Talk:Later-no-harm criterion

From Wikipedia, the free encyclopedia

Articles for deletion This article was nominated for deletion on March 8, 2007. The result of the discussion was keep.

Find sources: Later-no-harm criterion – news, books, scholar

Contents

[edit] Definition of ordering candidates preferredness

I think the following needs clarification: "less preferred candidate cannot cause a more preferred candidate to lose". What is the definition of preferredness of one candidate versus another? Can this be defined for all voting systems? For example, in range voting how do you order the preferredness of the candidates? It would be nice if less and more preferredness were links that give formal definitions. WilliamKF 16:28, 13 March 2007 (UTC)

Preferred has exactly the same meaning as in standard English. If candidate A is preferred to candidate B he/she is liked more than candidate B. In range voting if A has a rating of 100 and B a rating of 75 A is preferred to B. Dgamble997 16:08, 4 September 2007 (UTC)

[edit] Plurality and Later-no-harm

Tom I’ve undone your inclusion of Plurality in the list of methods that comply with later-no-harm. In the case of a single seat a voter cannot express a choice for a less preferred candidate so the criterion does not really apply. In the case of Plurality-at-large (with 2 or more seats being filled) Plurality actually fails later-no-harm.

Take the example:

1 A>C

319 A>B

310 C>D

185 E>C

185 E>D

With 5 candidates contesting 2 seats.

If the E voters use both their votes the winners are C and D.

A 320

B 319

C 496

D 495

E 370

If the E voters cast only one vote the winners are now A and E.

A 320

B 319

C 311

D 310

E 370

By voting for second choice, less preferred, candidates the E voters have caused their most preferred candidate to be defeated. Dgamble997 16:08, 4 September 2007 (UTC)

Later-no-harm criterion and Monotonicity criterion are the same. merge? —Preceding unsigned comment added by 72.155.188.56 (talk) 21:37, 11 January 2008 (UTC)

No, they are not the same. IRV fails monotonicity, but usually satisfies LNH. --Abd (talk) 03:02, 20 January 2008 (UTC)

[edit] IRV fails LNH if majority required

Apparently this has escaped notice, for the most part, but we can see from the IRV legislation proposed for Vermont by Terrill Bouricius when he had a seat in the Legislature, that there are conditions where adding an additional vote at a lower rank can harm the favorite. Suppose there are three candidates on the ballot, A, B, and C, and the voter votes for A alone. No candidate gains a majority; C is the runner-up to B, and B is one vote short of a majority. Under the Vermont constitution, the election fails and goes to the legislature for resolution by secret ballot among the top three. Thus A might win. If, however, the voter adds a vote for B, B gains a majority and the election is resolved, A cannot win. Thus LNH is violated. If the election is decided by plurality (which is the norm with IRV implementations), LNH is satisfied. --Abd (talk) 03:02, 20 January 2008 (UTC)

I'm sure you're very proud of your technicality, but I can't agree. I see IRV/STV as a black-box system of determining one or more winners via an elimination/transfer process. If no winners are picked, AND all elimiations are voided, it isn't IRV/STV any more. It's nothing at all in fact and the EM criterion no longer applies.
Admittingly it does sound insane to me - to allow a ballot majority failure (possibly caused by a single voter failing to rank among the final two as you show) to override all votes. Tom Ruen (talk) 06:17, 20 January 2008 (UTC)
Tom, the remark about my pride is uncivil, stop it. As to the substance, thanks for raising the issues. There is the black box system you describe, but it would be more accurate to say that there are a series of systems with varying characteristics and procedures, and it is the entire system, in context, that determines the true behavior of a system. Details count. By what authority would we claim that the Vermont system is not "IRV"? As you know or should know, the term "Instant Runoff Voting" was coined in 1996 in email conversation between Rob Richie and another, whose name is mentioned in IRV Talk by Markus Schulze. It was a political invention, not a term of art in the field of election methods, and FairVote has applied the term to just about any system using bottom elimination rounds, including two-round batch elimination, and when I pointed out that Robert's Rules doesn't recommend "IRV," but, rather, "preferential voting", and that the default under Robert's Rules would be that an absolute majority is required, and thus this was different from "Instant Runoff Voting" as it was then described by the article, I took a lot of flack for it. Yet this is precisely the case here, and it seems that Tom is popping up on the other side. As with Florida 2000, it seems it depends on whose Gore is being axed. Am I wrong?
Criteria are typically designed for single-ballot, deterministic systems; however, they may also be applied (and are applied) to multiround systems, considering them as black boxes, with the electorate being defined as the entire electorate that votes in the final poll. Indeed, "deterministic" is itself a criterion, and what can be said about IRV with an absolute majority requirement is that it is not deterministic. It still either satisfies or does not satisfy other criteria. It doesn't satisfy LNH, if an absolute majority is required, whereas it does satisfy the Majority criterion, for example.
Actually, while it is irrelevant here, my comment about the composition of the electorate, considering two-round elections as a black box, is incorrect. Given that I've stated this many times, and nobody has pointed out the error, I'll forgive myself! This is a problem that I'm not going to solve instantly, and, indeed, it may be most appropriate to say that some election criteria don't apply to two-round. However, that's not the case here, if we look at the IRV election as having a determination of, say, the top three, as the outcome, as in the case of Vermont, or of no determination at all, election failure, as in the case of Roberts Rules default.--Abd (talk) 20:43, 20 January 2008 (UTC)
Now, as to the alleged insanity of the situation Tom points to. That is indeed one possibility. I wrote the votes as truncated, because truncated votes are *common* in IRV. Voters vote truncated ballots because they have a strong preference , or, alternatively, they don't know enough to judge between the other candidates. I now have studies of the behavior of IRV in actual elections, and can compare this behavior to top-two runoff, and the difference is striking. I'm astonished, in fact, if it's true that nobody has noticed this behavior before. (It is hinted at in the argument against IRV that top-two runoff allows voters another chance to compare two specific candidates, and also in the argument that IRV not only does not guarantee a majority -- no method can, except by limiting the options to voters -- but it can miss a majority winner, a winner preferred by two out of three voters to the IRV plurality winner. *Requiring an absolute majority detects this condition, almost certainly.*
What I've discovered, simply by looking at the results of top-two runoff in San Francisco and Cary, NC, before there was IRV, is that all elections were either uncontested, won in the first round by a majority, or won by one of the top two in the first round, and two-thirds of the elections were won by the plurality winner of the first round. One-third of the elections that went to a runoff reversed the first-round result. (Yet to be done is a broader study of top-two runoff, but the sample size here is enough to justify the one-third figure as a general estimate.) However, after IRV, *all* elections have been won by the plurality winner of the first round.
It might be claimed that top-two runoff is unfair, because of the inconvenience to voters of a special election; however, the Cary situation actually turns this on its head. The first election is called a "primary," and this is also the name in San Francisco. In San Francisco, indeed, the runoff is scheduled in December, after the primary with the November elections. But in Cary, the *primary* is scheduled in *October,* with the runoff, if needed, held as part of the general election in November, so the runoff may be *more* convenient for some voters; turnout was roughly the same for the primary and for the runoff.
The conclusion is inescapable: in actual practice, IRV is defeating the purpose of requiring a majority, and the need for runoff was only there because of a desire that a majority be obtained. *In every IRV election, the plurality winner in the first round won in the end.*
Now, as to the example Ruen notes: I used truncation in the example. I could have used a fully ranked ballot with the same result. It was not truncation which caused the result, it was the failure of a majority to vote for one of the top two. Instead of truncation, let's assume that it was a fourth candidate that the voter voted for. There is a consistent attempt on the part of IRV advocates to blame voters for majority failure. If only those pesky voters would agree to vote for one of the top two, there wouldn't be any problem. Doesn't this sound familiar? Isn't that the argument for Plurality? In San Francisco, FairVote certainly calls what they have "IRV," but it is actually a three-rank limited system, and exhausted ballots are unavoidable, (with 22 candidates on the ballot!).
--Abd (talk) 17:02, 20 January 2008 (UTC)
I've argued in favor of a top-two primary system as superior to IRV on the grounds that it's less work for voters since they don't need to rank, no conditional preferences. The counter argument is that very few people vote in the primary so it's better to allow more choices in the general election. It's a subjective question what's better.
On majority failure it is a nonissues for me. You can never FORCE voters to rank (or vote in a runoff round they don't care about), so there's no guaranteed majority in any voting system.
I'm not overly impressed by the majority failure arguments. For anyone paying attention single-winner political elections will make it pretty clear which 2-3 candidates are strong, and if voters don't rank any of them, to me this suggests a protest vote - they don't want their vote transfered to any of the leaders. To allow such protest voters to block a winner from being determined in IRV or any election method to me is just dumb.
If you really want to block winners, you should use a NOTA explicit vote option to signal this choice. Even THEN I'd let the plurality winner stand. Only when the NOTA vote BEATS the plurality winner should we care to pick no winners. Tom Ruen (talk) 20:33, 20 January 2008 (UTC)

It's not an argument, it's a fact, and underneath the fact is a basic principle of democracy: that all decisions be made by a majority of those voting. Under Robert's Rules, which describe the default rules of most democratic organizations, meeting as peers, no decision can be made except with the assent of a majority of those voting. "Majority failure" means that no clear democratic basis for the decision exists. Now, for convenience, many organizations, including governmental ones, have decided to make the election of officers an exception to this, and it is widely recognized that this is problematic, hence the requirement for an absolute majority in many jurisdictions. At the least, top-two runoff guarantees a majority result, though with an obvious defect: just like IRV, it can fail to detect a majority winner with lack of "core support," so this candidate is eliminated in the first round. However, that situation is actually relatively rare. Exhaustion isn't rare, it is *necessary* in elections with more candidates than ranks (technically, with N candidates, there must be N-1 ranks expressable).

Ruen is focusing on majority failure due to ballot exhaustion from "protest" votes, i.e., voters who actively refuse to rank additional candidates. What he has not addressed is failure from *necessary* ballot exhaustion due to a voter voting sincerely. Further, there is the simple failure from voters who have no opinion between two candidates who turn out to be the top two. Because the race was "diluted" if it had many candidates, the situation is very different than if there were only two. And top-two runoff has the virtue of causing that focus to take place. Clearly, it does have a function. It is statistically unlikely, from the results I've found, that the voters would have chosen, in every case, in a real top-two runoff, the candidate who won under IRV.

Ruen hints at one reason: The counter argument is that very few people vote in the primary so it's better to allow more choices in the general election. and You can never FORCE voters to rank (or vote in a runoff round they don't care about), so there's no guaranteed majority in any voting system.

First of all, Ruen seems to have overlooked that the San Francisco situation was reversed: November is the primary (not partisan), and December was the runoff, if needed. In Cary, the situation was reversed: the primary was in October and the runoff as part of the general election. Complicating this in Cary is that the city elections are in off-years, i.e., *not* when the big elections are being held, so turnout is low for *both* elections.

However, one point has been missed, at least I've never seen it mentioned: holding elections at an inconvenient time causes a kind of sincere Range voting to take place. Essentially, weak preferences are devalued through differential turnout. This, I expect, a deep study would show that inconvenient elections, where the inconvenience is not so serious as to actually warp access, improve overall satisfaction with results. That is, a lot of voters who don't really care that much may not favor the result, but voters who do care prefer it.

Ruen wrote: To allow such protest voters to block a winner from being determined in IRV or any election method to me is just dumb. This is calling a basic principle of democracy "dumb." Wouldn't be the first time.... The majority requirement -- it is the default in Roberts Rules, and election by a plurality is strongly discouraged -- precisely allows voters to stop an election from succeeding, if a majority don't vote for the plurality winner. "NOTA" is not an appropriate answer, though it would extend this power (and I support it for that reason). Again, Ruen clearly hasn't followed prior discussion of this point. The standard for "absolute majority" is that all ballots containing a vote for an eligible candidate are counted as the basis for the majority. Is it a "protest" vote if the voter votes sincerely for a favorite, and, say, detests both frontrunners? Under Robert's Rules, there is an election de novo if there is majority failure, there are no eliminations, and new candidates may be nominated. And in public top-two elections, there is still a write-in option, providing a -- mostly theoretical -- out. However, NOTA would allow voters who actively don't want the election to succeed to vote to make themselves part of the basis for majority, without giving a vote to any of the explicit candidates, thus providing *increased* power to do what Ruen just claimed he thought was stupid.

And, finally, he wrote: For anyone paying attention single-winner political elections will make it pretty clear which 2-3 candidates are strong, and if voters don't rank any of them, to me this suggests a protest vote.... Again, Ruen isn't paying attention. In San Francisco, one election had 22 candidates on the ballot, and most had far more than the available ranks. After all transfers were completed, the winner among the 22 had 33% of the ballots containing a vote for that candidate. Further, one of the major selling points of IRV has been that voters may vote sincerely. Ruen now wants them to vote tactically, and thinks that they are not "paying attention" if they don't. I.e., if they prefer three candidates who are not the top two, they should vote insincerely in order to participate in the real election. Of course, this exact argument is used to justify Plurality voting: voters who are paying attention know who the top two are, and so will -- except as a protest vote -- vote for one of them. (The reward for protest voting is George W. Bush -- or, in another year, Clinton.) Voting for the third candidate isn't enough, it leads to ballot exhaustion for the last round.

Ruen has again and again complained about how much I write. However, he might learn something if he reads more. He is capable of independent thinking, he is nobody's puppet, but, too often, he hasn't made the effort.

Now, this is not much about Later-no-harm, except, once again, to underscore that IRV with an absolute majority requirement, as was the case with the IRV legislation Terrill Bouricius introduced in Vermont (and as with other later introductions by others), fails LNH, even though the legislation actually had the ballot claiming that it satisfied it. Something like: "Your vote for a lower ranked candidate cannot harm your first choice." Obviously, some people simply assumed that the general characteristics of single-winner STV -- with absolute majority not required -- would apply, and, beyond noting in the legislation how the election would pass to the legislature if there was a majority failure, the ballot instruction did not reflect the actual situation. Perhaps if the legislation had faced serious discussion, it would have come out. And perhaps not. --Abd (talk) 21:30, 20 January 2008 (UTC)

[edit] unless election requires a majority of ballots cast

This statement needs work:

Only Instant-runoff voting and Single transferable vote satisfy the later-no-harm criterion, unless election requires a majority of ballots cast.

The unless phrase was added by Abd in January: [[1]] and it is completely inexplicable as written, AND as best I can judge, it's a figment of Abd's over-excited imagination of technicalities which exist nowhere and don't deserve this confusion.

If I suggest it be removed, Abd is going to write 72 paragraphs re-explaining his complaint why it shouldn't be removed. I have no compromise idea how it can be included without being uselessly confusing.

Maybe we can agree for me to add as equally true or unless the moon is made of green cheese. to help clarify the nature of the first exception?

I suppose I should just call for a vote for deletion of this entire article, since it appears to a non-IRV-advocate as a criterion created for the specific POV to make STV/IRV look good and everything else look bad. Then we'd all be happy, right?! Tom Ruen (talk) 22:14, 6 February 2008 (UTC)

I added an unreferenced generalized explanation of Abd's technical exception"

NOTE: IRV and STV disallow an eliminated candidate a chance to win, however if a special rule was added that could reverse an elimination, then it is possible for lower choices to affect the election and the Later-no-harm criterion to fail.[citation needed]

Abd's majority exception isn't the critical fact to the issue, but a reversed elimination. If eliminations can be reversed for any reason, then a lower choice could change the result, possibly against the interests of the vote, in the same wa

s...You can skip this explanation unless you are curious)...Vermont's constitution states that if there is "no election", meaning no candidate receives the "major part of the votes" then the legislature must select one of the three candidates who received the most votes as governor. It is a principle of constitutional law that a law that makes a constitutional provision superfluous is constitutionaly suspect, so rather than crafting a bill that allowed IRV to always elect a "majority winner" I needed to preserve the POSSIBILITY that the legislature might still fulfill that function. Thus, I chose to define the phrase "the major part of the votes" to mean more than 50% of ballots cast for the race, including any exhausted ballots. Thus if no majority (by this definition) was achieved the legislature was allowed to step in and elect any one of the three top vote getters. But here is the key...that section was removed by the Senate and House, and the bill passed without it (later vetoed by the governor any way). So, this version of implementation was a single person's idea (mine), that was rejected, which hardly makes it an exception worthy of special mention in Wikipedia. if such a provision were to actually to be implemented somewhere, then it MIGHT be worth mentioning.

Tbouricius (talk) 14:24, 4 June 2008 (UTC)
Thank you for the information, Terry. I will remove the note. Progressnerd (talk) 01:32, 7 June 2008 (UTC)
I'm happy. Tom Ruen (talk) 04:22, 7 June 2008 (UTC)

Glad you are, Tom, but, unfortunately, like about everything else connected with Wikipedia, happiness is transient. Isn't that one of the Four Noble Truths? The text as I had it was simple, and accurate.

Here is the problem, folks. IRV is commonly failing to find a majority of ballots cast, whenever runoff takes place. I see seven out of nine actual IRV runoffs in the U.S. that did this. The Vermont constitution required a majority. The "last round majority" that IRV finds isn't a majority by any standard definition. So, indeed, the provision Bouricius had put in his legislation was removed. And that would have set up conditions for a constitutional challenged to the legislation. That legislation was quite remarkable for a preamble that clearly set up the desirability of a majority result, which it them proceeded to disregard. Robert's Rules, invoked so many times for allegedly "recommending" IRV, clearly requires a majority, it's blatant. So what is this "rare" method? Does anyone have statistics on frequency? What Progressnerd removed was a weird warping of a plain and clear statement. If a majority of ballots cast is required to elect, IRV does not satisfy Later-no-harm, period. This is an election criterion, and it's an absolute, i.e., if there is even one exception, the method fails the criterion.

Later-no-harm is incompatible with majority election requirements. Now, this is an article on a voting system criterion. It's not about legislatures and actual implementations; I've seen this argument advanced by IRV advocates many times, because actual implementation is what they are working for, and their strongest argument against what might be considered better systems is that they aren't being used. But you can hardly say that majority election requirements are rare, they are common. They exist, quite naturally, in connection with any top-two runoff system. (There is only one way to avoid possible majority failure with IRV -- or any preferential voting system -- and that is to require full ranking. Which is essentially to require voters to vote for a candidate whom they might detest, simply because there was a worse one in the race.) --Abd (talk) 05:58, 7 June 2008 (UTC)

Abd, it sounds like you're fighting other battles here. If a "majority of ballots cast is required" then it ceases to be IRV. (That is not what is meant when advocates claim IRV elects "majority winners", but that's not relevant here) The note it still purely about a hypothetical system. We should not scour the earth for every system ever proposed and judge it against this criterion. Progressnerd (talk) 01:08, 8 June 2008 (UTC)
What battles? Progressnerd, you appear to be a single-purpose account with a special interest in Instant-runoff voting. This isn't an article on instant-runoff. It's about a voting system criterion, and voting systems criteria are generally intended to apply to all methods, whether they are "hypothetical" or not. Majority requirements aren't hypothetical, though, they are fairly common, and, in fact, the default under Robert's Rules is that a majority is required, and when Robert's Rules describes "preferential voting," which IRV activists have for years cheerfully claimed is a description of "instant-runoff voting," they are describing a system which has a majority requirement. Whether or not this is what organizations are actually implementing is something that I don't have statistics on. Do you? But it's clearly what they describe, because they note that incomplete ranking can result in no candidate getting a majority of votes, and hence the election would have to be repeated.
Instant-runoff voting is a neologism, promoted by FairVote, and it's questionable whether this article should even mention it as such. For starters, it's not a well-defined method, it is a family of methods. So I'm going to take it out of this article, leaving Single transferable vote with the note about majority requirements. Thus it covers Preferential voting as described in Robert's Rules of Order. (Later-no-harm, if there is no majority requirement, is satisfied by Single transferable vote because lower preference votes are only counted if a candidate is eliminated, and this elimination is precisely what causes STV to fail the Condorcet criterion.)
Much of the theoretical work on Single transferable vote has been done with fully-ranked ballot versions of it, and majority failure cannot occur with a fully-ranked ballot. So, in fact, there is an error in what I'd put in, and I should fix it. Later-no-harm is satisfied by STV, if the ballots are fully-ranked; on the other hand, Robert's Rules and parliamentary authorities in general would stand in horror at creating a majority by forcing voters to essentially vote, as a contingency, for every candidate except for one. I think Australia is peculiar in requiring full ranking; they see a lot of Donkey voting as a result, and some provinces have implemented Optional preferential voting; and what are they seeing? A lot of plumping, voting for one only, and, of course, majority failure. But they don't require a majority, so later-no-harm still holds. Can of worms, Progressnerd, you appear to have attached yourself to. IRV is being proposed and implemented in the U.S. in jurisdictions where a majority has been required, and that has been dealt with by implying that IRV finds majorities while at the same time the implementation legislation removes the majority requirement. In no campaign I've seen has this been explicitly examined. IRV is sold as a way to find majorities without expensive runoff elections, but, in fact, it isn't finding majorities most of the time that it goes into instant-runoff rounds (it found a majority two times out of nine such runoffs in the U.S.)
But that's really dicta here. This is an article about a voting system criterion. If a majority requirement (explicit in Robert's Rules of Order) is not notable, because "hypothetical," why then do you swallow the camel of "Minimax Condorcet and "Descending Solid Coalitions"? To my knowledge, those are "purely hypothetical." --Abd (talk) 04:05, 8 June 2008 (UTC)
To be considered, the voting system should surpass a certain threshold of intellectual currency. Here's a voting system: rank the candidates, if none receives a majority of first choices, eliminate the candidate with the *most* first choice votes. Repeat. There now that the system has been presented, it should, by your own logic, be judged against every voting system criterion on Wikipedia. Reductio ad absurdum.
Some other battles you appear to fighting include: "is IRV a neologism?"; "what is the proper definition of 'majority winner'?"; "what system does Robert Rules recommend"; etc. Those debates are largely irrelevant to this article, and I suggest you find some place else to grind your axes. The system you are judging against LNH here is not sufficiently established -- either in voting literature or voting practice -- to be discussed.
The removal of Instant Runoff Voting from the list is also inappropriate. The algorithm for counting the votes on the Instant Runoff Voting page satisfies LNH. Progressnerd (talk) 05:13, 9 June 2008 (UTC)

[edit] Application to approval voting is dodgy

The application of LNH to AV seems dodgy to me. LNH is very much about preferences - if you prefer A to B and C, and A is the winner, then how you rank B and C should not affect the election of A. AV doesn't allow you to express preferences: you either approve of someone or you don't, so the only preference that can be expressed is between all the candidates you approve of, who you approve of equally, and those you don't, who you disapprove of equally. You may well have more nuanced preferences, but AV doesn't allow you to express them. To apply LNH to AV is thus meaningless, because there is no expressed preference order to consider. You can certainly criticise AV for not allowing you to express preferences, but that's a different matter.

Going back to Woodall's definition of LNH:

"Adding a later preference to a ballot should not harm any candidate already listed."

AV doesn't have a 'later preference', so you simply cannot apply this criterion.

Oh the other hand, if there's some authoritative external source which applies the criterion to AV, then fine - and we should cite it.

-- Tom Anderson 2008-04-16 1405 +0100 —Preceding unsigned comment added by 62.56.51.133 (talk) 13:05, 16 April 2008 (UTC)

No, Approval fails L-N-H in both the technical meaning and in common understanding. If I prefer A>B>C and I first mark my ballot for A, if I then add a mark for B I am adding an additional later preference, but risking aiding the defeat of my most preferred choice, A. The only way to get around the common-sense understanding is to pretend that voters don't actually have any preferences in their heads, but only what appears on the ballot (which with Approval is crudely unable to distingusih a top preference from later preferences). It is most useful to structure criterion so that they can be used to compare voting methods, so it is ill-advised to ignore voters' actual preferences and say that because the voting method does not allow for expression of ranked preferences that the voter must not have any such preferences. The ballot must be seen as the informational transmission device, rather than as the ORIGIAL SOURCE for preferences.
Tbouricius (talk) 14:35, 4 June 2008 (UTC)
Bouricius is correct on this point. We might easily assume a process: the voter first votes for the favorite, because, in Approval, there is never a reason to not do so. Then the voter thinks, "Should I add another candidate?" If the voter fears that the second candidate will, through the additional vote, beat the favorite, and this is a major preference, the voter may well abstain from it. The situation is more complicated in Bucklin voting, for, with Bucklin, the voter has already expressed the preference, but the election did not gain a majority. Should the voter participate in further process? Indeed, the second preference can again cause the first preference to lose to the second, and it is quite like Approval except for that clear first preference expression. But compromise is essential to collective action and cooperation. Many voters will, indeed, add additional preferences, and this was historically true with Bucklin, see Brown v. Smallwood. --Abd (talk) 19:07, 8 June 2008 (UTC)

[edit] Brown v. Smallwood

I have read Brown v. Smallwood. I don't see that Bucklin voting has been "declared unconstitutional partly because it violates later-no-harm". I rather have the impression that the members of the Minnesota Supreme Court didn't understand why, when a preferential single-winner election method is being used, a voter cannot give his first preference, his second preference, his third preference etc. to the same candidate. ("We do right in upholding the right of the citizen to cast a vote for the candidate of his choice unimpaired by second or additional choice votes cast by other voters.") Because of this reason, the Minnesota Supreme Court declared all preferential single-winner election methods (and not only Bucklin voting) unconstitutional.

Is there any reliable source that claims that Bucklin voting has been declared unconstitutional for violating later-no-harm?

If there is no reliable source, then this section should be removed from this article. Markus Schulze 21:00, 8 June 2008 (UTC)

(response by Abd (talk) begins here, not indented) I considered taking it out, because the claim is weak. Brown v. Smallwood does mention what can be interpreted, rather easily, as a "later-no-harm" problem. I did remove the "other voters" reference because this clearly didn't have anything to do with Later-no-harm. Indeed, the general thinking in BvS seems to be that any kind of alternative vote is unconstitutional, a conclusion which was not confirmed by other courts.

One complication is that "preferential voting" was the name given to Bucklin at the time. But that term was used elsewhere to refer to STV methods. Was the court outlawing all preferential voting? Probably. The later-no-harm reference was probably dicta. This is my analysis of the decision, which Warren Smith, of the Center for Range Voting adapted and put up. It includes a link to a pdf of the decision itself.

The interpretation that the reason for the decision was Later-no-harm appears to have been created and promoted by FairVote, and there is a clear motivation for that: if the problem was LNH, then perhaps IRV is not unconstitutional in Minnesota. But an attorney for the City of Minneapolis was asked to opine on the topic, and he wrote this, which considers that Brown v. Smallwood applies to IRV as well as to Bucklin.

Contrary opinion has been issued by Tony Anderson Solgård and Paul Landskroener, in a paper published in Bench and Bar of Minnesota,[2]. Solgård is chair of the board of FairVote Minnesota. This isn't exactly a neutral legal opinion! It's advocacy. Solgard and Landskroener wrote:

In the Bucklin vote-counting system, if no candidate received the majority of first choices, all second choices were added to the first choices already tallied, and vote totals were checked to see if any candidate reached the new majority threshold. Thus, in contrast to Single Transferable Vote, under Bucklin some voters' votes were counted more than once, and a second-choice vote for a candidate could work as a vote against one's first choice.

Actually, each vote is counted only once for each candidate. But the statement about a second-choice vote is true. In the Duluth implementation, there were three ranks. Overvoting was prohibited in the first two ranks, but the voter could ranks as many candidates as desired in third rank. Warren Smith points out that Madison proposed something like this system for the electoral college, in 1824.

[...] This is why, while 12,313 voters cast ballots in the 1915 Duluth election, the total number of "votes" counted (including first, second, and additional choices) was 18,860.

However, the decision has, quoting with approval another decision: "Our system of government is based upon the doctrine that the majority rules. This does not mean a majority of marks, but a majority of persons possessing the necessary qualifications, and the number of such persons is ascertained by the means of an election."

And then, perversely, the decision proceeds to make the comment above about the number of marks, saying further, "This was not an election of man against man." So ... what flaw of Bucklin? That the ballots, collectively, have more marks on them than there are voters? But, in the end, only one vote from each voter counts, the others could be erased without affecting the result in any way. All those extra marks are alternative votes, simply considered in a different way than with IRV. What stands in the end is that one candidate received a vote from more voters than all others. One vote and only one vote, at most, from each voter.

These flaws of the Bucklin plan -not present in Single Transferable Vote -- led the Minnesota Supreme Court to declare the Bucklin system unconstitutional. The Court first noted that the Minnesota Constitution provided that every male age 21 or older was "entitled to vote" in elections. The Court then said that, when the Minnesota Constitution was framed,
the word "vote" meant a choice for a candidate by one constitutionally qualified to exercise a choice. ... It was never meant that the ballot of one elector, cast for one candidate, could be of greater or less effect than the ballot of another elector cast for another candidate. It was to be of the same effect.

Now, consider this with respect to IRV. If as is normally done, there is a ballot only allowing limited ranks, and there are more candidates than can be ranked with the ballot, some ballots will be exhausted; also because some voters elect, for whatever reason, not to use all the ranks. In a top-two runoff election, with a majority required, these truncated ballots would essentially mean No to all other candidates, the voter preferring to see further process than to vote for one of the remaining candidates. Because of how IRV eliminates candidates and transfers votes, some votes get counted and some don't. It depends on who the voter votes for! So, by this argument, if we follow it, IRV would be unconstitutional. But the FairVote representatives certainly aren't going to so advise!

Guided by this definition of "vote," the Court concluded that Duluth's Bucklin voting system had the effect of giving more than one vote to some voters and was thus unconstitutional. The Court was particularly troubled by how the Bucklin system put voters in a position of undermining the prospects of their first choices when they indicated lower preferences:
The preferential system directly diminishes the right of an elector to give an effective vote for the candidate of his choice. If he votes for him once, his power to help him is exhausted. If he votes for other candidates he may harm his choice, but cannot help him.

(This is the mention of later-no-harm) Note the contradiction. The definition of "vote" was persons, not marks. The interpretation that Bucklin (of this form) gave more than one vote to voters was not followed elsewhere. But the MN Bar article is trying to establish that IRV won't be found unconstitutional, on the argument that IRV is "different." (This argument is also applied by FairVote activists against the city attorney's opinion above, on the claim that he did not understand the difference.) But the thrust of the argument in Brown v. Smallwood is quite contrary to this view. The mention of later-no-harm is brief and not central to the argument. Rather, the idea that a single vote by a voter voting a single preference might be "impaired" by second and third preferences cast by other voters, would be central, and it's repeated. Oddly, this phrase about impairment by lower-ranked votes from other voters, from the first decision, is often quoted by FairVote activists, even though it's a description of what happens with IRV.

In the request for reconsideration filed by the City of Duluth in this case, the court wrote this (130 Minnestota Reports, p. 508, in the pdf above):

We reached the conclusion that a system of voting, giving the voter the right to vote for the candidate of his first choice, and against the first choice of another voter, and, in addition, by manipulation of second and additional choice votes, vote for different candidates all against the first choice of such other voter to a number of times limited only by the number of candidates, was contrary to the intent of the Constitution; and that it was none the less so because such other voter was permitted to engage in a like manipulation of second and additional choice votes.... [We] do right in upholding the right of the citizen to cast a vote for the candidate of his choice unimpaired by second or additional choice votes cast by other voters.

This could not be clearer as a prohibition of single transferable vote. Later-no-harm was not mentioned in the reconsideration. And thus we can see the biased intent involved in this FairVote piece of propaganda, distorting what Brown v. Smallwood actually says. I've come to expect this from FairVote work. It's unfortunate. Solgard and Landskroener conclude:

Thus, a full reading of Brown v. Smallwood shows the Court invalidated the Bucklin system not because it was a preferential voting method per se, but because it had the effect of giving some voters more than one vote, and because it did not permit the voters to fully and effectively support their first choices. Because Single Transferable Vote does not share this fatal flaw, there is no reason to believe that the Supreme Court would hold that Brown v. Smallwood would prohibit a city from adopting Single Transferable Vote for its municipal elections.

In fact, a full reading does quite the opposite. If we pay attention only to carefully selected quotes from Brown v. Smallwood, we can easily conclude what these authors want us to conclude, but, in fact, the court rejected the very idea of alternative votes, no matter what method is used to count them, and this isn't marginal or obscure, it's clear and redundantly stated.

Now, does this mean that Minnesota's RCV system will be tossed out? Not necessarily. Brown v. Smallwood was quite possibly a corrupt decision, very poorly reasoned, not confirmed elsewhere, bucking the legal thinking of its time, and it is rather surprising that it has stood for all this time. The dissent was cogent and clear, and denied the interpretation that Bucklin violated one-person, one-vote. I consider it reasonably likely that the present court will throw out Brown v. Smallwood, reversing it as a precedent, if they are paying attention. Otherwise, led down the rosy path by arguments of such as Solgard and Landskroener, they might well say, "Yeah! Later-no-harm! Bucklin Bad! IRV Good!"

You never can tell nowadays.

Now, as to our present question: The relevance of Brown v. Smallwood to Later-no-harm is thin, certainly not deserving of so much text. There is a reference to later-no-harm in the case, but it was dicta, not central to the argument. On the other hand, a case could be made that the claim, being advanced politically by FairVote is notable, in which case it would have to be stated with attribution and balance.

--Abd (talk) 23:43, 8 June 2008 (UTC)

I can't read all of Abd's unsummarized writing, but interesting thought about Bucklin from whatMarkus Schulze wrote. If Bucklin allowed voters to rank the same candidate in ALL rankings, it becomes a strange form of cumulative voting. That would "fix" the later no harm issue because voters who bullet vote keep their full vote where they want it, and others are willingly compromising their favorite for another choice (splitting their vote). Strategy there is interesting. If all voters are stubborn, it becomes plurality. If some voters want to compromise, they are still "stuck" with their first choice, but can fractionally transfer part of their vote to another choice. I wonder if this would be considered constitutional?! It's certainly not a "runoff" system in my mind, but could be performed in sequential votes, adding results from the new round to all old rounds. OH - BUT it does fail a "majority" ideal since there's no promise any candidate has a majority of the voters. Each round the ending threshold gets divided further - 50%, 33%, 25%, 20%, etc... Tom Ruen (talk) 02:37, 9 June 2008 (UTC)
Well, I can't force Tom to read all that I write, but I'll say this: it's cheaper than a book on the subject, and none of it is a stupid waste of space. Well, almost none. :-)
No, Bucklin isn't cumulative voting. It's a phased Approval Voting. There was a form in Oklahoma that gave lower ranks fractional values, so Tom isn't the first to think of this, but, yes, that was found unconstitutional, and properly so. (It means that if your first rank vote isn't going to win, your second rank vote is only going to be one-half vote.) Bucklin was, indeed, a kind of runoff system. Let's look at the Duluth implementation: overvoting -- as I noted above -- was prohibited in the first and second ranks, voting for more than one was only allowed in the third rank.
So the first round was an FPTP round: vote for one. If any candidate gets a majority, and only one can do that, the candidate wins. But if no candidate gets a majority, then we have the "instant runoff." Votes from the second round are added in. You cannot vote in the second round for the candidate you voted for in the first round. It's not cumulative vote with three votes. It's one vote per candidate. So with the second round, it is Approval Voting with two votes allowed max. And if no majority winner appears, the third round is added in. At this point it is full Approval Voting, unrestricted. You can vote, for example, for "everybody but Bush." If you want. Or you can stick with a bullet vote, which, if your favorite is a frontrunner and you have no reason to want to encourage other candidates -- with the associated risk to your favorite -- that's prefectly reasonable. Bucklin is most simply described as "instant runoff Approval voting."
I see no reason to prohibit overvotes in the first and second rounds. Someone who does that is essentially saying "Don't discriminate against either of these two in favor of the other, I want to give them both the same chance to prevail in the first round, as well as subsequently." But Bucklin fails later-no-harm, definitely. And it *should*. That's part of the point about later-no-harm, there are experts who think that it is actually an undesirable characteristic. I just added a Woodall quote to the article on that point.... --Abd (talk) 03:24, 9 June 2008 (UTC)
I meant "effective fractional votes" in a Bucklin that allowed putting the same candidate in multiple rankings. Example, in "round 3", everyone has 3 votes, or 3 1/3 votes is the same thing. If I vote Fred-Fred+Fred, my vote is 1/3+1/3+1/3=1 vote for Fred. If I vote Fred-George-George, my vote is 1/3 for Fred, 2/3 for George. That's fractional cumulative voting within each round of voting. That's what Bucklin "COULD DO" if you could repeatedly vote for the same candidate in different rankings. In my mind it would be defendable as a "one person, one vote" system since everyone gets the same influence every round. BUT the failure of this (as I tried to say) is that Bucklin's measure of "majority" is lost. If a voter is PREVENTED from voting for the same candidate in two rounds, THEN, you know when a candidate gets 50% of the votes (counting as full votes by each rank), he has been supported by half the voters. With repeated rankings, that 50% could be reached by a subset of bullet voters, i.e. 50% of voters united in the first round guarantee a winner, 34% of voters in the second round (bullet voting) guarantee a winner (although there could be two pluralities fighting to win), and 25% in round 3, etc. Tom Ruen (talk) 05:56, 9 June 2008 (UTC)

[edit] Condorcet criterion?

I'm surprised by this claim:

It seems a contradiction. If Minimax Condorcet satisfies LNH, and it always picks a Condorcet winner, then something is fishy here! Perhaps the Condorcet criterion doesn't apply because it isn't a method that always picks winners? That's the only weakness I can see. Tom Ruen (talk) 17:40, 12 June 2008 (UTC)