User talk:Tomruen/A sensible argument against IRV

From Wikipedia, the free encyclopedia

[edit] A sensible argument against IRV

I moved this essay from the Talk:Instant-runoff_voting#Summary:_IRV_is_a_.22single_vote.22_in_Intro page, since it'd just get lost there. Tom Ruen (talk) 10:23, 8 January 2008 (UTC)

When faced by nonsense, the easiest thing to do is discount the speaker, tell him why he's wrong. The more interesting response is to discount the nonsense and see what's left. I mean it's always possible someone else has an intuitive reason to oppose something, but can't explain it logically. So let me try.

Who might not like IRV (as proposed as a bottom-up runoff) from the perspective of plurality or a top-two runoff?

  1. Major-party candidates: A candidate who expects to be the plurality candidate won't like it. He wins in plurality. He's guaranteed a head-to-head competition against a single opponent in a two round system. While in IRV, he (and his supporters) are supposed to sit idly by while who-knows-what-nonsense-conspires for the small possibility that two opponents will pass him up in the final round and disallow his fairly obtained advantage to be treated fairly in a head-to-head competition. (AND since major party candidates always EXPECT a chance to be the plurality candidate, they'll see NO reason ANY candidate below plurality-second should be given even a second look.)
  2. Minor-party candidates: A candidate who believes he has a shot to be the plurality candidate won't like it. He FEARS he may NOT be acceptable to a majority and is WILLING to win PURELY because his major-party competitors split his opposition vote.
  3. Strategic voters: Voters in an exhaustive ballot will vote strategically every round, using whatever information they have to decide IF and WHEN to compromise. A single preference ballot doesn't allow this degree of control. It is a difficult theorical problem to look at N candidates and try to imagine ALL POSSIBLE elimination orders to determine WHICH stategic compromise to make under WHAT conditions and WHICH are most likely to deserve a vote. Given a bottom-up elimination is hard to predict, strategic voters will likely support eliminating MORE candidates at the start, reducing the potential cascades of improbability from upsetting their efforts.
  4. Stubborn voters: Voters who DON'T like any of the candidates who stand a chance snowball's chance in hell to win and want an HONEST vote in the final round for their idealistic loser. They don't CARE that this vote won't affect the winner. They just want an honest count in the final round, showing their support and showing the TRUE support of the IRV winner. (IRV has no mechanism for registering such votes as valid. At least a TRUE runoff would allow protest write-in votes.)
  5. Lazy voters: Voters who KNOW their first choice are happy to vote in a two round system, but are given the uncomfortable task of ranking lower choices just in case their first choice is eliminated. Being lazy such voters may make a last minute decision, and regret their choice after they know who made it to the final round, and they'll be unable to pick a different lower choice in later rounds (like the two-round system would allow.)

So, that's all subjective of course, NOTHING to do with legal arguments for or against IRV, but people who may think that way will STRETCH whatever illogic they can find to support their feelings.

To support the legal argument against IRV, I don't believe you can argue one person, one vote is false, BUT you can argue that (1) Bottom up elimination has NOTHING to do with majority rule, allowing a tiny minority of the vote to repeatly change their votes (If the top-two candidates combined control 51% of the vote, majority rule COULD be used to force the elimination of all lower candidates in one step.) (2) A one-vote system like plurality rewards compromise before the election. An automated bottom-up runoff system that weakens this reward system weakens the incentive to pre-election compromise (OR weaken voter clarity towards this goal.)

From those two arguments, I've come to the conclusion that a top-two IRV election is the best you can fairly support (within the plurality ideals) since it rewards the STRONGEST two plurality candidates equally. Majority candidates will accept it as fair because they have a clear threshold for viabilility, AND they've got the confidence to believe they can win the final runoff round. Third party candidates will also accept it as fair - they are safely eliminated if they're third or lower, allowing supporters their second choice, and get EQUAL treatment when they reach top-two. Strategic voters are happy with a greatly reduced conditional result process to analyze.

A top-two IRV also handles the stubborn voter who refuses to rank below their favorites. In the first round they get a vote for ONE favorite. In the second round, they get their compromise opinion between the final-two OR they can keep their favorite counted as a protest vote, just like a real runoff write-in protest. (A bottom-up IRV could also COUNT a vote back to a FIRST choice after all rankings are eliminated, but it's sort of weird to get a FIRST, SECOND, and THIRD choice, and then returning to a FIRST choice after giving up!)

A top-two IRV doesn't help the lazy voter, whom would prefer voting twice in a two round system over ranking choices.

I'm just trying to express a rational defense why IRV (as promoted as a bottom-up) might not be a good idea, EVEN if it is good for multiseat elections of STV. Bottom-up elimination may NOT be a great idea, even if fair-minded people want to believe it is. It is an opinion that bottom-up elimination is best, NOT a fact, NOT defendable by majority rule. (I'm a fair-minded person myself. I think bottom-up elimination of IRV is quite good on average - it'll statistically produce a Condorcet winner more often than a top-two runoff, BUT I'm trying to defend a DIFFERENT opinion here - that plurality strength is worth protecting.) AND MOST EM-theorists don't know what I'm talking about since they're too busy trying to escape plurality to wonder about any virtue in it.

Unfortunately this detour has NOTHING to do with opposition to a one vote interpretation of IRV. I still believe that as a fact. I believe it is an important fact that distinguishes it from MOST other preference ballot methods and ought to be stated in the introduction.

Tom, what you believe is not a source for the article. I can source -- and have -- arguments contrary to the "one vote" interpretation, and one of them happens to be a state supreme court, if you've read the opinion. The other, most relevant now, is a position expressed by a voter rights organization in Minnesota which I think has actually filed a lawsuit, and I assume this is going to be one of their arguments, since the suit is based on Brown v. Smallwood. *Personally* I find the "single vote" interpretation a reasonable one, *but* it is quite properly controversial. An accurate and simple description of the actual *process* makes clear the "single vote" interpretation, relatively well, as does noting the origin of the term with "Single Transferable Vote," which is also accurate and not pushing the "single vote" interpretation. What was objectionable was taking the link to the STV article and reducing it to the display of "single vote," making it into an argument that appears to be a fact. Yes, it's subtle. Do understand that some very smart people were paid to develop all these "facts" and to package and present them in such a way as to lead naive readers down a particular path. You may think this is polemic, it's not. This is what happened. The very name "instant runoff voting" was suggested by someone, we think -- it is consistent -- who, apparently, came to regret it. The "single vote" in the introduction is, to me, a relatively minor point; were this the only problem, I might not even pay attention to it.
As to the arguments against IRV, yes, this is a whole class of argument. Used to be, the only Con argument allowed in the article by the cabal was that, summarized as a preference for the status quo. However, it neglects the entire world of election methods experts, *very few* of which support IRV. IRV is being promoted by political consultants and some political scientists, the mathematicians who study election methods are, as far as I can tell, *entirely* arrayed against IRV. Gotta leave now for a biopsy. --Abd (talk) 15:08, 8 January 2008 (UTC)
Obviously not ALL mathematicians are against IRV (At least for me, I have a BS in Math, and studied EM for some 7 years!). It's like saying 'Democracy is the worst form of governance and leave out except for all the others. Any critic can slam ANY solution as poor. IRV has good average behavior even if poor special case behavior.
I must pity you. Rather than risk leading readers down a particular path, you prefer to make an unreadable mess that says nothing understandble and uses a damned large number of words to do it. Tom Ruen (talk) 22:25, 8 January 2008 (UTC)
Tom, that's rude. Stop it. As to the point, you have a BS in math. Warren Smith has a PhD, and has specialized in the study of election methods for some years. Note that the American Political Science Association "uses IRV to elect its President." Of course, they apparently have *never* actually used it. On the other hand, the American Mathematical Society uses Approval voting. Actually uses it, since it is a method that requires no fancy ballot or rigamarole. I'm getting the impression, Tom, that you have not done much research in this area.
IRV has relatively *good* behavior *under special conditions.* Most notably, a two-party system. However, evidence is that it helps *preserve* a two-party system. Remember all the debate over that in the article? Would I prefer IRV over Plurality? For the election of the Presidential electors in 2000, damn straight I would! But Approval would have solved that problem with no expense, and, indeed, that election in Florida showed, quite clearly, the injustice involved in tossing overvotes. Those overvotes contained information that showed a different winner. There was more error created by discarding them than was avoided by counting them (and that's generally true). The Gore/Buchanan overvotes can be known, statistically and from voting patterns analysis, to be mostly intended Gore votes. What would have been the harm in counting them? (Aside, of course, from violating Florida law!). Approval has been studied for years, and the only articles I've seen that are critical of Approval are seriously flawed; for example, the so-called Burr dilemma article, published in a "peer-reviewed" journal, of which so much is made by FairVote, that terms the early U.S. Presidential elections as being "Approval" elections, which is patent nonsense. Where does IRV *not* perform well? Let's start with San Francisco! And I will. Watch. San Francisco conditions -- nonpartisan elections, a wide field of candidates -- are precisely where IRV can break down badly, and, from preliminary analysis of voting records, it seems it is doing just that. (And "Ranked Choice Voting," which is what SF is, is worse than full-ranked IRV because of the limited ranking, three ranks is pretty small when you have more than a dozen candidates on the ballot! Yet [FairVote]] trumpets San Francsico as a big success story. Sure, it saves runoff costs, but with huge setup costs and with, essentially, discarding the majority vote requirement that caused the many candidates to run in the first place, because it crimps the spoiler effect. My first analysis, already on the IRV talk page, of the 2000 election, shows that Plurality would have elected the same candidate in six out of the nine elections that went to runoffs, and it is fairly likely that, had the method been plurality, the other three would have gone the same way. People vote differently with Plurality, essentially anticipating the necessary compromise and voting for a frontrunner. But, of course, that's hard to prove, and certainly spoilers do occur with something like 5% frequency, I think is the estimate for U.S. Presidential elections, if it were 10% in San Francisco, that would still point to, likely, all three remaining elections coming up with the same result between Plurality and Top-two runoff.
What I do get from your posts is that you are nobody's meat puppet. You sincerely believe that IRV is a good method. I happen to think that you've swallowed a fair amount of propaganda, it can get the best of us. If you think IRV is a good method, why don't you take that idea to the Election Methods mailing list? See if it flies. You will find *some* support, particularly from an Australian writer, who *also* severely criticizes IRV. What you will find, though, is that you will be discussing IRV with a community that is *knowledgeable* about alternatives. IRV looks good if all you are thinking of is Plurality, and that has been, quite precisely, FairVote strategy, to always frame the debate in those terms. As you know, Bucklin resembles Approval in many ways, and, because it preserves ranking, it might be more acceptable in the U.S. than pure Approval. It was *not* discarded because it wasn't working, contrary to FairVote propaganda. It was working. Brown v. Smallwood reversed an election that was quite obviously just, and did it in a quite offensive manner (there were other options, such as accepting the result of that election and declaring, however, the method illegal, or, alternatively, holding a new election -- which is a more common remedy when an election has been found to be improper. Instead, the court declared the plurality winner in the first round to be the winner, which ignores the fact that voters *depended* on being able to honestly rank candidates, thus did not vote for the favorite frontrunner as would be normal with Plurality. Bad, bad decision.)
I think you could be of great help in improving the IRV article. I have no intention of making that article into an IRV hit piece. I want it to be fair and balanced; I just happen, because of the accidents of my browsing history, to be quite sensitive to the propaganda, I know what points -- that might seem to another to be some harmless quirk of wording -- are actually critical parts of a political campaign, because I know the arguments and the actual debates, having followed Richie around the internet. He makes the arguments, and when someone points out the flaws, he disappears, he knows that many readers lose interest after the first couple of posts in a forum, and he knows how to craft the sound bites. He's politically quite effective. Problem is, he is repeating arguments that he *must know* are misleading. Many, many observers have concluded this about him, it is by no means simply my own opinion. He's like many other political PR people, who only care about winning, not about truth....
So what I ask is simply that you continue to read and to improve your understanding, and to participate in the editing. I write a lot. You certainly don't have to read it all! But if you focus on the actual edits, and then particularly on the ones which start to become edit wars, it's a sign that there might be something important there. Perhaps you should look a little more carefully. --Abd (talk) 00:30, 9 January 2008 (UTC)
I'm sure you can find me in the EM archives [1], and yahoogroup IRV-Freewheeling [2]. I also sent you a second email, with some of my babble on the best compromise method - a two round Plurality-Condorcet hybrid, to maintain coalition advantages of plurality while being able to treat more than two candidates equally in the second (Condorcet) round. (Also I'm attracted to the simplicity of a two round Bucklin - a plurality-Approval hybrid, although in practice I expect the first round to mostly fail to find a winner, and the second round fail to produce many compromise votes.)
I still contend that your idea of fair-and-balanced seems more than to be skewed to making it unreadable. Well, I feel failing upon myself, and that's why I like adding pictures over writing. Tom Ruen (talk) 00:49, 9 January 2008 (UTC)

[edit] A sensible runoff

I'd ideally support a compromise that honors both Plurality and Condorcet. IRV fails both of them. It dishonors plurality strength by allowing candidates to be demoted, excluded from head-to-head competition even if plurality first or second. IRV dishonors Condorcet by failing to give proper accounting of 3 or more strong candidates.

My compromise, ideally is a 4 round runoff process, up to 3 rounds of voting.

Round 1
Round 1
  1. Primary: Overvote plurality ballot counted as an (equal&even) cumulative voting with an instant runoff - Eliminate all candidates below a given threshold (Either a fixed number of votes, like 1000, or a fixed proportion of votes, like 5%)
    • Write-in candidates are allowed in this vote.
    • Overvotes are counted as a divided vote. (Split vote divisors will reduce as choices are eliminated.)
    • One weakest candidate below 5% is eliminated at a time, allowing fractional transfers from voters who have overvoted.
    • I DISALLOW a full approval count to better reward voters who have worked hard before the election to gain a core following, AND disempower voters who want to mess things up in the second round by false overvotes in the first round. Having a penalty for overvoting means voters will only do it if they're confident in their mixed-support, and the bottom-up runoff best ensures a chance for their vote to count to at least one winner.
    • If any one candidate gains a majority on this ballot, a winner is declared immediately.
      Round 2
      Round 2
  2. General election: Instant Runoff Voting - A rank ballot vote among winners of round 1 - Eliminate one bottom candidate at a time until all candidates remaining exceed a given threshold (like 20%).
    • If any one candidate gains a majority support (including exhaustive ballots against the candidate), a winner is declared immediately.
    • Tied-rankings are allowed and will be counted as a split vote among a tie of highest ranked candidate.
    • If there's a tie for last in any round, eliminate one randomly.
    • Optionally a Single transferable vote count could be done here if a fixed number of winners is desired to compete in the runoff round. Then the threshold becomes the droop quota and surplus transfers will be included.
  3. Runoff election: Condorcet method - A head-to-head ballot vote among winners of round 2 - Ask voters their preference pair-wise between remaining candidates.
    • In each head-to-head question, include options: don't know and don't care, which won't affect the winner, but are tallied for the purpose of accounting and feedback to the real consensus.
    • A winner is declared if one candidate can beat all others head-to-head.
  4. Tie-breaker: Smith set - From round 3, find the smallest set of candidates which can defeat head-to-head all candidates outside the set. Consider these candidates in a virtual tie and pick a winner randomly from this set.

I offer this as three ballotings because voters are progressively asked more and more detailed information.

  • Round 1 (Take out the trash) This rewards candidates who can collect a solid 5% core backing. It rewards pre-election efforts to collect supporters and quickly eliminates candidates who fail to inspire sufficient support.
  • Round 2 (Mop up the stragglers) This uses IRV as a mop up activity, giving candidates a safe threshold for passing, and allows candidates who fail a final chance to rise on compromise appeal. Voters don't NEED to rank unless they're concerned their favorite might fail to reach the threshold.
  • Round 3 (Full evaluation of the contenders) This gives voters one last chance to focus on the final set of candidates. All candidates are treated as equals and there's no elimination. This set will be small, usually two, if needed at all, and so the process in such cases is reduced to a two round system. In the rare case when there's three strong candidates (over 20% support), voters are asked for three pairwise preferences so the consider them in detail. Each pairwise comparison is between proven candidates passing through the previous rounds. In the most unlikely case, it's possible for 4 candidates to survive the second round, in which case there'll be 6 head-to-head competitions. This could be done with a rank-ballot, but with only 6 head-to-head questions, there's no harm in such a balanced race to ask each explicitly.
  • Round 4 (Respect a failed majority) This comes as the most unlikely case, and can only happen if there's 3 or more candidates without any single candidate who is always above 50%. In such a case it is best to identify the set of mutually pairwise defeatable candidates and consider it a virtual tie. There's no reason to ask voters for anything more - they are mixed up, following different reasons and a single majority winner doesn't exist. So the tie is best broken by random determination, just a like any vote count tie among two.

This process for me seems fair and comprehensive, asking from voters only what is needed, and in each step further honoring the winners from the last step. There's no demotion except for candidate who fail the pre-election determined thresholds.

It COULD be done in a single rank balloting to speed up the process. It is possible to demand a ballot-majority process in each step, and in any round that the exhausted/undecided vote exceeds the power the influence the result, it could be halted and reballoted. Giving willing voters the power to speed up the process, while not taking away their option to reballot if enough are undecided, seems sufficient to me to honoring the voters time and interest in the process.

Finally notice that this process becomes STV for multiwinner elections, with rounds 3 and 4 unneeded.

... Post-thought... curious about the elimination of round 2, and IRV ranking at all. Round 1 with an approval ballot offers voters the chance to compromise with a divided vote, and transfer by elimination. The 20% threshold means "major party" core-voters can bullet vote safely like they want.

The interesting question is whether ranking ought to be considered a voter's right or not. IRV makes such poor use of rankings anyway, and if the top-vote means the most for most voters, why not just say forget the ranking and let compromisers split their one vote as they like. Well, it has some merit in my mind, and it offers the plurality merit argument clearly, with an easier goal of 20% for viability. Tom Ruen (talk) 05:14, 20 February 2008 (UTC)