for Bucklin voting

[edit]

I hope this is uncontroversial. That article did not even have a list of satisfied criteria. KVenzke 15:11, Jun 19, 2005 (UTC)

I agree that there should be a substantial page for Bucklin. I added a couple of the criteria that seem to be already approved or on their way to approval. I think that the strong defensive strategy criterion should probably be removed from that list, but I haven't done it yet. I have several objections to SDSC. Let me address them in a new section (see below)... Hermitage 22:02, 20 Jun 2005 (UTC)
I definitely oppose removing SDSC from Bucklin's list; it's the entire reason I consider it preferable to IRV. KVenzke 03:26, Jun 21, 2005 (UTC)
This discussion continues below, under the SDSC heading.

purge strong defensive strategy criterion from method pages

[edit]

part 1

[edit]

I have various objections to SDSC, including but not limited to the following:

(1) The criterion doesn't consider strategic truncation as "falsely voting two candidates equal", thus allowing several methods to pass that would not do so otherwise. I consider this to be misleading to the point of intellectual dishonesty. Since truncation in these methods has the same effect as voting candidates equal in last place, it should be considered to be a type of equal ranking for analytical purposes.

(2) The name is far too vague, and makes the criterion sound much more important than it is.

(3) The definition of "defensive strategy" that underlies the criterion is very vague and convoluted.

(4) It has never been published, and is mainly used by Mike Ossipoff and people who are strongly influenced by MO. Hermitage 22:10, 20 Jun 2005 (UTC)

(1) If truncation were considered to be "falsely voting two candidates equal," then no method could satisfy SDSC. I can understand if you find the wording misleading; I think the wording could even be changed to be more clear. (I may do this.) But suggesting that SDSC should be mean something else seems odd. KVenzke 03:42, Jun 21, 2005 (UTC)
What I'm suggesting is that the criterion shouldn't be referenced on individual method pages. I don't mind if it keeps its own page. --Hermitage 07:34, 21 Jun 2005 (UTC)
I believe it should be referenced on the pages of methods which satisfy it. That affects what, three or four articles? KVenzke 20:52, Jun 21, 2005 (UTC)
That might be a good compromise. Hermitage 22:13, 21 Jun 2005 (UTC)
Agreed, then? KVenzke 14:49, Jun 22, 2005 (UTC)
Possibly. Let's see what other people have to say. --Hermitage 22:40, 22 Jun 2005 (UTC)

part 2

[edit]
(2) The problem seems more to be that it isn't published, not that it isn't important. I consider a form of it to be the most important. KVenzke 03:42, Jun 21, 2005 (UTC)
Even if the criteria in its mathematical form is important, I still think that the name is very non-descriptive and vague. However, can you make an argument for why you find the criterion to be so important? As F451 suggested, I think that non-publication should place a somewhat higher burden of doubt on criteria. --Hermitage 07:34, 21 Jun 2005 (UTC)
If the name of the criterion is what damns it, then I give up. KVenzke
Names are fairly important. E.g. if the only available name for the Condorcet criterion was the "ideal democratic winner criterion", I would not want it on wikipedia, for reasons of bias. I feel that many of MO's criteria names have a similar effect. --Hermitage 22:13, 21 Jun 2005 (UTC)
Well, in the case of SDSC, I don't see the problem. The word "strong" differentiates from "weak," as in other criteria (possibly in other fields). Without that, you have "defensive strategy criterion." That sounds quite neutral to me, and it seems like a reasonable description. KVenzke 14:49, Jun 22, 2005 (UTC)
Again, using the term "defensive strategy" opens up a can of worms, because the definition of that term is so convoluted. Also, the word "strong" is problematic. There are several ways to make WDSC stronger; this is just one of those ways. --Hermitage 22:40, 22 Jun 2005 (UTC)
Eh?? It sounds like you feel WDSC has some legitimacy that SDSC doesn't; WDSC is also just one way to make SDSC weaker. I don't really see the problem with the fact that "strong" and "weak" don't define themselves in a word. KVenzke 14:23, Jun 23, 2005 (UTC)
Anyway, the term "defensive strategy" is still problematic.
I like the way Steve Eppley divided the criterion into three tiers rather than two. These are (1) non-drastic defense (similar to WDSC), (2) minimal defense (similar to SDSC), and (3) sincere defense (similar to SDSC without the truncation provision). It might make sense to create one page for all three of these criteria, and then redirect WDSC and SDSC to this page. --Hermitage 01:09, 24 Jun 2005 (UTC)
What would User:Hermitage like to call it? KVenzke
That's a good and difficult question. Maybe the "ease of exclusion criterion" (EEC)? Perhaps SDSC could be the "ease of exclusion by truncation criterion" (EETC), and WDSC could be the "ease of exclusion by equal ranking criterion" (EEERC). Remember, the criterion doesn't assure us that a particular candidate does win, but rather that a particular candidate does not win; hence I suggest something with a similar meaning to "exclusion", rather than "defense". I'm open to other suggestions. --Hermitage 02:21, 22 Jun 2005 (UTC)
In my mind, "defense" already means "defense against." Actually, I think "defensive strategy" and "minimal defense" are probably too modest; I consider this criterion a prerequisite for majority rule. KVenzke 14:49, Jun 22, 2005 (UTC)
I disagree with your take on the word "defense". I take it that you don't like the name "ease of exclusion criterion". I wouldn't mind working the word "majority" in there somewhere. Sigh, maybe you're right in that changing the name is more trouble than it's worth. --Hermitage 23:13, 22 Jun 2005 (UTC)
I understand that "defensive strategy" doesn't mean "defend against." But Wikipedia doesn't define "defensive strategy," so I think this could be fixed by suggesting (by wording in the commentary) under SDSC that it is defense "against." "Ease of exclusion criterion" does not seem descriptive to me. I do think "majority" should be worked in, so that it somehow contains the "majority defense against something" notion. KVenzke 14:23, Jun 23, 2005 (UTC)

part 3

[edit]
As far as an argument, I wrote the commentary under Minimal Defense criterion. This is from my webpage:
Minimal Defense deals with the issue of what a majority needs to do to get their opinion counted. Specifically, if they are united in preferring candidate A to candidate B, all they have to do is not rank B. They need not do anything special regarding A. For instance, on these ballots: (cut)
The A>B voters (i.e., the voters preferring A to B) are a majority, and do not rank B at all, so that Minimal Defense guarantees that B won't win. However, candidate B is the winner in e.g. Plurality, Instant Runoff, and Descending Solid Coalitions. Under these methods, if the A>B majority want to prevent B from being elected, they need to insincerely raise candidate A in their rankings.
A more general way to view this problem is by noting that this election is primarily a contest between A and B. In that light, it would be very undesirable for the election method to elect the wrong one of these two. Minimal Defense ensures that the method can't be "confused" by the introduction of weaker candidates preferred to the major candidates.
Thanks for that. Good points. I'll agree that there is an element of usefulness to the criterion. However, IMO various problems remain with the way the criterion is defined, named, presented, supported, etc. --Hermitage 22:13, 21 Jun 2005 (UTC)
My position is that the best situation would be to have a votes-only MD article. But since we can't, having the SDSC article is far, far better than having nothing. KVenzke 14:49, Jun 22, 2005 (UTC)
To summarize, SDSC and MD say that the method won't elect the wrong one of the two major candidates when the supporters of each candidate don't rank the other. The Four Majority Rule Criteria of James Green-Armytage are less important than this. By the way, is it true that the reason you prefer IRV to Bucklin is because Bucklin only satisfies one of the Four Majority Rule Criteria of James Green-Armytage, while IRV satisfies two? KVenzke 20:52, Jun 21, 2005 (UTC)
Why do you put "four majority criteria of JGA" in caps and quotes, with a dead link? I hope that it isn't an expression of rancor. Anyway, I list 5 criteria under the category of "majority rule" on my website. IRV passes majority, mutual majority, Condorcet loser; fails Condorcet and Smith. (3/5). Bucklin passes majority, mutual majority; fails Condorcet loser, Condorcet, and Smith. (2/5). No, Condorcet loser failure is not my main criticism of Bucklin. To be honest, I haven't thought seriously about Bucklin for awhile, and Bucklin isn't really at issue on this page, but I don't mind discussing it anyway, if you're able to be patient with me. First of all, as you understand Bucklin, are equal rankings allowed, or not? I think that Bucklin with equal rankings is much, much better than Bucklin without equal rankings. So far, the Bucklin page doesn't seem to mention how equal rankings are counted. Hermitage 22:13, 21 Jun 2005 (UTC)
Not rancor, just the impression that you're a bit dogmatic about criteria. But I did not expect you to say that Condorcet Loser is not the factor by which IRV trumps Bucklin, so my impression might be wrong. (I do consider these five criteria to be quite inadequate for ordering methods according to majority rule.) I don't understand Bucklin to typically allow equal ranking, but I agree it's better if equal ranking is permitted, from the perspective of incentive to vote sincerely. I don't necessarily consider Bucklin better than Approval. KVenzke 14:49, Jun 22, 2005 (UTC)

part 4

[edit]
(3) Did you see the commentary I added? I believe I made it much clearer. I would like to remove references to "offensive strategy," but both Mike and Steve Eppley use this notion to justify their versions of this criterion. KVenzke 03:42, Jun 21, 2005 (UTC)
It's not the SDSC page that I am critical of, but rather the criterion itself. Even if you make the SDSC page as good as possible, I would prefer that it was not referenced on method pages. --Hermitage 07:34, 21 Jun 2005 (UTC)
Well, I'm eager to hear you be critical of SDSC (the criterion, not the name). I got so far that (1) you want to change SDSC to be something similar but better, (2) you don't like the name, (3) you don't like to use the term "defensive strategy" in explanations, and (4) people "strongly influenced by MO" are particularly unimportant. KVenzke 20:52, Jun 21, 2005 (UTC)
Yes, I do think that the criterion should be reworded so that it's more obvious from the criterion itself that truncation plays a central role.
Your note ("A voter who does not explicitly vote for two candidates (i.e., by truncating them, leaving them out of their ranking) is not considered to be voting them equal.") is helpful, and I greatly prefer its presence to its absence, but I still believe that in general truncating should be considered as a kind of equal ranking for most analytical purposes. Hermitage
I find it baffling that you say this. It's like someone arguing that Majority criterion should be removed because a better criterion is MMC. KVenzke
First of all, that seems like a flattering comparison for SDSC, in that SDSC is not even close to being as well-established as MC.
But, more to the point, I'm arguing at this point about the analytical treatment of truncation, which is a more general concern than the definition of SDSC. In nearly all cardinal and ordinal methods, leaving a candidate off the ballot has the effect of voting them as tied for last place. If we say on one page of the wikipedia (the SDSC page) that truncation isn't considered to be a kind of equal ranking, then we seem to be making a general statement about truncation that is independent of the definition of SDSC. --Hermitage 22:40, 22 Jun 2005 (UTC)
You seem to have fixed this now in the SDSC article. KVenzke 14:23, Jun 23, 2005 (UTC)
The way I interpret SDSC does not place much emphasis on the meaning of "equal ranking." The emphasis is rather on "voting for" X and "not voting for" Y. KVenzke 14:49, Jun 22, 2005 (UTC)
My opinion is that the distinction of "voting for" a candidate versus "not voting for" them is primarily an artifact of the plurality ballot, with limited relevance to analysis of ordinal and cardinal systems. --Hermitage 22:40, 22 Jun 2005 (UTC)
I think this is quite wrong. Voters are never going to "vote for" all the candidates. If you force them to, they will just use random or alphabetical ranking. (This is exactly how I test votes-only MD on methods that disallow truncation.) Apparently you don't even believe LNHarm compliance would fix this. KVenzke 14:23, Jun 23, 2005 (UTC)
I think that the phrase "I voted for candidate X" loses its meaning on an ordinal (or cardinal) ballot. For example, let's say that I ranked all candidates in the election, but that I ranked Bush in last place. Does it make sense to say that I "voted for Bush"? In my opinion, it doesn't. --Hermitage 01:09, 24 Jun 2005 (UTC)
It only loses its meaning from the perspective of the vote counter. Voters still "vote for" and "don't vote for." I agree that it makes no sense to say you "voted for Bush." But if you were forced to rank all the candidates, then I contend that there are probably other candidates for whom we cannot say that you "voted." If you chose to rank all the candidates, then I'd say you "voted for" everyone other than Bush.
Anyway, my point is that this distinction doesn't become meaningless just because theoretically everyone can submit a complete ranking; particularly as strategically it doesn't usually make sense to submit a complete ranking. KVenzke 19:14, Jun 24, 2005 (UTC)
I find it extremely realistic to suppose that voters will not rank both major candidates, without even reflecting on strategy concerns. KVenzke 14:49, Jun 22, 2005 (UTC)
I tend agree with that, but it's not relevant to the question of whether truncating should be treated as equal ranking. --Hermitage 22:40, 22 Jun 2005 (UTC)
It's relevant to the question of whether SDSC is meaningful because it makes this distinction. As I said earlier, I don't believe you can do much with the "must have some way of voting" phrasing. KVenzke 14:23, Jun 23, 2005 (UTC)
Yes, I think that the name of the criterion should be changed into something more neutral.
Yes, I think that MO's definition of "defensive strategy" is a mess, and I think that the SDSC page suffers greatly from the use of the terms "offensive order reversal" and "defensive strategy", terms that are not defined on wikipedia, and that are IMO too convoluted and vague to be worthy of wikipedia.
No, I did not mean to imply that anyone's opinion was unimportant. --Hermitage 22:13, 21 Jun 2005 (UTC)
I have since rewritten the paragraph that used those terms. (That means all of the SDSC commentary on the page is written by me, and none is from electionmethod.org.) KVenzke 14:49, Jun 22, 2005 (UTC)
I'm glad that we can agree not to use those terms. I definitely think that your changes are an improvement. --Hermitage 22:40, 22 Jun 2005 (UTC)
Actually, I made quite a few edits to the SDSC page myself. Please let me know what you think. --Hermitage 23:13, 22 Jun 2005 (UTC)
Good. I don't know if you intentionally removed the italics, but I put them back. In the paragraph where you made this change: With SDSC, the majority need never explicitly vote two candidates equal (although they may need to truncate their ballots), I felt I should add another parenthetical to note that in WDSC, too, one may need to truncate. KVenzke 14:23, Jun 23, 2005 (UTC)
No, I didn't mean to remove the italics. I guess that happened when I got rid of the rather un-wiki-like html-ish formatting. --Hermitage 01:09, 24 Jun 2005 (UTC)

part 5

[edit]
(4) I guess you feel I am "strongly influenced by MO." KVenzke
I don't know; are you? While we're on the subject, I'm just curious: is there any direct evidence that "Mike Ossipoff" is not a pseudonym? --Hermitage 07:34, 21 Jun 2005 (UTC)
A pseudonym that he's used for ten years? I don't know. I'd just rather you not use guilt by association to criticize a criterion. KVenzke 20:52, Jun 21, 2005 (UTC)
Fair enough. --Hermitage 22:13, 21 Jun 2005 (UTC)
My preference would be to word SDSC more clearly, without Mike's permission (we shouldn't need it anyway), and/or to discuss Steve Eppley's version, Minimal Defense, for which he does provide a votes-only version which implies compliance with SDSC. I only use SDSC because there is no article for Minimal Defense. (Incidentally, Woodall has proven that Chris Benham's SCRIRVE fails Minimal Defense in a personal mail. I was very impressed.) KVenzke 03:42, Jun 21, 2005 (UTC)
I think that votes-only minimal defense criterion is more well-defined and well-named than SDSC. However, there are methods that pass SDSC, and pass a "stronger" SDSC (which considers truncation as equal ranking), but still fail votes only minimal defense. (E.g. approval-weighted pairwise, I think.) As with SDSC, I doubt that minimal defense is notable enough to be referenced in all voting methods articles on wikipedia. --Hermitage 07:34, 21 Jun 2005 (UTC)
"Stronger SDSC" would say that if a majority rank X over Y, Y can't win. An approval cutoff doesn't fix this; it actually breaks standard SDSC. KVenzke
No, DMC/RAV and AWP satisfy standard SDSC, and also SDSC without the "truncation addendum". That is, if a majority ranks X over Y, approves X, and disapproves Y, Y cannot win, even if all candidates are given strict rankings. I think that these methods do not satisfy votes-only minimal defense. The difference is that in SDSC, the majority has leeway to use the approval cutoff to help prevent the election of Y. For this reason, I prefer SDSC to votes-only minimal defense when dealing with anything other than a pure ranking method. --Hermitage 22:48, 21 Jun 2005 (UTC)
So what does your proposed criterion say? "If a majority prefer X to Y, then they should have a way of voting by which Y loses, and all preferences are voted sincerely"? My criticism of this, and of SDSC itself, is that "having a way of voting," some way of voting, is not very helpful. I don't envision that the majority benefiting from SDSC is discussing their strategy prior to the election. They just have the common sense to not list the opposition candidate as a compromise. KVenzke 14:49, Jun 22, 2005 (UTC)
I can't understand what your purpose is in discussing "stronger" SDSC in the middle of a paragraph arguing that unpublished criteria are not notable. KVenzke 20:52, Jun 21, 2005 (UTC)
Just to point out what SDSC would mean if it wasn't for that extra addendum saying that truncation isn't considered to be a type of equal ranking (when, in my opinion, truncation is effectively a type of equal ranking). --Hermitage 22:13, 21 Jun 2005 (UTC)
I edited SDSC as above. After reading the page again, I'm rather annoyed with your complaint (1). I had made it quite clear that truncation didn't count. Your complaint seems to be that one sentence, taken out of context, sounded too good to be true. KVenzke 04:34, Jun 21, 2005 (UTC)
Didn't you make the change after I made the complaint? If so, why would you be annoyed? Anyway, to sum up, let me repeat that voting the MO criteria pages for deletion is not my priority here. What we're doing is deciding on a standard list for criteria to be included on individual method pages, and I'm suggesting that SDSC (along with the other MO criteria I listed below) should not be on that list. --Hermitage 07:34, 21 Jun 2005 (UTC)
No, it wasn't. I'm saying the page was already clear at the time you wrote your complaint. But apparently what you consider "misleading to the point of intellectual dishonesty" is not what was written, but what it meant! KVenzke 20:52, Jun 21, 2005 (UTC)
I made my initial criticism of SDSC on this page at 22:10, 20 Jun 2005. At 04:17, 21 Jun 2005, you added the clarifying sentence: "A voter who does not explicitly vote for two candidates (i.e., by truncating them, leaving them out of their ranking) is not considered to be voting them equal."
I think that without the clarifying sentence, the definition of SDSC was unclear and misleading. --Hermitage 22:13, 21 Jun 2005 (UTC)
I didn't add it, I essentially moved it; there had been a "Definitions" section defining what it means to "vote equal," and I had already placed a sentence clarifying that truncation didn't count. So if one were to read the "Definitions" section prior to reading the "Statement of Criterion," they should not have had the impression that truncation counts. KVenzke 14:49, Jun 22, 2005 (UTC)
To sum up, I like your suggestion that SDSC should only be referenced on the pages of methods that pass it. However, I still think that there are serious flaws with SDSC as currently conceptualized. I would be more open to the criterion if
(1)the name of the criterion, and the page for the criterion, didn't use the poorly-defined terms "offensive strategy" and "defensive strategy".
Done. The reference is instead to the burial strategy on Tactical voting. KVenzke 14:49, Jun 22, 2005 (UTC)
I approve of those changes in the text. Of course, the term "defensive strategy" is still used in the name of the criterion itself, but perhaps there isn't much that we can do about that. --Hermitage 22:40, 22 Jun 2005 (UTC)
(2)the allowance for truncation was not an addendum to the criterion, but rather an integral part of the criterion --Hermitage 22:48, 21 Jun 2005 (UTC)
How about this for the definition:
"If a majority prefers candidate X to candidate Y, then they should have a way of voting that will ensure that Y cannot win, without any member of that majority (1) reversing a preference for one candidate over another or (2) falsely voting two candidates equal above last place." --Hermitage 22:48, 21 Jun 2005 (UTC)
This is in the right direction, but I'm not convinced that "last place" is clear. The distinction between "voting for" and "not voting for" seems undeniable, even if not expressed very clearly. KVenzke
I don't think that it is undeniable. I don't know how to word it better while keeping this distinction. I still prefer my definition. --Hermitage 22:40, 22 Jun 2005 (UTC)
Rather than "above last place," I'd like to say something like "above other candidates." I don't have a strong objection to your current suggestion. What do you think of Eppley's version? KVenzke 14:23, Jun 23, 2005 (UTC)
So far I have not tampered with the definition, since arguably Mike (or Russ, or whoever wrote this formulation) has the right to say how it is worded. KVenzke 14:49, Jun 22, 2005 (UTC)
If we necessarily have to conceptualize the criterion the way MO conceptualizes it, then I'm definitely opposed to having it on wikipedia at all, let alone linking to it from other pages. Ossipoffism may have its insights to offer, but it is too self-enclosed and polemical to be considered as encyclopedically neutral truth. --Hermitage 22:40, 22 Jun 2005 (UTC)
Well, Wikipedia doesn't say "SDSC is important," it just reports MO's idea. A criterion devised by X could be racist, sexist, harmful, and intuitively stupid, while a Wikipedia article reporting this criterion could still be "encyclopedically neutral truth." However, there might not be anything wrong with rephrasing the definition so that it says the same thing. My concern is with tampering with something that appears to be a direct quote. KVenzke 14:23, Jun 23, 2005 (UTC)
I don't think that any other voting criteria on wikipedia are expected/assumed to be a direct quote from the paper in which they were first defined. --Hermitage 01:09, 24 Jun 2005 (UTC)

part 6

[edit]
By the way, can you tell me which methods, aside from approval, pass WDSC but fail SDSC? Thanks! --Hermitage 22:48, 21 Jun 2005 (UTC)
Almost any method that allows equal ranking at the top. I believe it's been shown that ER-IRV(fractional) fails WDSC. Woodall's DAC and DSC fail it. In my opinion, WDSC and SDSC should be merged into one criterion which preserves the "vote for" vs. "not vote for" distinction, and drops the talk about reversing preferences. Then Approval and three-slot Schulze (for instance) could satisfy this criterion. WDSC is just not very encouraging. KVenzke 14:49, Jun 22, 2005 (UTC)
I'm confused. How about a list of methods that pass WDSC and fail SDSC? --Hermitage 22:40, 22 Jun 2005 (UTC)
Approval, ERBucklin (and MCA), limited-slot versions of methods which satisfy SDSC, MMPO, ERIRV(whole) (I'm pretty certain), margins (doesn't it?)... KVenzke 14:23, Jun 23, 2005 (UTC)
I'd be interested in seeing a SDSC failure example for ERIRV(whole) and beatpath(margins). Maybe it would make sense to condense WDSC and SDSC into one page, with proofs, failure examples, and a table with methods that pass (1) neither, (2) WDSC only, and (3) both. (Or, as I suggested above, use SE's names for the criteria on this page instead of MO's.) At this point, there seems to be less clarity about which methods pass WDSC and SDSC then there is with regard to the more traditional criteria, which may be another handicap when it comes to wikipedia use. --Hermitage 00:23, 24 Jun 2005 (UTC)
It's quite clear that ERIRV(whole) fails SDSC, since when no one uses equal ranking, it's equivalent to IRV. It's also quite clear that beatpath(margins) fails SDSC (look at a scenario illustrating the inability to use truncation as a deterrent). What I'm not 100% about is whether they satisfy WDSC. I think they probably do. I wrote that list off the top of my head, though; I'm either right or wrong, there is no ambiguity.
A small note: The relationship between votes-only MD and votes-only NDD doesn't seem to be the same, in Eppley's mind, as the relationship between SDSC and WDSC. Votes-only NDD seems to say that the majority may need to compress the compromise to the top, but not truncate.
Later today I'll get you some scenarios and consider your "one page" suggestion; I have to run at the moment. KVenzke 15:51, Jun 24, 2005 (UTC)
49 a
24 b
27 c>b
This scenario works fine for both ERIRV(whole) and anything(margins). A wins under both, and B>A voters have no sincere way to vote which defeats A.
I'm not really opposed to creating a single page on defensive strategy criteria. As far as which methods we could list in each category (limited to methods actually likely to be on Wikipedia): (1) IRV, Borda, FPP. (2) Approval, Minmax, margins methods (If I'm right that margins satisfies WDSC). (3) RP, Schulze, Bucklin. KVenzke 19:02, Jun 24, 2005 (UTC)
As far as I know, there are no methods currently on wikipedia that satisfy Eppley's "sincere defense" criterion. Just out of curiosity, which methods pass? I'm thinking DMC/RAV, AWP, CWP, probably approval margins, probably various methods with AERLO/ATLO... it would be nice to put some of those on wikipedia, but I don't know if any of them are widely-recognized enough yet.
I'm pretty darn sure that ERIRV(whole) satisfies WDSC. (If a majority prefers X to Y, and they rank X as tied for first place, Y will certainly be eliminated before X.) You were correct in saying that ERIRV(fractional) fails WDSC, although I believe that it passes WDSC in the three candidate case. I'm not sure if margins methods pass WDSC. --Hermitage 03:47, 25 Jun 2005 (UTC)
I had not put any thought into "sincere defense." Reading it now, it doesn't seem promising to me. I can't speculate on which methods satisfy it. Of the methods you listed, I would not currently want to see any on Wikipedia. DMC/RAV sorely needs a more descriptive name (like "Approval-Elimination Condorcet").
It does seem right to me that we should add methods that are interesting due to their satisfied properties. For example, I think Woodall's DAC almost deserves an article, since it satisfies MMC and Participation. KVenzke 04:36, Jun 25, 2005 (UTC)
I agree that approval-elimination pairwise or approval-elimination Condorcet is a better name than DMC or RAV.
I personally don't think that DAC deserves an article. I find it to be bizarre and impractical for public elections. I expect that it would have rather perverse strategic incentives. --Hermitage 28 June 2005 08:13 (UTC)
I don't advocate DAC either. But it satisfies a remarkable set of criteria (Majority and Participation, as I said; I also note Mono-raise-delete and Mono-sub-plump). I wouldn't call the strategic incentives "perverse"; it seems to me there is just strong incentive to compromise, and not vote for weak candidates. KVenzke June 28, 2005 21:20 (UTC)

part 7

[edit]

I support this. In fact, I didn't think we had to vote on this, because it's not on the voting system page, so it shouldn't be in articles in the first place. Does that sound like a reasonable position? RSpeer 05:47, Jun 22, 2005 (UTC)

It seems to me that listing SDSC under methods which pass SDSC is useful. KVenzke 14:49, Jun 22, 2005 (UTC)
I give KVenzke very full credit for arguing the merits of SDSC in good faith (and I find the discussion quite interesting) Certainly I would prefer KVenzke's limited presence proposal to having it on every method page (which I would find quite unnacceptable), but I still find the criterion to be rather problematic in general, so I'm not sure if I can endorse even the limited proposal. I'd like to hear more opinions at this point. --Hermitage 22:40, 22 Jun 2005 (UTC)
I support purging it from the methods articles.--Fahrenheit451 23:01, 22 Jun 2005 (UTC)
At this point, I'm undecided between my initial proposal (remove from all method pages) and KV's limited presence proposal (include on pages of methods that pass). --Hermitage 07:08, 23 Jun 2005 (UTC)
The point I want to make is that the fact that a method satisfies SDSC is, at worst, useful commentary on the method. Listing SDSC as a satisfied criterion is just the simplest way to convey it. KVenzke 14:23, Jun 23, 2005 (UTC)
This is a good point, although of course the same argument can be made for any voting criterion. --Hermitage 03:32, 25 Jun 2005 (UTC)
I don't agree with that. As an example, I don't think that "method X satisfies Condorcet Loser" would count as "useful commentary" on method X, especially if Condorcet Loser were not well-known. KVenzke 04:27, Jun 25, 2005 (UTC)
Sorry, but I don't understand the difference. --Hermitage 28 June 2005 08:15 (UTC)
The difference is whether there is any serious impact. If Condorcet Loser is not a well-known criterion, then I would say it is just trivia to note that a method satisfies it. It doesn't make guarantees to specific voters, and it just barely says something about the quality of the method's result. KVenzke June 28, 2005 21:15 (UTC)
The CL criterion is certainly more well-known than SDSC. Thus, you simply seem to be saying that SDSC is a more useful criterion than CL. I personally disagree, but I'd rather not debate it with you here or now. --Hermitage 29 June 2005 01:29 (UTC)


Removal of Local IIA from table and evaluative criteria

[edit]

Local IIA is a criteria that applies exclusively to Condorcet methods. Condorcet methods do not comply with real IIA, as any other ranked method fails. If it is a condorcet method, it is redundant to say it is local IIA. I propose to remove it from the table and where it is used to evaluate a method. This does not include the article that describes it.--Fahrenheit451 2 July 2005 23:32 (UTC)

It's not redundant for Condorcet methods. A method can satisfy Condorcet and Smith but still fail LIIA. KVenzke July 3, 2005 03:30 (UTC)
F451, your point is well-taken, but overall I think that local IIAC is interesting, and I'd rather not remove it. Certainly not all Condorcet methods pass local IIAC. I wasn't aware that Smith methods fail it. KVenzke, could you please give an example? Do Condorcet and clone independence together imply local IIAC, or not? --Hermitage 3 July 2005 07:36 (UTC)
I can give a hypothetical example. Suppose that on the ballots 45 A>B>C 30 B>C>A 25 C>A>B, the method elects A. Then suppose D is added in this way: 45 A>B>D>C 30 B>C>A>D 25 C>A>B>D. (D is not in the Smith set.) Suppose now B is elected. Then LIIA has been violated, but Smith hasn't. KVenzke July 5, 2005 14:20 (UTC)
Thanks, I do understand your explanation. I have modified my proposal however to include only the table.--Fahrenheit451 5 July 2005 19:30 (UTC)
O.K. Then, so we agree that it is interesting, but it May only apply to Condorcet methods, exclusive of non-condorcet methods? If so, then I shall modify my proposal.--Fahrenheit451 3 July 2005 19:40 (UTC)
I modify my proposal to remove it only from the table, as the column criterion is IIA, then for the two condorcet methods, the definition is changed in the blocks. This comes across as rather POV; no exceptional definitions are given for other methods, they either satify the criteria at the head of the columns or they do not. I think this is a fair proposal.--Fahrenheit451 4 July 2005 21:56 (UTC)

The modified proposal as stated above has no objections. This proposal has passed by tacit consent. The local IIA criteria will only be removed from the voting systems table. --Fahrenheit451 8 July 2005 16:41 (UTC)

Whoops. No, I don't agree with the modified proposal. I'm going to revert the edit until you can find some more supporters. Sorry for the misunderstanding. --Hermitage 9 July 2005 05:11 (UTC)

I hereby withdraw from this forum in protest. It has become a sham and a means to force POV agendas through. It looks like we go back to edit wars.--Fahrenheit451 9 July 2005 15:18 (UTC)

Nobody is trying to force anything through. I just didn't think that we had achieved consensus on that particular proposal, and that your 8 July 2005 16:41 post resulted from a misunderstanding. However, if nobody voices a strong desire to restore the local IIA to the table, I will leave your edit as is. --Hermitage 08:32, 13 July 2005 (UTC)[reply]
I think that local IIA should be in the table. - McCart42 (talk) 17:39, 14 July 2005 (UTC)[reply]
I agree. I suggest that you mention that explicitly on the general talk page, since F451 (the main opponent of local IIA) may not read this one anymore. If you put local IIA back into the table, I will support you. --Hermitage 21:16, 14 July 2005 (UTC)[reply]

Former proposals (successful)

[edit]

purge various other Mike Ossipoff criteria

[edit]

I propose that the strategy-free criterion, the generalized strategy-free criterion, the weak defensive strategy criterion, and the favorite betrayal criterion should be purged from the method pages, and perhaps purged altogether from wikipedia. Hermitage 22:15, 20 Jun 2005 (UTC)

WDSC and SDSC just survived VfD recently. I am not interested in SFC, GSFC, or WDSC. However, FBC has certainly received its share of attention on EM, from e.g. Forest and Alex. It's a pity that the only Wikipedia method which satisfies it are Approval voting, Range Voting, and MCA. KVenzke 03:29, Jun 21, 2005 (UTC)
Anti-plurality Voting, garbage method though it may be for several other reasons (most notably its failure of the independence of clones criterion), also satisfies it. Jack Rudd (talk) 18:59, 16 July 2008 (UTC)[reply]
FBC is interesting, but as I mentioned on the EM list, I think that its name (the "favorite betrayal criterion") is absurd, polemical, and insufficiently descriptive (why isn't ranking a less-preferred candidate equal to your favorite a form of "betrayal"?). Also, I think that limiting its applicability to "favorite candidates" is unnecessary, when one can make more general analytical distinctions (without the polemics) by talking about "compromsing-compression incentive" rather than "favorite betrayal".
I think that the name "strategy-free criterion" is the most absurd one of the bunch, as it very strongly implies that a complying method is free of strategy, which is obviously not the case.
Anyway, my point here is that whether these criteria keep their pages or not, I propose that they not be part of the list of criteria referenced on individual method pages. --Hermitage 07:44, 21 Jun 2005 (UTC)
I would be more enthusiastic about "compromising-compression incentive" if you could word it as a relative criterion. KVenzke
Okay, but I should have said compromising-reversal incentive, because that's the one that's closer to FBC.
Call it the "zero compromising-reversal incentive criterion". Assume that all votes have been cast except for those of a voter who is allowed n votes (n can be any number), and that the voter can read the full results thus far before voting. If the voter can ever get a preferable result by voting a less-preferred candidate above a more-preferred candidate, the method fails the criterion.
Approval and MM(PO) pass, I believe. Most ranked methods fail.
You can define similar criteria for compromising-compression (although it's extremely hard to pass), burying-reversal, and burying-compression. ZCRIC, ZCCIC, ZBRIC, ZBCIC. --Hermitage 23:08, 21 Jun 2005 (UTC)
This is still very hard to work with. I think that's the biggest problem with FBC. My rewording aims to fix that. KVenzke 15:08, Jun 22, 2005 (UTC)
I have offered a rewording of FBC under the name "Sincere Favorite" at Electowiki. As far as "why isn't ranking a less-preferred candidate equal to your favorite a form of "betrayal"?", it is, in the strong form of FBC. Even so, I don't understand that criticism. "Betrayal" does not have an obvious meaning in election methods; I can't see any reason to say that "betrayal" can't refer solely to insincere order reversal. KVenzke 21:33, Jun 21, 2005 (UTC)
No, my point is that polemical, non-neutral names should be avoided. I don't think that criteria should be given exciting-sounding "brand names" like "sincere favorite" or "favorite betrayal", but rather boring and neutral names like "consistency", "generalized majority", "Smith", "monotonicity", etc. I also think that the names should be as descriptive as possible. --Hermitage 23:08, 21 Jun 2005 (UTC)
Well, it seems to me that your descriptive substitutes fail to explain the purported significance of the criteria. That's part of being "descriptive," isn't it?
By the way, you asked why there should be such emphasis on favorites. I was asking myself the same question in wondering why I intuitively consider FBC more important than LNHarm. I think this is the answer: Cautious, strategic voters will uprank their most viable compromise to the top position. So protecting "favorites" (in the sense of "ensuring they can at least get the votes they're entitled to") ends up protecting every candidate preferred to the viable compromise. It essentially allows voters to vote at least as sincerely as under Approval. KVenzke 15:08, Jun 22, 2005 (UTC)

I support purging them as an evaluative criteria, but not the articles on these criteria.--Fahrenheit451 22:58, 22 Jun 2005 (UTC)

Yes. Anyway, purging the articles themselves is outside the domain of this discussion page. So, I should clarify my proposal on this page by saying that it only involves removing the criteria from method pages. --Hermitage 01:12, 24 Jun 2005 (UTC)

add Top-2 Runoff

[edit]

Currently, Top-2 Runoff is the second most wide-spread single winner election method (after First-Past-the-Post). Therefore, this method should be added even though it satisfies almost no important criterion. Markus Schulze. 22 Jun 2005

I agree that runoff voting has more encyclopedic significance then many of the more theoretical systems that we like, even though perhaps it isn't as intellectually fascinating. What should we add it to? It already has its own article and a place in the voting system article. --Hermitage 21:14, 22 Jun 2005 (UTC)
By the way, when Top-2 Runoff is added then also Condorcet loser should be added, so that there is at least one criterion satisfied by Top-2 Runoff and violated by First-Past-the-Post. Markus Schulze 23 Jun 2005


Majority Monotonicity Participation Condorcet Consistency IA independence clone independence
Approval No Yes Yes No Yes Yes Yes
Borda No Yes Yes No Yes No No (teaming)
Copeland Yes Yes No Yes No Local No (crowding)
IRV Yes No No No No No Yes
Plurality Yes Yes Yes No Yes No No (vote-splitting)
Range voting No Yes Yes No Yes Yes Yes
Ranked Pairs Yes Yes No Yes No Local Yes
Schulze Yes Yes No Yes No Local Yes
Top-2 Runoff Yes No No No No No No
Compliance of single-winner systems with some established voting system criteria.


Majority Monotonicity Participation Condorcet Condorcet loser Consistency IA independence clone independence
Approval No Yes Yes No No Yes Yes Yes
Borda No Yes Yes No Yes Yes No No (teaming)
Copeland Yes Yes No Yes Yes No Local No (crowding)
IRV Yes No No No Yes No No Yes
Plurality Yes Yes Yes No No Yes No No (vote-splitting)
Range voting No Yes Yes No No Yes Yes Yes
Ranked Pairs Yes Yes No Yes Yes No Local Yes
Schulze Yes Yes No Yes Yes No Local Yes
Top-2 Runoff Yes No No No Yes No No No
Compliance of single-winner systems with some established voting system criteria.


Maj Mon Part CW CL Cons IIA IC
Approval No Yes Yes No No Yes Yes Yes
Borda No Yes Yes No Yes Yes No No
Copeland Yes Yes No Yes Yes No Local No
IRV Yes No No No Yes No No Yes
Plurality Yes Yes Yes No No Yes No No
Range voting No Yes Yes No No Yes Yes Yes
Ranked Pairs Yes Yes No Yes Yes No Local Yes
Schulze Yes Yes No Yes Yes No Local Yes
Top-2 Runoff Yes No No No Yes No No No
Compliance of single-winner systems with some established voting system criteria.

Maj = Majority criterion

Mon = Monotonicity criterion

Part = Participation criterion

CW = Condorcet criterion

CL = Condorcet Loser criterion

Cons = Consistency criterion

IIA = Independence of irrelevant alternatives

IC = independence of clones

  • I'm not sure why these tables are here, but I will point out that this last, hyper-compressed form of the table is just about as bad as the original one that I replaced. Tables should be readable without a separate explanation. RSpeer 13:46, Jun 23, 2005 (UTC)
Of course, then I have to ask what we are talking about. Already now, the table at the "voting system" site is so large that we can add criteria only when we change the form of the table. Markus Schulze 23 Jun 2005
I agree that runoff and Condorcet loser should be added to the table. Runoff is a much more well-established method than most of the other methods there, CL is a more well-established criterion. I'm undecided between the big table and the compact table, leaning toward a more abbreviated table of some kind. --Hermitage 00:14, 24 Jun 2005 (UTC)
If we change the form of the table to something that requires cross-referencing with a legend to decipher, then we've lost more utility than we gain by adding a new criterion. Once there were enough criteria that I put them in two tables, which is what I'd suggest doing if we end up with that many criteria again. I put them back in one table when I got rid of the criteria that I knew were Mike Ossipoff's fault.
Note also that the table, in its current form, expands to fill some horizontal space, so don't take the amount of empty space on the right of it to be the absolute amount of space we have left. Section headings will wrap if the table reaches the width of the browser window. This is why I didn't think there was a problem with "Independence of irrelevant alternatives" as a section heading; it can wrap onto three lines if necessary.
But I will reiterate that I am strongly opposed to using acronyms as column headings.
RSpeer 01:46, Jun 24, 2005 (UTC)

The compressed table with acronyms is a very bad idea which I oppose. Only understandable tables please.--Fahrenheit451 03:16, 24 Jun 2005 (UTC)

The summability criterion has absolutely nothing to say about whether a method gives good results given a certain vote count. Advocates of the criterion claim that it has very important implications for the integrity of the count, but has this really been demonstrated? Do government elections that use IRV and STV suffer from more vote fraud than governments that use plurality, party list, etc.?

The argument given for the criterion's importance seems quite thin to me. It seems that the full (non-summable) IRV results for any given district should be able to fit quite easily on a single compact disc, which means that watchdog groups should be able to collect the full results for any given election directly. What's the fuss?

I notice that the only method mentioned on the page as failing this criterion is IRV. Does it make sense to have this criterion on every voting methods page, when only one method fails it? Why not just have a section on the IRV page that discusses the summability failure and resultant implementation challenges. Of course, that is already there, so I suggest that the logical course of action now is to delete references to the summability criterion from the pages of other voting methods. (I personally don't mind if the summability page itself stays where it is.)

Also, since other editors are interested in how widely used criteria are (esp. in journal literature), I'll ask how widely used the summability criterion is. I could easily be wrong about this, but my impression is that it's predominantly used by a smallish group of hard-core anti-IRV people. Hermitage 08:24, 16 Jun 2005 (UTC)

I oppose this. Failing summability is not damning, but is a significant feature of a method. --Fahrenheit451 18:15, 16 Jun 2005 (UTC)
I support it. This is a Mike Ossipoff criterion, isn't it? And it's on the criteria table for some reason. To disagree with Fahrenheit, for me failing summability is pretty damning. But I simply don't think this is a valid criterion, particularly in that it is only relevant to public elections. KVenzke 20:43, Jun 16, 2005 (UTC)
Yes, I suspect that it is a Mike Ossipoff criterion, and I'm also wondering how it got onto the criteria table. (To answer my own question, it looks as though the article was glutted with MO criteria until Rspeer's April 21 edit, and that "summability" was able to sneak through.) Even if failing summability is a big problem (which I doubt), it still doesn't make sense to mention it on every voting methods page when it's only an issue for one method. I say that we can let summability keep its page, and its reference on the IRV page, but delete it from the rest of the voting articles. Hermitage 21:10, 16 Jun 2005 (UTC)
Hermitage's suggestion sounds reasonable, so I support removing it. I had no idea it was a Mike Ossipoff criterion. RSpeer 23:17, Jun 16, 2005 (UTC)
Summability is a major concern. If the method doesn't meet summability, counting votes properly can become a major practical issue for implementation. You would have to transmit the data for every exact vote, which winds up being a great deal more information. See this page for more details. To sum up that page, for a Condorcet method you would need a maximum of N(N-1) vote counts, where N is the number of candidates. For IRV, you would need a maximum of N(2^(N-1)). This is a big deal for implementation, and summability should be returned to the table. - McCart42 (talk) 17:23, 14 July 2005 (UTC)[reply]
I don't agree, for several reasons. 1. Summability is a practical matter, which has little in common with other criteria, which tend to make guarantees of some kind. 2. Summability has no importance outside of public elections. 3. I don't personally know of a source which even calls summability a "criterion." 4. Methods which fail summability are currently in use in the world, so we know that failing summability doesn't make a method impossible to use.
The site you link to is reportedly written by Russ Paielli with much input from Mike Ossipoff. It would be nice to find some other sources on this topic. KVenzke 04:27, July 15, 2005 (UTC)
You talk about major practical issues, but where is the evidence that such problems have seriously compromised any real IRV or STV count? Please see my full argument above, and please try to be a bit more critical about what you read on the electionmethods.org site.
I don't consider summability to be useless as a criterion, but nor do I consider it to be among the most significant criteria. Hence, I remain opposed to putting summability on the table, especially when other criteria such as mutual majority, Smith, and invulnerability to burying are absent. --Hermitage 08:00, 15 July 2005 (UTC)[reply]
From what I understand of Australia's infrastructure, they actually have e-voting machines at every polling place, which are securely linked to servers, which then transmit the information to a central location. Here is where I first read of Australia's e-voting systems; admittedly, it doesn't provide much information. Their software is open source, a major plus, but something that might not happen in the United States, given our track record with e-voting. But here's my current concern: what do you put up as the voting results at each polling location? What numbers do you display to tell the voters how the vote went in their particular location? That's what the summability criterion means: that there is no simple way to reduce the information on a group of ballots to a meaningful set of numbers, aside from a number for each permutation of preferences. If you can't display the results of each geographic area, there exists the possibility for tampering on a large scale, because only one central location has to be compromised. That's worrisome, unless I'm wrong. - McCart42 (talk) 15:39, 15 July 2005 (UTC)[reply]
Also, you ask "where is the evidence that such problems have seriously compromised any real IRV or STV count" - a question which is similar to asking "where is the evidence that closed source non-voter verifiable e-voting machines have seriously compromised any real US election". If the source is closed, if the votes are only counted at a central location, then getting evidence that the method is being used in a corrupt or broken way is difficult. It's perhaps not a great analogy - we do have evidence that e-voting machines have screwed up major votes, such as that one out in California where 3000 some votes were lost. But there exists the possibility for tampering that we'd never know about because the source is not open, and there is no paper trail for recounts. Perhaps I'm making too big a deal out of this, and you're right that pretty much only one source exists for these complaints, but the reason I think this is a problem is because real opponents to ANY alternative voting system are going to use a complaint like this to stall IRV and scare the public, when and if any major movement proposes to make this the federal standard and gets any momentum. Incidentally, Congresswoman Cynthia McKinney (D-GA) just put up a bill for IRV in federal elections in May. - McCart42 (talk) 15:46, 15 July 2005 (UTC)[reply]
First, even if summability is a major concern, I still think that it is excessive to add it to the voting system table, and to add it to lists of criteria on method pages, because IRV is the only wikipedia method that fails it. --Hermitage
Second, you ask what kind of summaries can be made for IRV precinct results. I suggest that some fairly satisfactory summaries are possible. For example, one way could be to have the first choice totals for each candidate, as well as the totals for each candidate after each elimination that occurs in the overall race. --Hermitage
Third, I'm not sure that summability is as well-defined as the other voting criteria in the table. --Hermitage
Fourth, are there any journal articles written about summability? If it is a major concern, I don't see why it would remain unpublished. --Hermitage 00:40, 27 July 2005 (UTC)[reply]

Remove Copeland's method from table

[edit]

Does anyone really want Copeland's method to be there? The article on Copeland is a halfhearted stub that has barely been changed in two years. More importantly, the method itself cannot be used in public elections because it is seriously indecisive. I think that it should be removed and replaced with minimax. (Hopefully we'll write the minimax stub soon, right guys?) --Hermitage 10:16, 25 Jun 2005 (UTC)

I don't really want it there. I'll probably tackle minimax if no one else does within a few days. KVenzke 16:49, Jun 25, 2005 (UTC)
I support this proposal. It is a seldom used method.--Fahrenheit451 19:54, 25 Jun 2005 (UTC)
Three support, none oppose, none undecided. A consensus is established. As Hermitage proposed this, he may wish to remove it.--Fahrenheit451 30 June 2005 00:41 (UTC)
Done. --Hermitage 30 June 2005 23:36 (UTC)

I think that the Smith criterion is one of the most important criteria for single winner methods. In any given election, we cannot guarantee that the Condorcet winner will be elected (because there is not always a Condorcet winner), but we can guarantee that a member of the Smith set will be elected. We can't guarantee that no candidate who loses a pairwise comparison will be elected, but we can guarantee that any candidate losing a pairwise comparison but winning the election will have a chain of defeats back to the candidate who defeated him/her. In my opinion, this is a very important guarantee to make.

Assume that candidate S is a Smith set member that candidate N is not. S beats N in pairwise comparison, and there is no chain of defeats from N to S. Thus, I think that it is difficult to justify the election of candidate N over candidate S.

Some information about the relationship between Smith and other criteria:

If a method passes Condorcet, it doesn't necessarily pass Smith. If a method fails Smith, it doesn't necessarily fail Condorcet. If a method passes Smith, it passes Condorcet, majority, mutual majority, and Condorcet loser. If a method fails Condorcet, majority, mutual majority, or Condorcet loser, it fails Smith.

To say that a method passes Smith implies that the method passes all of these other criteria. However, not all wikipedia readers know this, so it makes sense for the wikipedia to state it explicitly (i.e. to list the other criteria in addition to the Smith criterion in the "satisfied criteria" section). To say that a method fails any of these other criteria implies that the method fails Smith, so in a sense it is redundant to say that it fails Smith. However, not all wikipedia readers know this, so it makes sense for the wikipedia to state it explicitly (i.e. to list Smith in addition to the other failed criteria in the "failed criteria" section).

BTW, Someone should expand the Smith set page. I'll probably do it eventually if no one else does. Hermitage 00:46, 16 Jun 2005 (UTC)

I support this. RSpeer 03:47, Jun 16, 2005 (UTC)
I oppose this. Criteria should be specifically named.--Fahrenheit451 18:06, 16 Jun 2005 (UTC)
I support having Smith criterion redirect to Smith set, and expanding the article. But I oppose putting it on the table of criteria. Considering the other criteria that are there, it seems like bias. KVenzke 20:30, Jun 16, 2005 (UTC)
In what way is the Smith criterion a biased criterion?? I've already argued above why the Smith criterion is important. Of course no one can prove that a criterion is important, but don't you find the argument above somewhat compelling? Not all Condorcet methods pass Smith, so it's not like Smith just beats on the point that Condorcet methods are superior. Rather, it makes a useful distinction between Condorcet methods that pass it and those that do not pass it. Hermitage 21:22, 16 Jun 2005 (UTC)
It's not a biased criterion, it's just superfluous to include on a table with little space available. For example, show me a method that satisfies Condorcet and (MMC and/or ICC) but fails Smith. I only find your argument (about Smith being important) convincing when I base everything on pairwise wins. I agree that Smith makes a useful distinction among Condorcet methods, but 1) we are not here only considering Condorcet methods, and 2) Wikipedia doesn't have articles for methods which satisfy Condorcet, fail Smith, but satisfy some criterion that isn't satisfied by e.g. Schulze. KVenzke 22:11, Jun 16, 2005 (UTC)
Clarification: I support adding Smith criterion everywhere except the table, and under methods which already fail Condorcet. (That is, I do not want to put it under methods which already fail Condorcet. KVenzke 07:14, Jun 17, 2005 (UTC)
I consent to KVenzke's modified version of the proposal. Hermitage 08:04, 17 Jun 2005 (UTC)
I oppose the modified proposal except for condorcet methods. Adding this elsewhere indicates a bias for condorcet methods.--Fahrenheit451 17:41, 19 Jun 2005 (UTC)
I consent to KVenzke's version. RSpeer 04:07, Jun 20, 2005 (UTC)
This closes tomorrow; should I consider it a consensus (all but F451) for KVenzke's proposal? RSpeer 04:07, Jun 20, 2005 (UTC)
Yes. Even F451's take is perhaps not so far from Kevin's. Smith will be mentioned on the pages of Condorcet-passing methods (which F451 is not opposed to). Smith will not be mentioned on the pages of Condorcet-failing methods (which F451 is in agreement with). Smith will be mentioned on the list of criteria on the voting systems page (1), but not the table. (3) If F451 disagrees, it is only with (1). As for me, although I would personally add Smith to all method pages, I consent to all aspects of Kvenzke's proposal. Hermitage 21:26, 20 Jun 2005 (UTC)

This proposal succeeded (in a limited form). Anyone who wants to can implement it.

This criterion doesn't even have a page yet, but it should.

First, it is intuitive. If a candidate would lose against any other candidate in a one-on-one election, it is intuitive that the candidate should not be elected.

Second, it has a long history. Jean Charles de Borda used the Condorcet loser criterion as an argument against plurality, in a presentation to the French Royal Academy on June 16, 1770. (See Hannu Nurmi, Voting Paradoxes and How to Deal with Them, p.11) It has been a fixture in voting theory ever since.

Third, not only Condorcet methods pass the Condorcet loser criterion, and some Condorcet methods fail the Condorcet loser criterion. Borda, IRV, Coombs, and Nanson pass; Minmax and Dodgson fail. Hermitage 01:53, 16 Jun 2005 (UTC)

I support this. RSpeer 04:14, Jun 16, 2005 (UTC)
I oppose it. I think having a flood of Condorcetish criteria just gives the appearance of bias. By all means give it a page, though. KVenzke 15:36, Jun 16, 2005 (UTC)
I support this. Condorcet loser criterion is a very significant property of a method. --Fahrenheit451 18:17, 16 Jun 2005 (UTC)
I'm a little confused about what "add" means. Surely not "to the table." Why is Condorcet loser a "very significant property" of a method? It seems to me it's a redundant criterion used to whack Approval voting and Minmax. I don't find it at all reassuring to note that IRV won't elect the Condorcet loser. Is he really that special? KVenzke 20:35, Jun 16, 2005 (UTC)
What does add mean? As I understand, it means to add a criterion to the Voting system page under the list of criteria, and to add it to the pages of all or most individual voting systems on wikipedia.
KVenzke, pretty much any ranked ballot criterion (aside from the "continuity criteria") whacks approval voting. Bucklin and Dodgson also fail Condorcet loser. Furthermore, I think that the fact that minimax fails it is quite significant, in that it illustrates something about the method that people might not realize at first. The criterion is definitely not a Condorcet-biased criterion, because as I mentioned various Condorcet methods fail it while various non-Condorcet methods pass it. Hermitage 21:41, 16 Jun 2005 (UTC)
You miss my point. If every criterion whacks approval, we don't need Condorcet Loser to do it too. Same with Bucklin and Dodgson, to a lesser extent. Minmax doesn't even currently have a page, and its defects are already evident from the failure of MMC and ICC. Condorcet Loser is not an isolated problem in Minmax. And in methods that satisfy it, it is not a meaningful guarantee. That is why I oppose putting it on the table. However, if all you want to do is list it under Voting system and nail Dodgson (etc.) with it, I don't object. KVenzke 22:23, Jun 16, 2005 (UTC)
Summary: I am of the opinion that the Condorcet loser criterion should be added to (1), (2) and (3), although my primary concern is with (1) and (2). Hermitage 08:07, 17 Jun 2005 (UTC)

The consensus seems to be to add the CL criterion to articles, and I think to the table. I can't tell about the table part. RSpeer 01:46, Jun 24, 2005 (UTC)

I've created an article for Condorcet loser criterion. At present, it still has a stub label. Your contributions to the article are welcome. --Hermitage 04:24, 25 Jun 2005 (UTC)
I think your coverage has already exhausted the topic. Maybe you could mention the historical significance of it.
MarkusSchulze also wants CL on the table, so I withdraw my opposition to this. KVenzke 04:47, Jun 25, 2005 (UTC)

This method doesn't have a page yet, and it's also got the problem that everyone calls it something different. But it's often used as an example of a simple Condorcet completion method, and it's been used by the American Pot Party (bastions of reputability that they are).

If you can't tell what method I'm referring to, it's the one where you elect the candidate whose worst defeat is the weakest. (If there's a Condorcet winner, they have no defeats, so they are elected.)

I'd make the article. Is the title Minimax Condorcet good? I don't want to use that double-slash notation (used in names like "Smith//Minmax") on Wikipedia, because it's unnecessary, ugly, and might not even be allowed by the software.

RSpeer 04:11, Jun 16, 2005 (UTC)

I agree that a page should exist for this method. As to the title, how about "minimax method"? (Other suggestions are "minimax voting", and "minimax voting method".)
BTW, some of its aliases are minmax, Simpson, Simpson-Kramer, "Plain Condorcet", and successive reversal.
I agree that the method should be called "minimax" rather than "minmax", although several people on the EM list refer to it as "minmax". The term "minimax" has roots in game theory and other branches of mathematics, whereas the word "minmax" does not. Hermitage 04:52, 16 Jun 2005 (UTC)
You can see Electowiki's article if you want to write a new article. I don't see why to add Minmax to the table, though, since it doesn't satisfy any properties that e.g. Schulze method doesn't. I could see adding Minmax(pairwise opposition) to the table if passed criteria such as Woodall's Later-no-harm criterion were also added to the table. KVenzke 20:20, Jun 16, 2005 (UTC)
I support adding Minmax basically anywhere. KVenzke 07:17, Jun 17, 2005 (UTC)
I don't feel strongly about adding it to the Voting system table, if there are space constraints, but I think that it should at least be added to the Voting system page. Hermitage 21:31, 16 Jun 2005 (UTC)

This proposal succeeded; consensus seems to be to give it an article and add it everywhere, including the table. (I wouldn't say there are space constraints. We have as much vertical space as we need.) RSpeer 01:46, Jun 24, 2005 (UTC)

I finally got around to writing the stub and adding it to the table... --Hermitage 08:06, 15 July 2005 (UTC)[reply]

Former proposals (failed or no consensus)

[edit]

add "pairwise defeat strength" page

[edit]

My opinion is that a single page should be created that defines both WV and margins. I suggest "pairwise defeat strength" as the title. I think that the difference between the two is certainly widely discussed and important enough to merit a page on wikipedia. --Hermitage 23:24, 21 Jun 2005 (UTC)

It seems to me you won't be able to do this usefully when there is no article for Plurality criterion, and you don't like to use SDSC. What information would this page have? KVenzke 14:54, Jun 22, 2005 (UTC)
Well, as long as the SDSC page (and pages for other criteria are a factor) is not deleted, WV and margins should be defined. Although there is no page for Woodall's plurality criterion, the Schulze page references it. Maybe there should be a page for Woodall's plurality criterion...
Also, I think that the terms have encyclopedic value in themselves. To begin with, I would just like to have the definitions of the terms. Secondarily, perhaps something can be said about strategy, counterstrategy, the Woodall plurality criterion, etc. --Hermitage 23:37, 22 Jun 2005 (UTC)
When you say as long as the SDSC page..., this makes me think that you believe SDSC is primarily about WV vs. margins. I doubt there should be a page for the Plurality criterion, since no method on Wikipedia fails it; you can add margins methods, but we don't even believe they're worth using. Finally, I can't imagine what sources you would cite regarding strategy differences; the value of defensive truncation as a deterrent to offensive order-reversal still seems disputed. I guess you could cite your own paper. I doubt you would be allowed to cite Eppley's paper. KVenzke 14:35, Jun 23, 2005 (UTC)

Okay, I've realized that a separate page is unnecessary. What I'm going to do instead is add a section on defeat strength to the Condorcet method page. I'll do it when I get some free time. --Hermitage 9 July 2005 05:24 (UTC)

I've done this now, as promised. Hope that everyone likes it. --Hermitage 09:05, 20 July 2005 (UTC)[reply]

What about IIA? I think it is one of those not necessarily damning criteria. I wonder if we should remove it as an evaluative criteria from the table and method articles? --Fahrenheit451 18:52, 19 Jun 2005 (UTC)

In my opinion, IIA can't be damning because nothing satisfies it meaningfully. I would keep it out of the method articles (except for those satisfying LIIA). It's not very informative on the table. KVenzke 22:30, Jun 19, 2005 (UTC)
Undecided, leaning toward oppose. IIAC failure is certainly not damning, but IIA is nevertheless an interesting criterion.
Approval and range voting seem to be the only methods that pass IIAC. This may provide a valid argument in favor of these methods. However, I question the real-world impact of this criterion satisfaction.
For example, if you remove a non-winning candidate after the balloting in these methods, it will certainly not impact the result. However, if you remove a non-winning candidate before the balloting, I believe that it could quite plausibly alter the result. Of course, in most real elections, candidates are not added or withdrawn after the balloting, so approval and RV's IIAC satisfaction have questionable practical significance.
Nevertheless, IIAC is an intuitive criterion, it is very well-published, and it brings up interesting theoretical issues.
In wikipedia, there is also the criterion "local independence of irrelevant alternatives", which I personally find interesting, and would probably not like to see removed. Hermitage 21:42, 20 Jun 2005 (UTC)
I oppose this; by way of Arrow's theorem, it's one of the most discussed criteria there is. RSpeer 01:33, Jun 24, 2005 (UTC)
That's true it is often discussed by voting method theorists, but all ranked methods fail it, and it does not really apply to rated methods. Methinks it certainly deserves an article, but not be used as an evaluative criteria.--Fahrenheit451 03:23, 24 Jun 2005 (UTC)

Two support, one uncertain, one oppose. No consensus. Proposal fails. --Fahrenheit451 30 June 2005 00:35 (UTC)

Fair enough. Of course, if someone wants to bring it up again later, that's fine. For now, I'll move it down to the failed proposals section. --Hermitage 9 July 2005 05:17 (UTC)

I'll begin with a proposal that may seem out of character: I want to add Majority Choice Approval to the list and the table.

Yes, I made a VfD against it once, but active editing on that page has convinced me that the method is well-known to theorists and even has a possibility of being implemented in the real world. With the amount of effort going into its Wikipedia page, it should be included on the voting system page. RSpeer 05:13, Jun 13, 2005 (UTC)

(Add your comments below, indicating whether you support or oppose this change.)

Support.--Fahrenheit451 18:00, 13 Jun 2005 (UTC)
Abstain. Oppose. MCA satisfies the same criteria as Approval voting; the exceptions are debatable. (Specifically: Whether either is clone-independent, and whether MCA really satisfies Majority when there are more than three candidates.) If it's going to be on Wikipedia, I suppose we should list it. But personally I don't feel that Forest's method is a real improvement over Approval voting: Yes, it can elect a majority favorite where Approval might not, but also, it gives few voters any good reason to use the middle slot. KVenzke 06:27, Jun 15, 2005 (UTC)
Clarification: I still oppose adding MCA to the voting system list, criteria articles, and the table. This is largely because MCA is similar to Approval; also because it's very debatable which properties MCA satisfies. I don't think it's worth going into when Wikipedia has just one three-slot method. KVenzke 07:11, Jun 17, 2005 (UTC)
Oppose. I personally have never understood the appeal of the MCA system. If you like Bucklin's properties (which I don't anyway, but that's a separate issue), why not just use Bucklin? Otherwise, why not just use approval? Hermitage 00:32, 16 Jun 2005 (UTC)

This proposal failed to reach consensus after 5 days. RSpeer 04:07, Jun 20, 2005 (UTC)

I consider this to be a useful criterion. It is not identical to the majority criterion; for example, plurality and minmax both pass majority, but fail mutual majority. I find it worth mentioning that IRV passes mutual majority, and that Black, Dodgson, and Copeland fail it.

Mutual majority is a very intuitive criterion, in my opinion. For example, if the majority of voters strictly rank all Labour party candidates above all other candidates, it seems intuitive that a Labour party candidate should win the election. Hermitage 00:57, 16 Jun 2005 (UTC)

I disagree, weakly. I haven't seen this criterion used much. RSpeer 04:13, Jun 16, 2005 (UTC)
The problem is that Mike Ossipoff coined the term "Mutual Majority" if I'm not mistaken. Woodall just calls it "Majority." Markus Schulze calls it "Majority criterion for solid coalitions." I'm sure he is quoting someone, but I don't know who. I support adding the criterion although I hate the name "MMC." KVenzke 15:35, Jun 16, 2005 (UTC)
I oppose this proposal carte blanc. However, would support it only in specific cases where the method passes majority criterion.--Fahrenheit451 18:09, 16 Jun 2005 (UTC)
I think that Mike did coin the term "MMC". According to Mike, it was based on a criterion called "generalized majority criterion" that was defined by Bruce Anderson. A google search reveals no such proposal, but nevertheless I think that the term "generalized majority criterion" makes sense. I would support changing the name from MMC to GMC, redirecting MMC to GMC, adding the criterion to the Voting Systems page, and to the pages of methods passing the majority criterion, if not all methods. Hermitage 21:28, 16 Jun 2005 (UTC)
I would prefer to use a term that is actually in print. Otherwise the name MMC is just being changed for aesthetic reasons. KVenzke 22:15, Jun 16, 2005 (UTC)
Okay, but weren't you the one who said that you hate the name MMC? If you want to change it to GMC, I support you. If not, that's okay with me too. Or did you have a different name in mind? --Hermitage 04:47, 22 Jun 2005 (UTC)
Clarification: I support adding MMC everywhere except the table and under methods which already fail Majority Favorite. (I had thought that perhaps MMC should go on the table, but when you consider that Majority Favorite plus Clone-Winner implies MMC, it looks redundant.) KVenzke 07:16, Jun 17, 2005 (UTC)
I consent to KVenzke's modified version of the proposal. Hermitage 08:05, 17 Jun 2005 (UTC) Also, I would be happy to call the criterion "generalized majority" rather than "mutual majority" everywhere on wikipedia, and to redirect MMC to GMC. Hermitage 08:10, 17 Jun 2005 (UTC)

This proposal failed to achieve consensus after five days. (The modification was added fairly late, though, so I wouldn't be surprised if the modified version passed if it were proposed again.) RSpeer 05:45, Jun 22, 2005 (UTC)

RSpeer, I don't think that I understand your process. What do you mean when you say that the proposal "failed to achieve consensus after five days"? I was under the impression that we had reached consensus to add MMC to the pages of methods passing the MC. I already added MMC to the voting methods page, along with Smith, while subtracting summability. --Hermitage 06:17, 22 Jun 2005 (UTC)
Okay. I counted 2 votes for, F451's vote against, and my weak vote against. In most Wikipedia discussions, that wouldn't be a consensus. (There is no hard and fast rule for consensus, but "unambiguously over two-thirds support" seems to work as a guideline. I'm not sure what the result would be if we had a proposal where two people voted for and one voted against.) It wouldn't be a rejection either, of course; it means that the status quo remains, possibly until the proposal is clarified and proposed again. RSpeer 01:32, Jun 24, 2005 (UTC)
I'd like to add it to the table. - McCart42 (talk) 17:38, 14 July 2005 (UTC)[reply]