Talk:Kemeny–Young method

Archive 1 content
Archive 1 contains the following topics:


 * Example
 * Reply to Henrygb
 * VoteFair ranking moved to Kemeny-Young method
 * Relationship between VoteFair ranking and Kemeny-Young method
 * Cleanup
 * Arrow's Theorem
 * Corrections
 * Wording
 * Table reorganization
 * Characteristic table
 * VoteFair cleanup
 * Reinforcement?
 * Condorcet method?
 * Polynomial runtime
 * Is K-Y NP-complete?
 * Runtime section
 * Some modifications
 * Reliable sources
 * Is Kemeny-Young even in NP?
 * Schulze conflict of interest
 * Characteristics
 * Reversal symmetry

Tally number not correct
The math doesn't work in the example. All rows tally 100 except one. (Anonymous)


 * Awhile back someone vandalized some tally numbers. Now the tally numbers again sum to 100.  VoteFair (talk) 15:12, 27 June 2009 (UTC)
 * Thanks. CRGreathouse (t | c) 21:32, 28 June 2009 (UTC)

Kemeny-Young relationship with Slater ranking?
Kemeny-Young seems to be very similar to another method called Slater ranking. The similarity is described in.

I'm slightly surprised because of the discrepancy between EPFL's intelligent agents course (which only talks about Slater rankings) and Wikipedia (which does not seem to know about them but has abundant details about Kemeny-Young). However, before I invest a lot of time creating and modifying articles, maybe a domain expert could comment on this?

Best, Jonas Wagner 128.178.249.77 (talk) 17:22, 4 November 2009 (UTC)


 * The Slater ranking seems to be much inferior: it's just like K-Y but it throws away the magnitude information. Since computing the Slater ranking is also NP-hard, I don't really see a reason to prefer it!
 * But it would be nice to have an article on it -- though I wonder if it's identical (in the sense that it induces) to one of the Condorcet methods in one of our articles?
 * I grabbed a copy of Slater's 1961 paper and will study it later.
 * CRGreathouse (t | c) 19:04, 4 November 2009 (UTC)


 * I think it's a bit simplistic to say that Slater ranking is inferior. What is better: inversing the order of one pair of candidates with many votes difference, or inversing the order of two pairs of candidates with few votes difference each? In the graph terminology of, is it better to inverse one strong link or two weak links? K-Y prefers the second, Slater the first. gives an example that shows that the two can produce completely opposite result in the sense that the Slater Winner is the Kemeny-Young loser and vice versa.
 * Jonas Wagner (now with Wikipedia account) Sjlver (talk) 10:32, 5 November 2009 (UTC)


 * I can't think of any circumstance in which I would prefer to ignore more votes just because they are less disparate. To my mind removing a few scattered votes is better on both counts.
 * On the other hand, if the Slater ranking could be calculated much more quickly than K-Y, I could certainly see use for it. In fact I have a practical application in mind now... CRGreathouse (t | c) 04:44, 6 November 2009 (UTC)


 * Oh, here's another terminology issue: the "similar candidates" in your link are described in Wikipedia (and, as far as I've read, in the literature) as "clones". CRGreathouse (t | c) 05:45, 6 November 2009 (UTC)

Ties -- in sequence scores
I would be interested to see what happens in the event of a tie, where there is no unique sequence that achieves the highest sequence score. Does Kemeny-Young specify a procedure to resolve such a situation? (Perhaps treat the tying sequences as ballots in a new election, and resolve that with K-Y, or some other method?)

Presumably there are some mathemtical results about the circumstances under which this can happen. A worked example would be ideal. Cheers. Grover cleveland (talk) 23:08, 24 October 2010 (UTC)


 * Here is a simple example that involves a tie in the winning sequence. Suppose the sequence A>B>C>D and the sequence A>C>B>D have the same highest sequence score. Then the winning sequence is A>B=C>D.


 * Cycles that produce the same highest sequence score also can be resolved as producing a tie in the winning sequence.


 * In the first, simple example, a look at the pairwise counts would reveal an exact tie between two choices. (The number of voters who prefer B over C exactly equals the number of voters who prefer C over B.)  In a cyclical case, exact ties may not (directly) appear in the pairwise counts.


 * I've corrected the recently added paragraph to repeat the fact that the winning sequence can involve ties. VoteFair (talk) 17:56, 26 October 2010 (UTC)
 * Thanks. Two questions:
 * What, exactly, is the algorithm for resolving ties in the general case? I guess an obvious one is that each candidate gets the highest ranking it achieves in any winning sequence, but that's just my speculation.
 * Is there any proof that ties can only occur in relatively simple situations like the one you describe? For example, could there ever be a tie between A>B>C>D and C>D>B>A but no other sequences?  I haven't been able to find much if any discussion of this via Google Books, but perhaps you have access to a better source. Cheers. Grover cleveland (talk) 21:25, 26 October 2010 (UTC)


 * Your speculation about one way to identify ties is correct. I've added the details to the article.


 * Ties can arise in very complex ways. For example, a pairwise tie can occur within a cycle among a few choices, and yet another cycle can occur in another part of the overall ranking. And it's probably possible for some few-choice cycles to appear within a cycle involving a larger number of choices. When such ties arise, there are many sequences that share the same highest sequence score.


 * You ask about a "tie" between A>B>C>D and CB>C>D and C<D<B<A having the same highest sequence score, there have to be many other sequences that also share the same sequence score. Although I'm curious about this kind of correlation, I'm not in the academic world, so I have no way to get paid for deriving such a mathematical correlation.


 * Much of my understanding of the calculations arose as I developed the software that is now available at VoteFair.org. That occurred before I found out that Kemeny had come up with what was mathematically equivalent to what I came up with, which I now call VoteFair popularity ranking.  From what I can tell, academic literature about the equivalent methods is very limited regarding ties in the overall ranking. VoteFair (talk) 20:50, 27 October 2010 (UTC)

Output of the Kemeny–Young method is a set of linear orders. Example: 3 voters vote ACBD; 8 voters vote ACDB; 3 voters vote BADC; 6 voters vote CBDA; 6 voters vote DBAC; 4 voters vote DCBA. The linear orders with the best Kemeny scores are ACDB and CDBA. However, there is no partial order PO such that we could say that a linear order LO is a winning order if and only if LO is a completion of PO. If e.g. someone said that PO = (C>D>B) and that LO is a winning order if and only if LO is a completion of PO, then this would imply that also CADB and CDAB were winning orders. By the way: Output of the Schulze method is a partial order PO such that a linear order LO is a winning order if and only if LO is a completion of PO. See sections 2.1, 4.1, and 5 of my paper. Markus Schulze 20:57, 27 October 2010 (UTC)


 * If I understand you both correctly, then VoteFair is saying that there is a procedure to resolve ties (order candidates by the highest rank received in any winning sequence -- this would result in an overall order of A=C > D > B in the example given above). However, Markus Schulze seems to be suggesting that this is not the case (which is consistent with what I have found scanning the literature).  VoteFair -- do you have any reference for the tie-resolving procedure?  If there is no reference, then I propose that the article should simply state that Kemeny-Young can give multiple rankings in the event of a tie.  (Although if it were to be used in real-life elections then obviously some tie-resolving procedure would be needed). Grover cleveland (talk) 22:21, 27 October 2010 (UTC)


 * Yes, the example presented by Markus Schulze is (as far as I know, without time to confirm it) a case in which the method described by Kemeny and Young does not yield an unambiguous result.


 * In such cases, an academic implementation of the "Kemeny" method might be programmed to produce a message stating that the "Kemeny" method does not yield a full ranking sequence -- starting at the level at which such a cycle begins. (The cycle might not affect which choice wins first place, which choice is in second place, etc.) A more useful ("real-life") implementation of the method would resolve the ambiguity, just as Condorcet methods do not give up when there is no Condorcet winner. Of course the VoteFair ranking software at VoteFair.org resolves the ambiguity at whatever levels are affected.


 * As I previously stated, one way to resolve such ambiguity is to keep track of the highest ranking position for each choice when each newly considered sequence produces the same highest sequence score, and then rank the choices at those choice-specific highest levels.


 * Another way to resolve such ambiguity is to keep track of which choices appear at different ranking levels in the different highest-score sequences, and regard all such contiguous (adjacent) "moving-around-one-another" choices to be tied. For example, if the sequences with the highest score are A>B>C>D>E and A>C>D>B>E and A>D>B>C>E, then choices B, C, and D are contiguous choices that move with respect to one another, so the result would be A>B=C=D>E.


 * For the example presented by Markus Schulze, which is entered at VoteFair.org [here], the first ambiguity-resolution method yields A=C>B>D and the second ambiguity-resolution method yields A=B=C=D. Either result can be defended as being correct.


 * There are yet other ways to resolve these ambiguous cases, and they are also defensible.


 * Of course a useful implementation of the "Kemeny" method must choose how to resolve such ambiguity, just as a useful implementation of IRV must choose how to handle a round in which two choices have the same fewest number of votes.


 * Personally I advocate sharing the pairwise counts -- in a tally table -- so that voters can see for themselves how their favorite choice compares with the "winning" choice. This is equivalent to the current approach of sharing plurality counts so that when the results are nearly tied, voters (and media and lawyers) can discuss (and argue) about who really should have won.


 * When I get time to work out a clear wording, I intend to edit the article to clarify the ambiguity that can arise as a result of these tie-related issues. I presume that's needed because this discussion is expressing interest in this topic. (In response to anyone who thinks the issue is "irrelevant" to the "Kemeny method", consider that much of the Condorcet methods article discusses how to resolve cases in which there is no Condorcet winner.) VoteFair (talk) 17:24, 2 November 2010 (UTC)


 * Dear VoteFair, you wrote: "The example presented by Markus Schulze is (as far as I know, without time to confirm it) a case in which the method described by Kemeny and Young does not yield an unambiguous result." Please confirm the result of my example. It doesn't make much sense to discuss different tie-breaking schemes when you cannot confirm the result of this example. Markus Schulze 18:12, 2 November 2010 (UTC)


 * I certainly will analyze the example further when I have time. It involves using software that isn't running on this computer. VoteFair (talk) 18:27, 3 November 2010 (UTC)


 * VoteFair: You say that your software "resolves the ambiguity" in case of a tie. Does it do this by breaking anonymity, neutrality, or both? CRGreathouse (t | c) 21:14, 2 November 2010 (UTC)


 * What do you mean by "anonymity" and "neutrality"?


 * The ambiguity -- in which multiple sequences have the same highest sequence score -- is resolved by using the pairwise counts in the tally table. It does not involve looking at any specific ballots, so that might answer your question about anonymity. VoteFair (talk) 18:27, 3 November 2010 (UTC)


 * See, e.g., May's theorem. An anonymous, neutral, universal social choice function cannot guarantee tiebreaking (for obvious reasons), so if you can break ties in all cases you must give up anonymity (treating some candidates differently), neutrality (treating some candidate differently), or universality (using randomness or restricting allowable vote configurations). CRGreathouse (t | c) 01:54, 4 November 2010 (UTC)


 * Isn't that a misrepresentation of universality (also known as universal domain)? Doesn't universality simply allow each voter to submit any order of preference? That doesn't imply determinism. It has nothing to do with whether ties are resolved using a procedure that involves randomness. Traditionally, a coin flip is used. (I don't recommend simple randomness, however, since it can lead unnecessarily to violations of Independence of Clone Alternatives.) SEppley (talk) 21:52, 12 March 2012 (UTC)


 * I think there is confusion here about the word "tie". Neither the "Kemeny" method nor VoteFair ranking breaks ties. The kind of "tie" being discussed here is when there are two sequences with the same highest "Kemeny"/sequence score. (In the article, to avoid confusion, the word "tie" is not used for this second meaning.) When such a "tie" -- as in multiple highest scores -- occurs, the meaning/interpretation of the matching scores must be resolved. In most cases this resolution amounts to recognizing a tie -- in the usual sense of the word -- in the overall ranking (i.e. two choices are equally preferred). But in some cases, mainly ones that involve circular ambiguity (cycles), there can be uncertainty about which choices are tied, and at what ranking levels those ties occur.


 * In a broader context, remember that the Condorcet-Kemeny method produces a full ranking of all choices in all cases. This contrasts with other methods, which normally identify a winner, and then additional repetitions of the method, or intermediate calculations (in the case of the Condorcet-Shultze method), can be used to identify rankings at lower levels. So, the way in which matching highest scores are interpreted amounts to a third step that is required in cases that have multiple highest-score sequences. VoteFair (talk) 17:22, 4 November 2010 (UTC)


 * Which "other methods" are you talking about? There are plenty of other methods that give a full ranking, such as Copeland's method, are there not? Grover cleveland (talk) 01:37, 5 November 2010 (UTC)


 * Although Copeland's method, Minimax, Nanson's method, Ranked Pairs, Schulze method, IRV, Borda, etc. can produce a full overall ranking, they all start out with a focus on which candidate has some desired characteristic regarding his/her competition with the other candidates. In most cases these methods determine the second-place winner by omitting the first-place winner and repeating the same process (with the remaining data) to find the second-place winner. (If I don't mention it here, Schulze will, so I'll add that the Schulze method develops a full overall ranking sequence as part of identifying the first-place winner; but even this method focuses on identifying which candidate has the desired characteristic, which is related to the "weakest pairwise defeat" numbers.)  By contrast, the Kemeny method looks at sequences of candidates (not specific candidates), and identifies which sequence is a best fit (based on the pairwise numbers).  I used the phrase "other methods" instead of "all other methods" to allow for the possibility of another sequence-oriented method that I don't know about. VoteFair (talk) 19:39, 5 November 2010 (UTC)


 * That's a misrepresentation of Ranked Pairs. Like Kemeny, Ranked Pairs finds a "best" order of finish; it doesn't pick a winner and then start figuring out which alternative finishes second. Ranked Pairs constructs the order of finish a piece at a time by considering the majorities one at a time, from largest majority to smallest majority. The largest majority might, for example, cause the candidate that ends up finishing second to be placed in the (partial) order of finish ahead of the candidate that ends up finishing fourth. An equivalent way to define Ranked Pairs, which more closely resembles the language of Kemeny, is to find the order of finish that minimizes the largest thwarted majority (in the minlexmax sense). Kemeny, on the other hand, finds the order of finish that minimizes the sum of the sizes of all thwarted majorities. (Which makes Kemeny manipulable by nomination of clones. The information about voters' preferences regarding clones is redundant and thus should be irrelevant, but it fools Kemeny's estimation of maximum likelihood.) Although Ranked Pairs sounds much like Kemeny given this brief description, the definition of Ranked Pairs also specifies how ties (majorities that are the same size and/or pairwise ties) are resolved, so that Ranked Pairs always produces a single order of finish. SEppley (talk) 22:37, 12 March 2012 (UTC)


 * Your clarification of the Ranked Pairs method in another article is helpful in understanding this point. I can see your point. I also can see that Ranked Pairs can be derived (originated) in a way that attempts to find the "best" winner and happens to yield a full ranking, which contrasts with the Condorcet-Kemeny method's approach of only yielding a winner who is the top-ranked choice in the sequence with the highest sequence score. VoteFair (talk) 18:16, 13 March 2012 (UTC)


 * It does seem curious that the literature doesn't say more about the situation where Kemeny-Young produces multiple winning sequences. I mean -- if this method is ever to be used in a practical way (e.g. for elections) then this situation has to be addressed.  I understand that VoteFair's adaptation of Kemeny-Young does deal with this, but is this in a referenceable source?   Or are there other referenceable sources on the topic?  Grover cleveland (talk) 06:21, 10 November 2010 (UTC)


 * I imagine the literature does address it, but it isn't an issue in real elections, where the possibility of such a tie is tiny -- just calculate the odds of any tie with, say, 1000 voters. CRGreathouse (t | c) 14:40, 10 November 2010 (UTC)


 * Yes, multiple highest sequence scores are not "commonplace," yet they do arise often enough that any real implementation must either resolve them (as ties in the output ranking) or generate a message that the issue has arisen and is not being resolved. In the VoteFair American Idol polls I host, multiple highest sequence scores do happen fairly often -- especially because the results are updated as voters vote in the poll, not just at the end of the polling time.


 * I'm not surprised that they are not discussed in academic literature because such efforts usually focus on the bigger problem of figuring out how to do Kemeny calculations quickly -- which I've already figured out how to do. When the right opportunity arises I'll be happy to share what I've learned, but at the moment it's a chicken-and-egg problem of me not being in the academic world, so I don't have a channel through which I can share my understanding.  Here in Wikipedia is as close as it comes to providing such a channel. VoteFair (talk) 19:04, 10 November 2010 (UTC)


 * Why is everyone assuming that the literature doesn't discuss ties?
 * As for your method of calculating Kemeny scores, VoteFair, I don't know how efficient it is compared to the best published method. I imagine the best published method is asymptotically faster than yours, but since I don't know how you're calculating I can't be sure in that assertion. CRGreathouse (t | c) 20:27, 10 November 2010 (UTC)


 * Followup: I don't have time at the moment to look into specifics, but I see that Kemeny-Snell 1962 mentions ties in the context of Kemeny's voting method (here, the "Kemeny–Young method") on pages 9, 12, 13, .... So ties are certainly "in the literature" on K-Y. CRGreathouse (t | c) 03:35, 11 November 2010 (UTC)


 * Second followup: Mathematics without numbers does discuss ties in (what you call) sequence scores on pp. 589-590. The discussion finishes:
 * We conclude that the requirement of a unique social ordering is too restrictive. It suffices that we should arrive at a unique ordering "in most cases."
 * which nicely parallels my earlier statement. :D CRGreathouse (t | c) 16:19, 11 November 2010 (UTC)


 * How about this (pure original research, for which I apologize): If application of K-Y is ambiguous (i.e. there is more than one ranking with the highest score), then consider all rankings in which exactly 1 pair of candidates is tied.  There will be n!.(n-1)/2 such possible rankings.  In calculating the scores, preferences between candidates tied with each other are ignored.  If this yields a unique winnng ranking, then that becomes the K-Y winner.  If not, then consider all rankings in which exactly 2 pairs of candidates are tied, and so on.  If all else fails, then the K-Y winning ranking will tie all the candidates.  I've tried this out with a few simple cases, and it seems to produce what is intuitively the right result.  (Note:  I'm not in a million years suggesting adding this to the article, just sharing :).  Grover cleveland (talk) 19:53, 22 March 2011 (UTC)
 * That's somewhat similar to the use of top cycles. CRGreathouse (t | c) 23:24, 22 March 2011 (UTC)
 * What does your tie-breaker do in the example above (3 voters vote ACBD; 8 voters vote ACDB; 3 voters vote BADC; 6 voters vote CBDA; 6 voters vote DBAC; 4 voters vote DCBA)? Markus Schulze 09:34, 23 March 2011 (UTC)
 * As explained above, there is more than one way to resolve tied-sequence cases. Earlier I added info to the article about how tied sequences can be resolved, and that was removed, so it seems we have wandered off the topic of what can be included in the article.  VoteFair (talk) 01:06, 6 May 2011 (UTC)

Reinforcement
The current description of reinforcement seems unclear to me. I much prefer the previous description, which was also sourced. Why was this change made? Grover cleveland (talk) 22:17, 27 October 2010 (UTC)


 * Explanations in an encyclopedia are intended to be much clearer than what is written in academic journals, and the current explanation is much clearer than the previous version. For example, the previous version was incomplete because it did not say that the two results had to be from the same election/contest.  As for sourced, the source reference remains.  You say the current version is unclear.  What is unclear? VoteFair (talk) 16:30, 7 November 2010 (UTC)

Even better, we should create an article for the voting principle called "reinforcement." That would enable a rigorous mathematical definition to be presented here in Wikipedia -- for those who want that level of detail. An unrelated article named "reinforcement" already exists, so the new article's name needs to be qualified somehow. I'm not familiar enough with Wikipedia's naming rules about such an overlap, so I'll have to let someone else create the stub for such an article. VoteFair (talk) 18:32, 11 November 2010 (UTC)

Peyton Young defines Reinforcement in his book Equity: In Theory and Practice. Reinforcement says that if two collections of votes produce the same order of finish, then the combined collection must produce that same order of finish. It's weaker than the Consistency criterion, which says that if two collections of votes produce the same winner, then the combined collection must produce that same winner. Neither Reinforcement nor Consistency are important, in my opinion, since the rules can easily prevent a minority from having the power to partition the voters; thus a minority won't be able to exploit scenarios where these criteria are not satisfied. Contrast that with the ease with which a tiny minority can exploit scenarios where the Independence of Clone Alternatives criterion (ICA) is not satisfied; this means ICA is much more important. SEppley (talk) 16:50, 12 March 2012 (UTC)


 * Partitioning of the ballots is a useful way to describe the criteria, but it's not (necessarily) what's relevant in elections. Consider that in a U.S. Presidential election, if the same ranking occurs in every one of the 50 states, then it is advantageous that the overall ranking matches the unanimous result.  In that case, nobody partitions the ballots. VoteFair (talk) 18:33, 12 March 2012 (UTC)


 * Advantageous for whom? Can you provide an example that illustrates why the society would be better off? (Feel free to simplify the example; 50 states shouldn't be needed.) It needs to show why this is really advantageous to society rather than some insignificant aesthetic notion of consistency. (There are a slew of such aesthetic notions of consistency and no voting method can satisfy all of them. Advantage on one of these criteria can be cancelled by disadvantage on another, so you'd need to demonstrate a net advantage.) SEppley (talk) 19:48, 12 March 2012 (UTC)


 * In single-winner cases there is no advantage to society. I'm thinking of future (much-more-advanced) voting situations in which the ranking of candidates (beyond the first-most-popular choice) affects an election/decision outcome. VoteFair (talk) 18:21, 13 March 2012 (UTC)


 * Please be clearer about how the order of finish will be used in those situations, and the advantage that satisfaction of Reinforcement would allegedly provide. SEppley (talk) 17:23, 15 March 2012 (UTC)


 * Note that this article does not claim any relative level of importance for the Reinforcement criteria, so this question and my reply are not consequential to editing this article. Having made that disclaimer, I'll point to the software negotiation tool at NegotiationTool.com as an example of an advanced method of voting, which in this case improves the fairness of decisions made by a group of decision-makers (such as in a legislative body). VoteFair (talk) 17:18, 16 March 2012 (UTC)

Algorithm
It might be interesting to have some discussion of how the K-Y winning sequence could be calculated efficiently. The brute force method, simply considering every possible sequence, would presumably have a complexity of O(n!) which isn't great. I presume that there are some algorithms that substantially improve upon it. Grover cleveland (talk) 02:09, 26 February 2011 (UTC)
 * The calculation of the Kemeny-Young ranking or the Kemeny-Young winner is NP-hard. Markus Schulze 04:22, 26 February 2011 (UTC)
 * That doesn't mean there aren't algorithms which dramatically improve on the naive one! CRGreathouse (t | c) 19:47, 27 February 2011 (UTC)
 * Grover cleveland is correct; there are ways to substantially speed up the calculations. The surveys/polls/elections at VoteFair.org use such faster calculation methods. I'm working on steps that will lead to sharing how that's done. Note that the archives already contain earlier discussions about this topic. VoteFair (talk) 20:24, 27 February 2011 (UTC)

"Sequence" vs. "Order" vs. "Ranking"
I recommend that, to stay consistent with the terminology at Wikipedia, the term "sequence" should be replaced by "order" throughout the article. Markus Schulze 22:55, 27 February 2011 (UTC)


 * Agreed. CRGreathouse (t | c) 00:46, 28 February 2011 (UTC)


 * I strongly disagree. A "sequence" is a kind of "order," but an "order" is not necessarily a specific (kind of) "sequence."  Also, the word "order" -- in U.S. English at least -- has other meanings that are not similar to the word sequence.  (See [].)  Regarding consistency, what Wikipedia articles are you wanting to be consistent with?  Even the article linked to above (order) is titled "total order," not simply "order".  (The need for the word "total" clarifies that the word "order" alone is not unambiguous.)  VoteFair (talk) 05:10, 1 March 2011 (UTC)


 * "Order" is what is used in the literature, with different forms based on what problem is being addressed (ties or no ties).
 * There's a 1-1 correspondence between sequences and total orders. But since we discuss ties we may be in the case of weak orders instead, which sequences cannot handle.
 * It may be worth mentioning at this point that I'm an editor for an encyclopedia devoted entirely to sequences, so I'm rather familiar with the terminology.
 * CRGreathouse (t | c) 07:31, 1 March 2011 (UTC)


 * Maybe, we could replace the term "sequence" with the term "ranking". This would still be in accordance with Wikipedia's terminology. Markus Schulze 15:14, 1 March 2011 (UTC)


 * The word "ranking" is much better than the word "order." The word "ranking(s)" -- sometimes preceded with the word "possible" -- would fit in many of the places where the word "sequence" now appears.  However, there are many places that now refer to "sequence scores" and I don't know if "ranking scores" is the best choice.  I'm open to suggestions -- except that I oppose calling the scores Kemeny scores because Kemeny described scores that measure opposition instead of support.  My goal is to achieve clarity for all readers, not just mathematicians.  What do you think CRGreathouse?  I continue to have a high regard for your sharp and unbiased thoughts. VoteFair (talk) 19:23, 1 March 2011 (UTC)


 * Ranking would be fine. Order allows more precision, but if you think that will cause the nontechnical readership problems we can go with ranking instead. CRGreathouse (t | c) 05:18, 2 March 2011 (UTC)


 * As requested, I changed the article's wording from "sequence" to "ranking". VoteFair (talk) 18:20, 25 March 2011 (UTC)