For the second time in a week or so, fling93 has asked, why IRV (alternative vote) and not Condorcet or approval vote? It’s high time I addressed the question!
It just so happens that I was thinking of this very question earlier today, as I reflected more on the 48th congressional district special election of December 6. It is now apparent to me that even if Marilyn Brewer (the more moderate Republican in the all-party primary, who apparently obtained votes of some Democrats wanting to block both the more conservative Republican, John Campbell, and the extremist Jim Gilchrist) was, in the first round, the “consensus pick,” she would have had a hard time winning under IRV.
Let’s assume that Brewer was the Condorcet winner–defined as the candidate who would have beaten any of the others in a two-way race. If that is the case, then Condorcet’s method (which extracts from the voters’ collective expression of ranked choices a simulation of a series of one-on-one contests) obviously would have revealed that and given the election (in one round of actual voting) to Brewer.
Approval voting might also have done so. Under the approval vote, a voter may vote for as many candidates as he or she “approves” and these votes are not ranked; a vote is a vote, adding to a candidate’s total. The winner is the candidate who obtains the most votes, i.e. is most “approved.”
I very much disapprove of the approval vote! It makes voters’ strategy very unclear. It asks voters to divide their preferences dichotomously: approve vs. disapprove. In a multi-candidate field (which is almost guaranteed by approval vote) it is not at all clear, a priori, where a voter should draw this line to maximize the voter’s chance of seeing his or her most-preferred choice elected.
As for Condorcet, I must confess to knowing too little about its actual mechanics and how a voter might respond to it. I think a flaw that both Condorcet and approval share is the likely favoring of colorless, offend-no-one candidates. While I dislike plurality for encouraging negative campaigning, polarization, and–if there is not strict two-candidate competition–minority-preferred winners, I am no more favorable to a method that encourages candidates to avoid taking any clear stands in order to avoid losing in the simulated one-on-ones (of Condorcet) or driving up his or her own “disapprovals” (in approval voting). Both methods have the further, and closely related, flaw of possibly allowing a relative unknown, who is the first choice of a small minority, to emerge victorious.
Neither Condorcet nor approval has ever been used in a legislative (or executive) election at any level of government anywhere in the world, to my knowledge. IRV, on the other hand, has a long record, for instance in Australian lower-house elections. I think that is a nontrivial point in favor of IRV and against the other two methods.
In the case of IRV and the 48th district, and given Campbell’s 45% of first preferences (taking the primary as a proxy for such preferences), Brewer would have won only with some transferred votes from Gilchrist. Unlike with Condorcet, no second choices from the leading candidate, Campbell, would have been transferred to a trailing candidate like Brewer. IRV transfers second preferences only when the voter’s first-preferred candidate has been eliminated (and transfers lower ranked preferences only when the voters’ higher ranked candidates have been eliminated.) Perhaps this is an argument against IRV. Nonetheless, I am yet to be persuaded that either Condorcet or approval is a good choice for elections to public offices. I hereby place the burden of proof on fling93 (or other advocates) to make a convincing case for Condorcet or approval, rather than IRV, as an option to improve upon the plurality system.
Thanks! I hope I wasn’t being a pest. It’s been an issue that’s been bugging me for a while.
No, I can’t make a convincing case. You, of course, already know that every voting system is flawed (but that plurality is flawed more than almost any other). The main issue I had with IRV is that a centrist candidate can be squeezed by candidates on either side and thus lose in the first round because of a lack of 1st place votes.
So I was a strong fan of Condorcet and Approval (and indeed planned to write posts on them) until I’d read a variation of the “likely favoring of colorless, offend-no-one candidates” argument somewhere else. Ever since then, I haven’t been sure what to support.
I haven’t found too many people blogging about electoral reform, so I really wanted to hear what you thought. Thanks!
Seed planted by fling93 — 07 December 2005 @ 18:58
Oh, I think it was this place where I read that. I recall also they advocate the use of Condorcet within legislatures to select from among competing bills, thus preventing killer amendments and riders (and bills, unlike candidates, can’t hide their views too easily). But also to select chairpeople to center-weight a legislature selected by PR.
Unfortunately, it’s a very badly organized site. And I’m not sure who’s behind it. But a lot of it made sense to me.
Seed planted by fling93 — 07 December 2005 @ 19:02
The major problem with Condorcet is what happens if there is no Condorcet winner. They just don’t have a consensus position on how to resolve the issue. The various solutions that Condorcet advocates do propose are hair-raisingly vague and effectively impossible to count or scrutineer..
Voting demands transparency. The competing methods to resolve a Condorcet paradox are so complex that they defeat the idea of transparency.
Seed planted by Alan — 07 December 2005 @ 19:24
The so-called “center squeeze” issue is a defect of IRV only if you accept the Condorcet definition of “majority rule”, i.e. the candidate who beats all other candidates in pairwise contests. That definition leads directly to a strong bias in favor of the median candidate. Honest advocates of Condorcet understand this and frankly state that it appeals to them because it elects moderates. Other advocates of Condorcet hide behind the (unprovable) assertion that “winner of all pairwise contests” somehow implements “majority rule” better than other election rules.
IRV’s concept of “majority rule” is to count each voter’s ballot for the highest-ranked candidate who still has a chance of winning. That’s more appealing to me (although just as unprovable) as the Condorcet concept–probably because I do not want the median candidate to win almost every election.
IRV proponents describe the “center squeeze” situation as one in which the middle-of-the road candidate doesn’t have the majority support needed to win, but whose supporters do get to choose between the candidates to the right and the left of their own. That’s real power.
Seed planted by Bob Richard — 08 December 2005 @ 09:49
Bob Richard: The so-called “center squeeze†issue is a defect of IRV only if you accept the Condorcet definition of “majority rule…
Since the President is, in my opinion, supposed to represent the entire country and not just the right side or the left side, I believe a centrist candidate is preferable. Of course, the flaw that Prof. Shugart points out is that candidates can easily pretend to be centrists by merely hiding all of their views. And I don’t think there’s an answer to that.
Still, if there were a way to select a true centrist candidate, I believe that would be far preferable than selecting one that is unrepresentative of half of the electorate. And if candidates could not hide their opinions, Condorcet would achieve this far better than IRV.
Bob Richard: IRV proponents describe the “center squeeze†situation as one in which the middle-of-the road candidate doesn’t have the majority support needed to win, but whose supporters do get to choose between the candidates to the right and the left of their own. That’s real power.
I think it’s a bit of a tautology to claim a candidate has “real power” just because they would win in the system that you favor. Plurality proponents could describe winners in their system as having “real power” because the winner always has the most first-place votes.
I think the “center squeeze” situation is more likely to happen when you have a polarized electorate with not too many moderate voters. And I think the centrist candidate would be a *much* more preferable President for this electorate than one selected from one of the two poles.
Just look at our system, which swings from left to right to left to right depending on who wins. Each side undoing the changes that they can do from the other side. This isn’t progress. This is a huge waste of time, effort, and resources.
Alan: The major problem with Condorcet is what happens if there is no Condorcet winner.
In my mind, this only happens when voters act irrationally to create a cycle of preferences, preferring A over B over C over A. Now, voters do act irrationally all the time, but it is impossible to devise a system that handles irrational votes in a fair way.
Also, the various tie-breaking schemes are not vague. They are specific and always result in a clear winner. Some of them satisfy the monotonicity criterion (which IRV fails). And besides, I would think that limiting the field to the Smith set or the Schwartz set would result in a pretty good set to pick from, making the specific tie-breaker chosen not that important.
Yes, the complexity is an issue. You’d need computer counting. But transparency could be achieved by making the ballots (which should be both human-readable and computer-scannable, but that’s a separate issue) and the computerized vote totals available for third-party verification using their own software (and undoubtedly, an open-source solution would become available).
Getting the electorate to understand the system is the tougher problem, but I think the “round-robin tournament” analogy could go a long way towards that. But this problem isn’t worth solving unless we can fix the “colorless, offend-no-one candidates” problem first, and I don’t know that it’s fixable.
Seed planted by fling93 — 08 December 2005 @ 13:03
fling93: “I think it’s a bit of a tautology to claim a candidate has ‘real power’ just because they would win in the system that you favor.” I’m sorry that I didn’t make my point clear. In the center squeeze election we’re talking about, the centrist candidate would lose (not win) under IRV, but his/her supporters would nonetheless be positioned to make the choice between the candidates on either side.
fling93: “In my mind, this [no Condorcet winner] only happens when voters act irrationally to create a cycle of preferences, preferring A over B over C over A.” The individual voters don’t have to be irrational at all. Cycles can happen in the vote totals even when each individual voter’s preferences are transitive (A over B and B over C imply A over C). This is the “paradox of voting”. I believe that cycles are more likely to happen when there are multiple issue dimensions rather than one continuum from right to left.
Seed planted by Bob Richard — 08 December 2005 @ 15:28
Bob Richard: the centrist candidate would lose (not win) under IRV, but his/her supporters would nonetheless be positioned to make the choice between the candidates on either side.
That’s also true the other way around. The right-wing candidate would lose under Condorcet, but his/her supporters would be positioned to make the choice between the centrist and the left-wing candidate. Ditto for the left-wing candidate’s supporters. I think most voting systems with ranked-choice ballots have this feature.
Bob Richard: I believe that cycles are more likely to happen when there are multiple issue dimensions rather than one continuum from right to left.
I actually think the number of dimensions in the political spectrum has little to do with it, because you can still find the center of a two-dimensional or three-dimensional political field.
I think it’s more likely to happen when candidates stake out conflicting political beliefs, so voters have to prioritize and compromise. But this is less likely to happen in a multi-party election, because I think it’s mostly an artifact of a two-party system where each candidate knows they have only one alternative to beat.
I’m by no means an expert in Political Science, though. Just a voting-theory geek (who might end up studying PoliSci).
Seed planted by fling93 — 08 December 2005 @ 15:56
I think you’ll find that Condorcet cycles are extraordinarily common and the principal reason Condorcet has never been used for any governmental election. There is actually no way that a Condorcet voter can vote their true preferences and at the same time vote to avoid a Condorcet paradox.
The prospect of the media launching into a learned discussion of whether Candidate A is likely to achieve the Smith set criterion or Schwartz set criterion is somewhat less than zero, and certainly less than the likelihood of any given election resulting in a paradox.
There is simply no way to justify an election method where you need to launch into set theory to explain why Candidate A was elected.
Seed planted by Alan — 10 December 2005 @ 22:40
Although I disagree with “no way that a Condorcet voter can vote their true preferences and…avoid a Condorcet paradox,” I don’t have good responses to the other points, and I think I’ve already spent too much time arguing what is probably a moot point anyway.
Seed planted by fling93 — 12 December 2005 @ 13:57
a) I can’t imagine why you’d expect cycles to be common; certainly the academic work I’ve seen estimates that they would happen in less than 0.1% of races where there are more than two contenders, and of course having more than two real contenders in the first place is still fairly uncommon.
b) The idea that there’s no satisfactory way to resolve cycles is mistaken as well. Schwartz Sequential Dropping, aka the Schulze Method, has been used for years by the Open Source community. There are still certainly arguments over what is the BEST method for resolving cycles, but it’s really not a huge deal — in case of a cycle, IMHO, all candidates have a decent claim to victory, and a random method like drawing straws would be fine.
c) The idea that Approval and Condorcet favor “colorless” candidates is countered by simple observation — in 2000, in a race between Bush, Gore, and McCain, McCain would’ve won Condorcet or Approval quite handily, precisely because he was NOT colorless — he was seen as an idiosyncratic, plain-spoken man who fought for his ideas. This kind of “champion” candidate, who may have trouble surviving partisan primaries, would do BETTER under Approval or Condorcet.
d) For IRV’s effects on campaigning, and what kinds of candidates would be attracted, you might check my long post on this topic.
Seed planted by RM 'Auros' Harman — 18 November 2006 @ 18:46
‘Auros’: regarding your point ‘c’–that “simple observation” reveals that Approval and Condorcet would not produce “colorless” candidates, as I had sugegsted–the problem with “simple observation” is, of course, that it is, well, simple.
We can’t assume that the candidates we had in the actual nomination and general-election stages in 2000 would also be the candidates under approval, Condorcet, or other rules. I agree that McCain was a potential Condorcet winner (assuming, as discussed above by Bob, that that means the candidate at the median) among the three candidates from 2000 that you mention, and holding constant the campaign strategies these three men actually used in 2000. However, there were many other candidates at variuous points that year, and possibly still others who did not run who might have entered under different rules. And, with different rules and different candidates, the strategies each employed also would have been different.
Simple observation, in which only one variable is allowed to vary, simply won’t get us very far in trying to predict outcomes in a nationwide election held under rules that have never actually been employed in any large electorate.
On the possibility of cycles, I defer to some of the previous propagators of this planting, who already addressed the issue with respect to Condorcet.
Seed planted by MSS — 19 November 2006 @ 11:16
Your claim about “colorless” candidates from Approval Voting is flawed, for the same reason as your point about the “difficulty” of picking an approval threshold. It is up to each voter to decide where he wants to draw that line. If he approves only the candidates he would give a “7 or above” to, then he’s maximizing his chances of getting such a candidate. But if he approves everyone he likes more than average, he improves his chances of getting such a candidate – even though he lowers the odds that the winner will be outstanding. It’s up to each voter to decide that compromise for himself – that’s democracy.
What we know about Approval Voting is that it has a much higher social utility efficiency than IRV, meaning that it statistically increases your expected satisfaction from elections, as compared to IRV. So it’s better at giving voters whatever they say they want. So if you do not want “colorless” candidates, you are free to approve only the candidates who you perceive to be colorful, who get fired up on the issues. If other people also like those colorful candidates enough, they’ll win. If people choose to use a low threshold, because they’d rather have a good chance of getting a tolerable candidate than a 50/50 chance to get a really great or really awful candidate, that is their choice. That is democracy. If they like colorless candidates, that’s what they get. I actually think a lot of people feel the opposite way. A lot of people I know, for instance, like Ron Paul _and_ his Democratic counterpart Mike Gravel because they are both so far from mainstream, despite having diametrically opposed views on most economic issues (like Medicare). So I think you underestimate people, and disregard fundamentally crucial issues like social utility efficiency.
Seed planted by Clay Shentrup — 09 June 2007 @ 19:00
Clay, “social utility efficiency” is your Center’s computer model. It also puts range voting ahead of IRV, according to the graffiti you’ve left on pretty much any blog or news article to mention IRV and end up indexed by Google.
What leads me to seriously doubt Warren’s operationalization of “social utility” is this. Range voting still beats IRV when voters are voting strategically.
That is, when a vast minority of voters overrules the majority of voters.
As I said today on FairVote’s blog… if getting third parties elected is what you want, you’d be much better off helping FairVote, CfER and others advocate for PR. Running around trashing IRV doesn’t accomplish anything.
Especially because the sheer insanity of range voting means it won’t be used for public elections anywhere ever.
Seed planted by Jack — 11 June 2007 @ 18:32
Jack,
The computer models used to calculate social utility efficiency have been quite rigorous, testing over 60 voting methods over the range of 0% strategists to 100% strategists, ignorant voters to informed voters, few candidates or many candidates, etc. In *all* 720 combinations of the 5 fundamental parameters used, averaged over billions of simulated elections, Range Voting beat out all the other common voting methods. IRV comes in second-to-last, right behind plurality voting. You can get the source code to the software at RangeVoting.org and analyze it for flaws if you like. We’d welcome the chance to improve it even more.
> What leads me to seriously doubt Warren’s operationalization of “social utility†is this. Range voting still beats IRV when voters are voting strategically.
Yes, the benefit of Range over IRV _doubles_ if voters go from 100% honest to 100% strategic. This makes perfect sense, since Range Voting strategically degrades toward Approval Voting – which is still a great voting method – whereas IRV strategically degrades to plurality voting.
This is discussed in some detail here:
http://rangevoting.org/TarrIrv.html
http://rangevoting.org/RVstrat2.html
> As I said today on FairVote’s blog… if getting third parties elected is what you want, you’d be much better off helping FairVote, CfER and others advocate for PR. Running around trashing IRV doesn’t accomplish anything.
Incorrect. IRV leads to two-party duopoly, which doesn’t help us to get proportional representation. But Range and Approval Voting make it quite feasible to break out of duopoly, which _will_ help us get PR. And a vast number of posts (e.g. mayor, Senate, governor, President) are single-winner, and will be two-party dominated with IRV. But RV/AV give us a chance to escape that conundrum.
Please read this:
http://RangeVoting.org/PropRep.html
Once you understand why IRV is so harmful to democratic reform, maybe you’ll join the choir and start helping to educate the public away from its pitfalls as well.
> Especially because the sheer insanity of range voting means it won’t be used for public elections anywhere ever.
Range Voting is simpler/cheaper than IRV, reduces spoiled ballots (whereas IRV massively increases them), reduces the risk of ties (whereas IRV increases them), and can be done on all standard voting machines without the expensive upgrade costs of IRV. If anything is insane, it’s IRV. And I’m a huge proponent of Approval Voting – which is CERTAINLY far simpler than either, and still has all the good fundamental properties of Range Voting, and is almost as good as Range Voting (and much better than IRV).
Your objections are highly typical of the standard IRV advocates, full of the misconceptions in the form of talking points. I urge you to look at the facts scientifically and reassess your position.
Regards,
Clay
Seed planted by Clay Shentrup — 13 June 2007 @ 06:15
There is a very large literature on electoral systems, not only that by political scientists, but also by mathematicians. (Strangely, the lits don’t always address one another, although there are distinguished scholars who publish in both fields.)
Approval voting, of course, has its advocates among the experts in electoral systems. However, despite its having been introduced almost thirty years ago (by Brams and Fishburn in APSR, 1978), it has not been used or even seriously proposed for any public jurisdiction during this time that I am aware of. There are good reasons for that, some of which Jack or I have already mentioned or that you can locate in the literature. Range voting, as far as I know, is advocated by no one not affiliated with the organization linked by a now-frequent propagator above. A search of the academic articles data bases in political science, economics, mathematics, and computer science turns up exactly zero articles on range voting. Until the method has been subjected to independent scholarly scrutiny, it simply is not worth being taken seriously–at least here on a blog devoted to analysis of actual elections and to the discussion of scientifically sound alternative electoral systems for actual democracies. It does have its uses in other settings and, in fact, I oversaw the implementation of a range-voting system for a committee that selected some student award winners this past year.
A very recent article on approval voting (which has drawn considerable scholarly attention over the past thirty years) in one of the top journals of the discipline comes with the following interesting abstract:
(Finally, I want to note that the posted comments advocating range voting have been triggering my spam filter because of some aspect of their formatting. I will not promise to continue retrieving them from the virtual orchard’s compost heap.)
Seed planted by MSS — 13 June 2007 @ 14:23
MSS: A search of the academic articles data bases in political science, economics, mathematics, and computer science turns up exactly zero articles on range voting.
Range voting is discussed in the academic literature here: Nicolaus Tideman, Collective Decisions and Voting: The Potential for Public Choice (Ashgate Publishing, 2006): pp. 174-176 and 236-238. For what it’s worth, Tideman rates it “unsupportable” (his word), and rates IRV best among the methods that do not require heavy-duty algebra.
This looks like a very good book for those who are inclined toward a mathematics/social choice theory approach rather than a political science approach. As MSS notes, these traditions don’t engage each other much. I would love to have a thread about the causes and consequences of that disengagement.
Seed planted by Bob Richard — 15 June 2007 @ 22:41
Clay,
Two brief points for you.
1) You mention: “IRV leads to two-party duopoly, which doesn’t help us to get proportional representation.”
This is a presidential democracy with directly elected governors at the federal level… a country of single-winner elections. Bashing IRV and STV, two real-life-tested voting systems, will not get us to PR. And I implore you exercise some realism and revisit Alan and Matt’s point that elites change voting systems to accommodate new party alignments. The contrary is counterintuitive.
2) I struggle with a measure of public happiness that holds a wrong-winner election, or a plurality reversal, is a mass-desirable good. I’m sure the source code in Warren’s computer models is well vetted. My problem is with the models’ fundamental assumptions about voter opinion.
Seed planted by Jack — 18 June 2007 @ 22:10
Jack and Clay,
Another fundamental problem with Warren Smith’s computer models — and with range voting in general — is whether it is meaningful to aggregate the “utilities” of individuals based on cardinal measurement. Kenneth Arrow explicitly rejected this approach (Social Choice and Individual Values, 2nd ed., pages 9-11) and decided to restrict his work to aggregation based on ordinal measurement. I would expect to find that there is still some disagreement about this among economists (Jeremy Bentham probably still has a few disciples), but I would also expect to find that the dominant view is that Arrow was right.
Anyone who wants to replace the principle of majority rule with the principle of maximizing the sum of individual “utilities” needs to do some homework in the field of welfare economics and report back on why the objections of Arrow and many others are invalid.
Seed planted by Bob Richard — 19 June 2007 @ 13:32
Touche, Bob. What is the difference between ordinal and cardinal measurement of aggregated individual utilities?
I sense more blogging on the horizon.
Seed planted by Jack — 19 June 2007 @ 21:59
The nice thing about IRV is that it absolutely can never elect a Condorcet loser, which would be an improvement over the US situation today.
The bad things are
1. the potential for strategy – If Royal’s supporters in the recent French elections had betrayed her and supported (the condorcet winner) Bayrou instead, they wouldn’t be stuck with Sarkozy today. Favorite betrayal is a particularly ugly strategy which is necessary in Plurality and IRV, unnecessary in Condorcet, approval and range.
2. also the centralized counting. A little algebra is far better than a hundred million ballots which must be gone over several times in a central location.
3. Non-monotonicity. It may sound innocent, but if a candidate won on a nonmonotonicity distortion, I bet neither he nor the shining new electoral system would have lasted very long.
Seed planted by Harald Korneliussen — 06 August 2007 @ 06:54