Rebuttal to Arguments in Favor of IRV

This is a reply by Mike Ossipoff to a web article by an IRV advocate who here will be known only as "Irving". No disrespect is intended by the use of this pseudonym. Irving's original text is indented.

(1) The Pairwise system is a strong and fair single-winner election system. In fact, on purely theoretical gorunds, it can be said that no system is superior to it.

So far so good. But I'd add that when Irving says "Pairwise", he's referring to the general classification of pairwise-count methods. Actually some of those are excellent and some have little merit. The excellent ones are the versions of Condorcet's method, named after their original proponent, the 18th century founder of voting theory. I won't devote a lot of space to introducing that method here, because that's done elsewhere at this website. Let me just say that, with pairwise-count methods, if more people rank A over B than vice-versa, then A is said to "beat" B. And if one candidate beats each one of the others then he wins.

Pairwise-count methods differ in what they do when, due to a "cycle", where, for instance, A beats B beats C beats A, no candidate is unbeaten. One of the refined interpretations of what Condorcet meant to do in that instance is known as Sequential Dropping (SD). SD probably has the most support among Condorcet advocdates, as the best Condorcet proposal. Here's its simple rule for when there's no unbeaten candidate:

"Drop the weakest pairwise defeat that's in a cycle. Repeat till there's an unbeaten candidate."

(The strength of B's defeat by A is measured by how many people ranked A over B. In other words, a defeat's strength is the number of people who voted for that defeat).

Although Irving is only comparing IRV here to pairwise-count methods, and so that's what my reply is about, I should add that there's another method that's also much better than IRV: The Approval method. Because I might mention it later, let me briefly define it: Approval is like Plurality except that the voter can mark more than 1 name on the ballot if he so wishes. He can mark as many as he wishes, and by doing so he gives 1 whole vote to each candidate whom he marks. I like Condorcet best, but Approval is a close 2nd.

(2) However, when other factors are taken into consideration, I have a preference for IRV ("Instant Runoff Voting", also known as the Instant Run-off System, IRO, Majority Preference Voting, the Alternative Vote, and other names).

(3) As a whole, while single-winner election system reform is certainly important, it is much less important than making the move to Proportional Representation.

People who consider single-winner reform to be of secondary importance should leave it to others. I urge Irving to promote Proportional Representation, and to leave single-winner reform to people who take it seriously enough to inform themselves about it. Some of us feel that IRV advocates are trying to push IRV on you because it resembles a PR method that they want to lead into. Sometimes they admit it. Irving gets into that later in this article that I'm replying to.

I don't oppose PR, but recent decades have shown that it isn't winnable in the U.S., and certainly is nowhere near as winnable as single-winner reform. PR is a whole different concept of representation. Single-winner reform would just improve the fairness of the same elections that we now conduct. People express great suspicion of their representatives and the interests behind them. A complete change in the concept of representation is viewed with suspicion therefore. So it makes more sense to work on the kind of reform that's winnable. Contrary to the claims of IRV advocates, the use of a genuinely better single-winner method would make a tremendous difference. Admittedly that wouldn't be true of IRV. Irving continues with a list of comments.

(1) The positions in the paper are based on my experiences, readings, discussions with people strongly advocating various systems, values, and judgements. Since different people have different values, judgements, etc., I don't expect this paper to necessarily convince someone who strongly advocates Pairwise that they should change their position. Rather, the idea is that they understand where I, and to some degree the PR movement as a whole, are coming from on this issue.

Yes, different values & judgements. For instance, we all agree that majority rule is important, and that we'd like to get rid of the lesser-of-two-evils problem, but Irving has his own unusual definitions of those things:

Irving, along with other Irvies, seems to feel that if, after we've eliminated all but 2 of the candidates, if we then elect the one of those 2 who is ranked over the other by more people than vice-versa, then we're carrying out majority rule. What about if we've eliminated a candidate who, from the start, had majorities over each one of the other candidates? That doesn't seem to bother Irvies.

Say I take you to a restaurant, and give you the menu, and say "You can have whatever you want." You point to something and say "I'll have that." I say "You can't have that. Which of these 2 do you want?" If you believe that I've given you what you want, then you'll believe that IRV carries out majority rule.

Say we define the following method: Determine which 2 candidates are the 2 most despised of all the candidates. Hold an election between just those two. If a majority then vote one over the other, then we can justify electing him by saying that we've carried out majority rule, right? Well maybe not :-)

I'll comment about Irving's concept of the lesser-of-two-evils problem when he talks later about evidence.

(2) This paper officially reflects my position only. The leadership of the PR movement is not terribly interested in spending a lot of time debating Pairwise vs IRV.

If the leadership of the PR movement had spent more time on differences like that, then maybe they wouldn't now be promoting a meritless method. But I hope the leadership of the PR movement can forgive me for commenting on such things as IRV vs Condorcet, IRV vs Approval, and IRV vs Plurality. Before you adopt a new single-winner method, it's best to hear the arguments pro & con. Don't let a "reform" be pushed on you by someone who doesn't reply to discussion of its problems, or who objects if his opponents tell you of the proposal's disadvantages.

Of the leadership of the PR movement, I believed I've studied the issue the most. We've pretty much decided to push IRV, because it is equal to Pairwise on theoretical grounds, and superior on pragmatic grounds

Irving neglects to say what sort of a theory he used for determining that IRV is theoretically equal to Condorcet :-)

Elsewhere at this website is a page that defines & discusses criteria for evaluating & comparing single-winner voting systems, such as IRV, Condorcet, Approval, & Plurality. Check out that Criteria webpage. You'll find that IRV is inferior to Condorcet by every criterion that distinguishes betweent the two. And IRV is inferior to Approval by every important criterion.

A criterion is a precise yes/no test for comparing what voting systems will or won't do, as a way of comparing & evaluating their merit. Criteria are the way of saying something precise about what a method will or won't do.

(2) Pairwise has a bit of a problem with handling circular ties.

Wrong. Circular ties aren't specifically a problem of pairwise- count methods. When, due to a cycle, every candidate is pairwise-beaten, that remains true no matter what the method is. Sure, IRV ignores the pair defeats, because it doesn't reliably count everyone's preferences. Ignoring something doesn't make it go away. So, no matter what voting system is used, when there's a circular tie, any candidate whom you elect will be someone who is pairwise-beaten. The difference is that Condorcet deals with that, by dropping weakest defeats, and that Condorcet is the method that reliably counts all of the preferences that you vote.

All methods that I've heard of for handling them have problems: 2nd elections, "cheatproof", etc.

In typical form, Irving neglects to tell us of the problems that he claims that every circular tie solution has.

(3) Note that IRV winners will tend to be those with enthusiastic support, while Pairwise winners will tend to be compromise candidacies.

Irving is completely missing the point of rank balloting. The whole idea is so that you can cast a vote for a compromise against someone less-liked, while still casting a vote for your favorite over the compromise. If we're a rugged noncompromiser like Irving, and we aren't interested in compromises, then we have no reason to be interested in rank-balloting. Plurality would then be fine for us: Just vote for your favorite, you courageous noncompromiser!

But if you do choose to rank a compromise, you do so in order to vote for him against someone you like less. You want that vote counted. Sorry, Irving, but it isn't for you to say that that preference doesn't get to be counted. And of course, all you rugged noncompromisers can refuse to compromise in Condorcet if you want to, by only voting for your favorite.

Neither, therefore, is inherently superior to the other, it is simply a matter of approach, a value judgement.

...except that IRV is drastically inferior to Condorcet by every criterion that distinguishes between the two--and only a few of the trivially weak criteria fail to distinguish between them.

(4) Pairwise has a potentially fatal flaw: Since a voter's second choice could easily help defeat their favorite candidate, campaigns might very well ask their voters to bullet vote. This is reality - it killed the Bucklin system.

Also, with IRV, it will often be necessary for you to insincerely vote your 2nd choice in 1st place, to prevent someone worse from winning. But if you do that when, unknown to you, your 1st place could have won, then you'll give the election away to the 2nd choice. When you do that, your support for your 2nd choice _is_ making your 1st choice lose.

With Condorcet, when your favorite has a clear win, when your favorite is the candidate who'd beat each of the others in separate pairwise contests, voting a 2nd choice can't hurt your favorite's win. Actually, all that can happen is that, when your favorite _isn't_ that universal winner, and if for some reason there's so much indifference that the universal winner doesn't have majorities against the other candidates, then, by not ranking that candidate, you could gain victory for your own favorite instead. That's the only sense in which you can gain by voting a short ranking instead of a complete ranking. But note that it's an offensive strategy that can only work if the there's so much indifference about the universal winner that he doesn't have a majority against your candidate.

As for Bucklin, Condorcet isn't Bucklin, and Irving's statements about Bucklin have no relevance to Condorcet. Besides, in Bucklin, voting only for one's favorite is the best strategy when you believe that your favorite has a win. Maybe when Bucklin was used, people believed that their favorite had a win. Maybe , in those municipal elections, they were above using strategy, and preferred, on principle, to vote only for their favorite. There's nothing wrong with that choice. It certainly doesn't mean that Bucklin failed. This Bucklin thing is a common mis-statement that we frequently hear from the promoters who are trying to push IRV on you.

Luckily, there are a couple of ways to handle this: (a) use the "cheatproof" tiebreaker [need to get the definition from Mike Ossipoff again], (b) require voters to fully rank the ballot. However, both of those methods have their own disadvantages.

Again, Irving forgets to tell us what the alleged problems are. Condorcet is by far the most problem-free voting system. IRV is a problem-ridden strategic mess.

(5) IRV has a huge advantage over pairwise from the point of view of PR activists. When we write up ballot language for STV, we can use exactly the same language for IRV. We've handled multiple-winner reform and single-winner reform in one fell swoop. If we proposed STV for multiple-wnner elections, and pairwise for single-winner elections, ...

Here is where Irving admits why he wants to sell you an inferior single-winner method: He, and his other Irvie friends, like IRV because it leads right into something else that they want. IRV would provide precedent for a PR system that the IRVies want to get enacted. Those of us who take single-winner reform seriously resent the IRVie contemptuous attempt to use single-winner reform as a stepping stone for something else.

"...multiple winner reform & single-winner reform in one fell swoop"? A very clumsy fell swoop, a one-size-fits-all sales-pitch that is a lie, because IRV is quite inadequate as a single-winner reform. I encourage multiwinner method advocates to promote multiwinner methods, but they should be ashamed of trying to sell you a worthless single-winner reform because they think it would lead into a multiwinner method that they like.

Actually, some of us believe that wouldn't happen. We believe that IRV won't lead into anything except embarrassment. IRV will discredit rank balloting, electoral reform, and the uninformed promoters who are pushing it.

We'd have to write language for two systems, and explain two systems - not very attractive.

Ok, so Irving wants to use a PR system for single-winner elections, to save himself from having to write separate language? :-) And the fact that his PR method is no good for single-winner elections doesn't stop him from trying to waste your time with it.

(6) IRV is tested and proven in public and non-governmental elections. Pairwise, to my knowledge, has never been used in a public election, and is very rare in non-governmental elections.

In what sense is IRV proven? It's been used in Ireland for their Presidential elections. But an Irishman told me that that office is a ceremonial office. IRV is used for one house of Austrialia's Parliament. But Irving hasn't offered any data to show that IRV doesn't do, in those elections, the undesirable things that I've been saying that it is sure to do. "Proven"? Irving hasn't proven anything. As I said, parties in Australia rarely run more than 1 candidate in IRV elections, in spite of the fact that, when IRV was instituted, it was hoped that they would, so that voters could have more selection. Parties apparently fear a "spoiler problem". I've been told by an Australian that parties have a difficult time getting their voters to vote for their favorite candidate in 1st place, instead of giving 1st place position to whichever of the big two parties' candidate they like better than the other--a problem that I'd predicted, and one that's familiar to U.S. voters.

(7) We need to remember that voters typically have a strong preference for one candidate. They are not dispassionate machines simply ranking preferences. Pairwise advocates tend to forget that if I rank the candidates, B, A, C, D, that it might very well mean that if I had one dollar to spend to help each candidate that I might give 90 cents to B, 10 cents to A, and zero cents to C and D.

Nonsense. You might just as well rate your 1st 2 choices equally, and rate your 3rd choice much lower. In any case, it's common for voters to dump their favorite in order to support a 2nd choice compromise. They'll do it in IRV too, because they'll have to. They won't have to in Condorcet or Approval.

8) I strongly suspect that in the real world, the Condorcet winner and the IRV winner would be the same person about 90% of the time. So realistically, is the difference really worth much? I doubt it.

So we should choose a new voting system based on Irving's "I strongly suspect..."? IRV fails the voter often enough that there's no reason to settle for that meritless method.

(9) Note that anyone can come up with scenarios which favors their position, in any debate.

Good, so where are the scenarios that favor IRV? :-)

/If anyone can come up with scenarios that favor his favorite method, then it's curious that Irving doesn't include any such scenarios that favor IRV. And if it's true that anyone can do that, would that make all methods equal? Not unless you think all the "scenarios" are equally important, or that the arguments based on them are equally valid or important.

Instead of referring to "scenarios", I refer you to the webpage, at this website, about criteria. With critera, we can talk about what methods will or won't do.

Therefore scenarios have their uses, but have huge limitations. Much stronger evidence includes real world experience and studies that follow good scientific and statistical principles.

Irving neglects to say what those limitations are. He speaks of much stronger evidence that includes real-world experience, including scientific & statistical studies, but Irving, true to form, doesn't offer us any such evidence.

(10) I can not cite the source, but it has been shown mathmatically that in a public election it is virtually impossible for a voter to vote strategically in any useful manner when using IRV.

Irving can't cite the source because there isn't one, unless it's Irving or some other Irvie. Not only can you vote strategically in a useful manner in IRV, but sometimes IRV forces you to do so. Sometimes you'll have to use the strategy of dumping your favorite in order to prevent your last choice from winning.

(11) Pairwise advocates often throw up all kinds of IRV horror stories and suggestions of strategic voting. But in reality, none of this has ever happened, despite IRV's extensive use in Austrilia and Ireland, much less non-governmental elections.

Irving offers us no proof for his claim that those predicted failures didn't happen. Why doesn't he cite a source of information here? Why doesn't he tell us what data support his claim, or how they support it? Because he has no data that support his claim.

I doubt that the kind of data are reported & recorded, in countries that use IRV, that would show that IRV does or doesn't do the undesirable things that I & others say that IRV must do.

Irving has no data.

But it's known that in Australia's IRV elections, parties rarely run more than 1 candidate, even though, when IRV was adopted there, it was hoped that its use would encourage parties to run more candidates, to give the voters more selection. But it didn't happen. Apparently the parties fear that their candidates will split the vote & lose. In other words, IRV in Australia has demonstrated a spoiler problem.

An Australian told me that in Australia's IRV elections, the parties other than the big 2 have a difficult time getting their voters to vote their party in 1st place, because the voters feel a need to vote tactically, voting their more preferred of the big-2 candidates in 1st place. In other words, IRV voters suffer from the lesser-of-2-evils problem.

No, I haven't gone there myself, and my only information is from that Australian. But the person who told me of that problem was someone who had no incentive to make up an IRV problem.

The rarity of parties running more than 1 candidate in IRV elections is well-established.

Ireland uses IRV too, for its Presidential elections. But an Irishman told me that he wouldn't expect any strategy incentive anyway in that election, since that election is for a ceremonial office only. And again, Irving offers no data from Ireland, just as he offers me no data from Australia.

(12) We need to avoid the mistake of thinking that voters think like we do. The typical voter is anything but a mathmatician and election system analyst.

So what? It doesn't take a mathematician or an election system analyst to know that he needs to dump his favorite and vote for a lesser-evil. It happens in Plurality, and it will happen in IRV. And if you don't sometimes dump your favorite in IRV, then you'll regret it (because, in the U.S., some of us will be publishing the data that will show IRV's failures and violations of majority rule, and its failures to count your preferences that are important to you).

Say there are 2 extreme sides, and a Middle. Sometimes it's necessary to for 1 extreme to insincerely vote Middle in 1st place to avoid the election of the opposite extreme. If your extreme makes sure that it only errs in the direction of sincerity, in the direction of not putting Middle in 1st place when you should, and if the opposite extreme makes sure that it errs only in the conservative direction, in the direction of sometimes putting Middle in 1st place even when it isn't necessary, then your side is being had: IRV's swings to extremes will only go to the opposite extreme, never to yours. If you keep voting so sincerely, if you keep being had in that manner, then you're a sucker.

If both extremes start out voting sincerely, or trying to err only in the sincere direction, then one side will surely start taking advantage of the other side's naiveness. And that victimized side will soon putting Middle in 1st place more, erring conservatively, if they know what they're doing.

So sincere voting in IRV is an unstable situation. It won't last.

Does Irving think that no one will tell the voters how they can improve their outcome by insincere voting? No one will tell their faction how to get a better result?

(13) In Pairwise, the middle candidate is the King. In IRV the middle candidate is the Kingmaker. I find either situation to be fair and acceptable.

I prefer the electorate to be "king", and to have their expressed majority wishes honored, and to have all their individual expressed preferences counted. I don't want Irving to play kingmaker with our elections.

(14) One potential ugly result with pairwise could occur. A virtual unknown rogue candidate could insert themselves into what is effectively a two-person race, position himself as the middle candidate, and win.

This is Irving's "unknown middle" argument.

First of all, the voters are adults, and we must assume that they're qualified to make their own evaluations & choices.

If someone is a "rogue", then no one forces you to vote for him. If you don't believe that someone is a rogue, then it should be your right to vote for him.

I find it a little scary that someone in this country can say that you shouldn't be allowed to have your preference for Middle vs Worst to be counted, because you might not know enough about Middle. That's for _you_ to judge, not for Irving to decide for you. Perhaps Irving would like to take the right to vote away altogether, because you might make a bad choice, or not correctly identify a disguised rogue.

Additionally, the voter median position will be a popular & crowded place for candidates, especially as Condorcet or Approval encourages more people to run. So how likely is it that the only person at the voter median position will be some unknown rogue? There will be wellknown & well-understood candidates there.

Also, aside from all that, the media will call into question candidates' claims of their positions or their background. Opposing candidates won't keep quiet about that either. Irving is being silly when he says that you won't have any way of knowing about the candidates near the voter median position.

Once elected, who knows?! IRV does not have this problem because it requires that you get a lot of #1 votes in order to win.

"IRV does not have this [nonexistent] problem" because it often ignores your expressed preference for your 2nd ranked candidate over your last choice. Thanks a lot, IRV :-)

IRV erratically & capriciously swings to the extremes when it ignores your expressed preferences. Almost as scary as Irving's reluctance to count your preferences is IRV's swings to extremes. What kind of candidates will IRV elect when it does that? Let's not find out.

(2) "Let's say that 41% of the voters vote A,B,C, 39% vote C,B,A, 10% vote B,A,C, and 10% vote B,C,A. With IRV A would win. Yet most voters prefer B to A! How can you call this democratic???" First of all, as Note 3 (above) points out, IRV rewards enthusiastic support, while pairwise rewards compromise candidates. In IRV, A wins because A has a combination of ...

I don't know about you, but I don't want Irving to decide whom to support, which candidate to favor, whose support is more important. I want a method that simply counts the preferences that I vote.

In IRV, a wins because A has a combination of the most enthusiastic support, along with enough support from B's supporters to put his/her candidacy over the top. Neither is inherently superior to other. I'm very comfortable with Pairwise, and would be happy to see Pairwise used, and B win in this scenario. I'm also very comfortable for IRV to be used, and A to win in that scenario. Secondly, see Notes 8 & 9 (above).

I included the preceding passage so that I won't be accused of selectively leaving out argument. What I said above applies to the above passage.

(3) "Referring to Response 2, just above, if B were to run directly against A, B would win. Clearly then, B should win the election, refuting the claim that IRV is as democratic as pairwise." You don't know that this would happen in reality. The fact is that the dynamics of a multi-person race are different than the dynamics of two-person races. For example, if voters were more focussed on just candidates A and B, and B came under more scrutiny, then B might lose a lot of votes. A lot of voters might be.

Here Irving is desperate. He seems to be claiming that the preference that you voted isn't valid because you might change your mind and later hold a different preference (!!??).

He says that if you knew more about the candidates, you might not have the preferences that you have, and I guess he thinks that therefore it's ok if he & his count method ignore your preference.

Here Irving, like an fugitive driven into a corner, is trying a bizarre & desperate ploy.

A lot of voters might be putting B as their #2 ranking simply because the main race is A v.s. C, and they are really attacking each other, and B is the "other" candidate. This is the problem with scenarios - they are too simplistic.

Irving is suggesting that some voters will try "offensive order-reversal". But please note that, with IRV, you'll need defensive order-reversal even if no one's using any offensive strategy. IRV itself will often make you sorry if you don't insincerely vote some compromise in 1st place, dumping your favorite.

Offensive order-reversal would be a risky and devious thing to try. It will tend to backfire, and so it will be well-deterred.

But in any case, Irving shouldn't be saying anything about strategy, when his method is the one that often forces you to reverse your preferences & dump your favorite. The criteria at the Criteria webpage at this website specify IRV's failures in that regard, as compared to Condorcet & Approval.

4) "I am a registered Republican, but I lean toward libertarianism. In a pairwise voting system, I would tend to vote for the libertarian first and the Republican second (depending on the actual candidates, of course). However, in an IRV system, I would vote for the Republican first and the Libertarian second. Why? Because in the IRV system, nothing beyond the first choice matters unless no candidate gets a majority. If the Democrat gets a majority, I "wasted" my vote on the Libertarian just like I would in the current system. That is, I failed to help prevent the Democrat from getting the majority." This shows a misunderstanding of how IRV works. In any system, if the Democrat has support from a majority of the people, it wouldn't much matter how you vote, except for registering a protest or a statement. In reality, if IRV was used, and you voted Lib, Rep, Dem, your vote would be registered for the Libertarian, then be transferred to the Republican once the Libertarian was defeated. Exactly what you want.

Wrong. The only way you can be sure that your vote will transfer to the Republican (from your Libertarian favorite) is if your favorite is a sure loser. Because if your favorite has any chance of winning, then he could just as well eliminate the Republican and then lose to the Democrat.

The reason why we want to support our favorite, in 1st place, is that we hope that he might eventually (or now) have a chance of winning. So then, how much sense does it make to offer you a method that makes it safe to vote your favorite in 1st place only if he's a sure loser. That's the kind of offer that could only come from someone who is contemptuous of single-winner reform, someone who is uninformed about it, and someone who is rather contemptuous of you too. Someone who wants to sell you garbage in order to further a goal of his, in which his he hopes that his garbage single-winner method will provide precedence for something else that he wants to get. You, my friend, are being used.

(5) "With IRV, your second ranking has zero meaning, zero power, unless your first choice loses. With pairwise each ranking has power, so pairwise is superior." This is both pairwise's biggest strength and its biggest weakness. The fact is that your second choice could help defeat your first choice, by forcing a circular tie. That is terrible, anad must be dealt with, or pairwise would never work in the real world, because campaigns would ask their supporters to bullet vote (see Note 4, above).

"That is terrible"? It's also untrue. Your vote for your 2nd choice won't "force" a circular tie, or cause one. That was a naive statement from someone who is talking about things that he isn't qualified to talk about.

Now, what _could_ cause a circular tie would be if you dropped from your ranking the candidate who'd otherwise win by beating everyone. But the only way you could gain from doing that would be if that "universal winner" who would have beaten everyone doesn't have a majority against your own candidate, due to a huge amount of indifference.

In other words, it isn't that you can cause a circular tie & suffer because you voted a 2nd choice. All that you could lose thereby would be a possible opportunity to try to _cause_ a circular tie, in an attempted offensive strategy that couldn't work anyway unless that universal winner from whom you'd like to steal the election is regarded with so much indifference that there's no majority ranking him over your own candidate.

I'm sorry if that's wordy or complicated. I wanted to fully answer Irving's claims, since he'll no doubt be using his faulty argument again.

Now, Irving said that Condorcet's practice of actually counting your preferences is a weakness. Does that sound as bizarre to you as it does to me?

However, it doesn't follow that having one advantage means one system is superior to another.

Condorcet & Approval don't just have one advantage over IRV. By every important criterion, IRV is inferior to both Condorcet & Approval. IRV pretty much fails every criterion that's strong enough to actually distinguish between the proposed methods. I'd like Irving to try to name a criterion that IRV meets but which Condorcet doesn't meet.