When reformers start to build momentum to pass electoral reform, they often attract a Voting Theorist who comes out to argue for some other method. He (it’s often a he) uses technical jargon and mathematical theorems and points to Burlington Vermont’s 2009 mayoral election to prove reformers should support his preferred method and oppose Instant Runoff Voting (IRV). His preferred method could be Approval Voting, Score Voting, or Score Runoff Voting. (If you are lucky your campaign might attract one of each and they will argue with each other!)
He makes some interesting points, but also some convenient exaggerations and he cares more about math and theory than about human behavior and empirical evidence. Here are seven claims that advocates for Score Voting, Score Runoff Voting, and Approval Voting make about Burlington, Vermont in particular and IRV more generally, and responses.*
Score and Score Runoff proponents claim the Burlington election proves IRV is inherently flawed because it did not elect the Condorcet Winner—the candidate who would beat all other candidates in a head-to-head contest.
Response 1: True. But the Voting Theorist’s method doesn’t guarantee the Condorcet Winner either and he’s OK with that
IRV fails the Condorcet criterion, and in Burlington it elected a non-Condorcet winner. But Score, Score Runoff, and Approval Voting all also fail the Condorcet criterion, so they could elect non-Condorcet winner as often—or more often—than IRV. If the Voting Theorist actually cared about guaranteeing the Condorcet winner, he could push for one of the methods that does that. But he doesn’t believe the Condorcet Winner is necessarily the “right” winner. (With good reason—Condorcet does not necessarily mean most support.) His objection that IRV fails this criterion falls a bit flat once you realize he actually doesn’t believe it is a critical criterion.
IRV has proven, in the real world, to elect the Condorcet winner much more often than plurality and runoff voting—the two systems in widespread use in the United States and Canada.
Score Runoff Voting advocates point to one study to support their claim that IRV will create non-representative results in about 20 percent of close elections.
At first blush, this claim may read as 20 percent of all IRV elections will fail to elect the Condorcet Winner. But, no, the study found that it could happen in 20 percent of close elections, where three candidates each win about one-third of the vote. The vast majority of elections do not have three evenly matched candidates, so the occurrence rate in all elections is much lower. Second, the study counts cases where there was no Condorcet Winner—these cases are like games of rock-paper-scissors where no candidate can beat all others. But in these cases no method could elect the Condorcet Winner because there is not one.
In the real world, more than 99 percent of IRV elections pick the Condorcet Winner. In the 140 American elections have used IRV and released full data (thousands of elections have used IRV, but Condorcet failures are impossible to identify based on election results; you have to scrutinize the full data to find them.). Of the 140 elections with full data, all but one (Burlington in 2009—that’s why all the Theorists talk about it) elected the Condorcet Winner. For more information about what we can learn from Burlington—and what we can’t—see this detailed article on the topic.
That’s a real world occurrence rate of well less than 1 percent. If we eliminate the elections where a failure was impossible because there were only two candidates and look just at the 84 IRV races with three or more contenders, now the real world occurrence is just over one percent.
So yes, if more American and Canadian cities adopt IRV, some tiny fraction of elections will elect a non-Condorcet winner (although some of the those will be because there is no Condorcet winner). Probably around one percent of the time. And, as the below scenarios suggest, the Theorists’ methods might produce a non-Condorcet winner more often.
Claim 3: Some other untested voting method would likely have elected the Condorcet Winner in Burlington
Score Voting advocates claim it is “likely” that Score and Approval would nonetheless have elected the Condorcet winner in Burlington. Score Runoff Voting advocates believe it is “very likely” that the Condorcet winner would have won. Or if he did not win, they will argue that with Score Voting, the voters would have felt the outcome was fair (but the same voters would have felt the same outcome was unfair under IRV).
Approval, Score, and Score Runoff Voting all fail the “Later No Harm” criterion, meaning giving an honest score to a less-preferred candidate could cause your favorite to lose. In elections with three strong candidates, many voters will score or approve only their favorite rather than cause their favorite to lose. If voters try to help their favorite win, Score and Approval methods will not only not elect a Condorcet winner, Approval and Score might do what plurality voting does and elect the least representative candidate, the Condorcet Loser—the one who would lose to both other candidate in a head-to-head. This is the worst possible outcome.
Instant Runoff Voting and other runoff methods (including Score Runoff), will never elect the Condorcet Loser.
None of the three methods the Theorists like have been used in a public election, so we don’t really know how voters and candidates would respond. But Approval Voting has been used in professional organizations and student body elections, so we know that in those settings most voters vote for just one candidate. Using the data from Burlington’s elections and some estimates about how campaigns and voters would use the three theoretical methods, here’s which candidate they might have elected in Burlington.
Approval Voting—most likely Republican Wright, possibly Progressive Kiss
In Approval Voting you can vote for every candidate you approve of. If Burlington voters gave an approval vote to every candidate they gave a rank to in the IRV election, Democrat Montroll would have won by a landslide 6,589 votes, Kiss would have trailed with 5,694 and Wright in last place amongst the top three candidates with 5,380.
But both the Republican and Progressive campaigns in Burlington would presumably have recognized this obvious threat from the center candidate and alerted their voters that, if they wanted their favorite to win, they should not approve the Democrat. If all voters had only approved their favorite of the top three contenders (they could also have safely approved of the Independent and Green candidates), Republican Wright would have won handily, 2,945 to Kiss’s 2585 and Montroll’s 2,045. Republican Wright was the Condorcet Loser—the candidate who would have lost to every other candidate in a head-to-head. So to the extent theorists believe that electing a non-Condorcet Winner like Kiss is a fail, electing Condorcet Loser Wright should be a catastrophe.
In student body and professional organizations that have used Approval Voting, often 80 to 100 percent of voters “bullet vote”—only approve of one candidate. If that pattern held in Burlington, Wright would have won. Even if Burlington voters were more generous with their approvals and only 70 percent bullet voted, Wright would still have won.
If Republican and Progressive voters voted only for their favorite, but all Montroll voters approved both him and their second choice, then Progressive Kiss would have won.
Score Voting—most likely Republican Wright
In Score Voting you can give each candidate a score, for example between one and nine. In Burlington, there were five candidates in the race and some voters ranked all five. If voters had given their first-ranked candidate a 9, their second choice a 6, third a 4, fourth a 2, and fifth a 1, Democrat Montroll would have won with a score of 41,020 to Kiss’s 37,515 and Wright’s 37,504.
But as in Approval Voting, the campaigns would have likely have recognized this threat (again, it’s rather obvious) and encouraged their voters to give low or no scores to all but their favorite. In fact, nearly 40 percent of Republican voters did not give either of the other two candidates a rank, so telling Republican voters not to score the other candidates would have been an easy sell. Even if some voters gave scores to all candidates so the average scores were just lower—say an average of 9, 4, 3, 2, 1—Republican Wright would have won.
Score Runoff Voting—likely Progressive Kiss, possibly Democrat Montroll
In Score Runoff Voting, as in Score Voting, you can give each candidate a score. But then the two candidates with the top total scores advance to an automatic runoff. Score Runoff advocates prefer a scale of zero to five, so let’s say that voters’ honest scores were to translate their actual rankings for all five candidates into a 5 for their favorite, 4 for their second, 3 for third, 2 for fourth, and 1 for fifth. Democrat Montroll and Progressive Kiss would have gone to the instant runoff where Montroll would have won.
Again, the campaigns would have understood the threat and likely let voters know they should give their favorite a 5 and their back-up a 1. This would minimize the chance their back-up would bump their favorite out of the runoff, but still give themselves a vote in the instant runoff in case their favorite didn’t make it. If everyone voted this way, Progressive Kiss would have won. Just like in IRV. Even if about half the voters didn’t get the strategy—say, that, within the top three contenders, voters averaged a 5 for their favorite, a 1.5 to their second-choice and 0.6 to their third—Kiss would have won.
Or say Republican and Democratic voters gave their second-choice an average of 2, but the Democrat didn’t spread the strategy since he was polling in third place, so his voters gave an average score of 3 to their second choice. Kiss would have won—the same as IRV.
Find this article interesting? Support more research like this with a gift!
Or say polling showed voters that the Democrat would clearly beat the other candidates in a head-to-head. Republican and Progressive voters would realize their best bet was to keep the Democrat out of the runoff. They could strategically give the Democrat a low score and give their opposition candidate a high score, in hopes of sending the weaker opponent to the runoff where their favorite could defeat him. If even one-third of voters pursued this strategy, Wright and Kiss would have gone to the runoff and Kiss would have won.
Voting Theorists claim that Score Voting, which has never been used in an election, “appears to inspire honesty to a surprisingly large degree.” Some Score Runoff Voting Theorists confidently claim that with Score Runoff (a.k.a. Star) Voting it is always safe to vote honestly because strategic voting doesn’t work.
Sorry, strategic voting can work in every system. As the above scenarios show, strategic voting would have worked in Burlington in all the proposed methods.
Although honest voting was clearly the best strategy for 83 percent of Burlington IRV voters, critics focus on the 17 percent of Burlington voters whose favorite was Republican Wright and whose least-favorite was Progressive Kiss. Had these voters somehow known the Republican (who was polling first place) would lose, they could have strategically ranked the Democrat first to send him to the runoff instead of the Republican and at least keep the Progressive out of office. But the same strategy would have worked under Score Runoff Voting. Republican voters could have strategically inflated the Democrat’s score and also strategically given a low or no score to the Republican to boost the Democrat into the runoff where he could beat the Progressive.
Score, Score Runoff, and Approval Voting advocates all say Burlington proves that IRV maintains the spoiler effect, where a candidate wins with minority support because the majority of voters split between two other candidates.
IRV eliminates what most people think of as the spoiler effect, where a minor party candidate splits votes with a similar major party candidate, causing the other major party candidate to win with less than a majority. The 2000 American presidential election between Bush, Gore, Nader is the classic example. Oregonians may think of the 1990 Gubernatorial election where a conservative Independent split votes with Republican David Frohnmayer, allowing Democrat Barbara Roberts to win with less than 46 percent.
The mathematical definition is the independence of clones criterion: the winner does not change due to the addition of a non-winning candidate who is similar to a candidate already present. IRV passes this criterion: if Gore was going to beat Bush, Nader joining the race wouldn’t have changed that with IRV.
But IRV critics are implicitly re-defining “spoiler effect” to mean: the winner does not change due to the addition of a non-winning candidate even one dissimilar to the winner. By this definition, critics are correct that the Republican “spoiled” the election for the Democrat in Burlington.
But a Republican acting as a supposed spoiler to a Democrat was not a real problem since the Progressive winning was still representative of the majority of voters who leaned left. The similar-candidate spoiler is worrisome because it leads to an unrepresentative result.
Score Runoff advocates say Burlington had an unrepresentative outcome because not all votes were equal—voters for Republican Wright and Progressive Kiss only had their vote count for one candidate, whereas voters for Democrat Montroll got to vote for him, and then for someone else once Montroll was eliminated.
Wrong. IRV counts votes in rounds, just like a primary and a general. If your favorite advances you vote for him in the first round (primary) and in the next round (general), but if your favorite is eliminated after the primary, you have the chance to choose between the candidates in the next round. Everyone gets one vote per round, although some voters just get to vote for their favorite twice, just as they do in primary and general elections now. Two stage runoffs do, in practice, give voters who vote in the primary two bites at the apple while voters who only vote in the general only get to vote in that one round. By letting voters rank their preferences on a single ballot, IRV ensures that everyone who votes in that single election gets an equal voice.
Take note: this is actually a problem for Score voting methods which do let some votes count more than others, because you can give a candidate a score of, say, 8, which will count for twice as many points as another voter’s 4.
Claim 7: If minor parties would just subsume themselves into the two major parties, plurality would work fine
Anthony Gierzynski, the lead author of the Score Voting critique of the Burlington election, actually prefers plurality voting. He appears to have paraphrased the guy in the XKCD cartoon below and said “I prefer Plurality Voting, but since Burlington seriously did Instant Runoff, then I’ll argue for Score Voting instead.” He thinks IRV’s flaw is that it allows the Progressive Party to run candidates who compete with Democratic candidates. Plurality voting should force Progressives to become a part of the Democratic Party and then Democrats like Montroll could win. I’m sure you’ll be surprised to learn that Gierzynski was the Democratic candidate for the Vermont legislature in 2002 and lost to—guess who?—Progressive Bob Kiss.
If Americans were happy with the Democratic and Republican candidates, they would just vote for them and this whole argument would be moot. But Americans are not happy—close to half don’t affiliate with either party. Candidates outside the two major parties, or on the wings of the major parties, can broaden the number of issues and solutions that are discussed. IRV can enable more voters to cast a vote for a candidate they are genuinely excited about.
IRV has largely withstood the tests of real elections.
I hope, my dear fellow nerds who have read this far, it is clear I am not saying IRV is perfect. Indeed, no single-winner system can overcome all forms of strategy nor guarantee the “right” winner, especially because there sometimes isn’t one (recall rock-paper-scissors).
But the claims about Burlington and IRV all have something in common: they are attacking a reality in favor of a theory. After hundreds of public IRV elections, Burlington happened and now critics can point to it. But after hundreds of Score or Score Runoff or Approval Voting elections, would one of them elect a non-Condorcet winner? Probably yes. But no public elections have happened, so the Voting Theorist can blithely claim his method would “likely” elect the Condorcet winner and “always” encourage honest voting. IRV is like a real building that has been through decades of use and stood up to the tests of weather and gravity. It’s easy to point out the tiny cracks. Untested methods are like architectural drawings, perfect in every colored pencil detail, but who knows what will happen when they exist outside of imagination and paper.
Maybe some of those theories will work out in the real world. But in the meantime, don’t let the Voting Theorists derail real-world reform.
*Note that IRV advocates now prefer the term Ranked-Choice Voting (RCV) and Score Runoff advocates now prefer the term Star Voting, but I use IRV and Score Runoff throughout for clarity.
I definitely agree that we shouldn’t kill momentum, and people probably shouldn’t vote FOR Plurality just because the only other option is IRV. Still, I think many other methods show much more promise, and would like to see them thoroughly tested, which we aren’t getting much of right now.
1) It’s true that many supporters of Cardinal voting systems (Approval/Score/SRV) don’t care about the Condorcet criterion, because they don’t believe that that Condorcet winner is always the best, nor do they believe that the Condorcet loser is never the best. But IIRC some of the things IRV supporters say makes it sound like they would really want a Condorcet method, or even think IRV guarantees them that. But even if that’s what they want, they still shouldn’t use IRV, they should use a Condorcet method.
2) So there have only been 140 elections that released full data? And in 40% of those, there were only two candidates? Of the 84 with three or more “contenders”, how many were strong? Were there still practically only two strong candidates in most of them (other than Burlington)? Did you already account for that in those numbers?
A fear of many of these theorists is that IRV still devolves into two-party domination. If this is true, then most of the time there will be a Condorcet winner, because most of the time there will only be two strong candidates.
3) I know a point among some Cardinal supporters is that, under certain assumptions about voter strategy (which they assert is “reasonable”, although it also works if everyone is perfectly strategic), the Condorcet winner will always win when they exist. And since Condorcet methods aren’t immune to strategy even when there is an honest Condorcet winner, Cardinal methods might elect Condorcet winners more often even then true Condorcet methods. Very interesting to us nerds.
But even if this is true this might not necessarily sound great, since the Condorcet winner might not always be the best. With more honesty, someone even better might win. So if a Cardinal method fails to elect the Condorcet winner, it’s not necessarily a bad thing. As long as there was a good reason for it, which IRV might not give.
IRV satisfies Later-No-Harm, but this is incompatible with and therefore must fail Favorite Betrayal (and the Condorcet criterion, for those who care).
As mentioned, it’s disputed that the Condorcet Loser is the “least representative candidate”, it might even be the best. Of course Plurality could elect the Condorcet loser for no good reason, while Cardinal methods should hopefully only elect them if there is a good reason.
By the way, I think you have a typo in Response 3, where it says “Instant Runoff Voting and other runoff methods (including Score Voting), will never elect the Condorcet Loser”. I think you means “Score Runoff”, not simply “Score”.
4) The author of the page in Claim 4 could probably have chosen their words better. Their image clearly shows that even SRV doesn’t not reach the maximum score under “honesty”, while a higher score is given later, so presumably they knew and were in some sense admitting that it’s not “always safe to vote honestly”. They could have made that clearer.
5) I’m having trouble following the Independence of clones criterion. It seems to be saying that clones are candidates in a subset who rank each other before ranking anyone outside the set? So if Alice and Bob are clones in an election which also includes Zelda, everyone who puts Alice first will put Bob second, and everyone who puts Bob first will put Alice second? I suppose this would also include later ranks, i.e. clones are always adjacent to each other in any voter’s rankings?
If I’m right, then the Burlington example was not a clone scenario. But you can be most similar to another candidate without being a clone of them. Republicans may have essentially all put Democrats adjacent (so they were most similar to the Democrat), but Democrats did not all put Republicans adjacent. In fact, I think Burlington still more-or-less matches your definition of the spoiler effect, where a non-winning candidate splits votes with the candidate most similar to them, causing both to lose when the other could have won.
But what’s in a name? If IRV still encourages two-party dominated, this effect could be the reason, and would be bad by any name.
6) If I understand correctly SRV’s criticism is fairly specific, it involves any single ballot being able to be negated by another single ballot. To put it in perspective, even Plurality doesn’t meet this.
7) I have actually seen an American Libertarian say something along the lines of Plurality being better than IRV (though IIRC he still preferred Cardinal methods to both). His argument was that IRV makes it harder for minor parties to be a spoiler without increasing their chances of winning, thereby reducing the minor parties “blackmail power” (like, “if at least one of you doesn’t at least throw us a bone, we’ll sabotage you”). Of course, even if this is true, IRV would still be better than Plurality for anyone who prefers a major party.
The problem, I think, is that while many of these theorists acknowledge that IRV has been tested, but they consider IRV to have “failed” the test. They want to test something new, something that they think will be better.
But by all means, don’t let them (or me) kill your momentum!
Cody, I agree with what you said but I would like to add even more detail to your arguments.
1) Exactly, and one of the reasons it sounds like IRV supporters want a Condorcet method, is because they usually use talking points like “candidates should win by majority, not by plurality.” The reason they say this is because after you eliminate enough candidates, there is always a winner that is preferred by a majority of voters to all of the other candidates in the small subset of remaining candidates. But saying that the winning IRV candidate won by a majority is looking at the election with rose tinted glasses because IRV winners only win a majority when you pretend that many candidates in the race don’t exist. Under the same logic, you could argue that IRV always elects a candidate by unanimity by modifying the definition of IRV from eliminating candidates until one candidate has a “majority” to eliminating candidates until one candidate is every voter’s “1st choice”. The definition of winning by a majority is not that a candidate is preferred by a majority of voters in one small subset of candidates, but that they win by majority when you include all the candidates. If you care about which candidate wins by majority in a small subset of candidates, then you should care about electing the Condorcet winner, because the Condorcet winner is the candidate that wins by majority in every subset of candidates in which there are only two candidates.
2) Your point was that the elections in the bay area were not competitive enough to observe the effects IRV has on competitive elections. This is a huge understatement. Out of the 138 elections in the bay area, while 100% of them elected the Condorcet winner, in 94.9% of them, the plurality winner happened to be the Condorcet winner. So a different way of looking at this data, is that the results of 94.9% of IRV elections elect the same person in plurality voting as they do in plurality voting. However that would be an unfair claim to make because this data does not include elections that are competitive enough to see the “glitches” in both plurality voting, and IRV.
And finally, you respond to Kristen’s final point by making this counterpoint: “The problem, I think, is that while many of these theorists acknowledge that IRV has been tested, but they consider IRV to have “failed” the test. They want to test something new, something that they think will be better.”
Voting theorists consider IRV to have failed this test because has failed this test. Australian analysts even list of promoting a 2 party system as one of the disadvantages of IRV: http://www.rangevoting.org/AustralianPol.html#indepsimpler. Here is a pie chart of the seats won during Austraila’s federal and state seats under IRV: http://www.rangevoting.org/AustralianIRVseats.png. But what is even more impotent about a voting system is how voters feel about it. So how do Australians feel about IRV? Well, there are 20% MORE Australians who want to have plurality elections then continue to use IRV in their house of representatives (57% plurality, 37% IRV, 7% unsure): http://rangevoting.org/AusIRV.html
Correction to a small typo I made. I said:
So a different way of looking at this data, is that the results of 94.9% of IRV elections elect the same person in plurality voting as they do in plurality voting.
Here, I meant to say:
So a different way of looking at this data, is that the results of 94.9% of IRV elections elect the same person in instant runoff voting as they would have in plurality voting.
Thank you, great article!
Anthony Gierzynski, the lead author of the Score Voting critique of the Burlington election, actually prefers plurality voting. (and the text “prefers to plurality voting” linked to this: https://www.uvm.edu/~vlrs/IRVassessment.pdf#page=4)
I am guessing you are referring to the quote “IRV proponents claim that despite these problems IRV works better than the standard plurality system with a runoff” on page 4 of the article because that is the page of the article that your link directs you to. If this is the quote that you are referring to then note that he is not comparing IRV to plain plurality voting, but he is comparing it to plurality with a RUNOFF. Thus it is untrue to claim that Anthony Gierzynski is a fan of plurality voting. Here is a good article comparing IRV to a delayed runoff: http://scorevoting.net/HonestRunoff.html
He appears to have paraphrased the guy in the XKCD cartoon below and said “I prefer Plurality Voting, but since Burlington seriously did Instant Runoff, then I’ll argue for Score Voting instead.” He thinks IRV’s flaw is that it allows the Progressive Party to run candidates who compete with Democratic candidates.
Where did he say this? Can you provide me with a link?
The power of money in politics comes from undemocratic voting systems, as partially explained by Warren D. Smith in . IRV is undemocratic. It shares the principal problem of FPTP, the unequal distribution of political power.
As for your assertion that Score voting is untested, it is tested by every bee colony and seems to have survived the rigors of Darwinian selection.
Of course I join the other commentators in agreement that none of us should let anything kill our enthusiasm for real correction of the voting systems in use in the US so as to defeat the political rule of capital.
I’d support any of SRV/Score/Approval, but we need to drop this stupid bee argument. Bees don’t vote. They aren’t people. Their dancing has about as much to do with voting as the length of a peacock’s tail.
Let’s focus on human examples like product/hotel/restaurant ratings, Olympics scores, psychometrics and Likert scales. Humans have been rating things for a very long time, and the meaningfulness of those ratings were never much disputed, until the same concept was applied to political candidates. But it’s just as valid here as it is there.
Bees do indeed vote.
I’ll chime in too. I’m mostly an elapsed election science nerd but here is what I’ve written on IRV and single winner election methods. TL;DR: To me IRV is completely unpalatable to me and about equally bad as plurality.
IRV is a terrible way to elect politicians. This is where we are in agreement. But you linked to a petition to repeal IRV in North Carolina however this petition does not mention other voting methods, so when you sign this petition, you are effectively saying that you would rather have plurality voting then IRV, and that is where I disagree with you.
I am not a supporter of IRV, but because it passes the independence of irrelevant clones criterion, it is better then plurality.
Many Bayesian Regret simulations have IRV tied with plurality with 100% strategic voters, but this is only when all voters have accurate information about the election and are exactly 100% strategic, thus it is unrealistic that all voters will have accurate information about the election and that they will all be able to figure out how to vote 100% strategically in IRV. Also, being able to rank third parties on a ballot encourages voters to at least learn more about third parties which gives them a much greater voice then plurality gives them.
Warren D. Smith (PhD)
Re Eberhard’s claim (2) “IRV elects non-Condorcet winner less than 1% of time in real world”:
I’d be interested to know what evidence she has for that, but see http://www.rangevoting.org/Aus07.html for evidence that in the 150 Australian House races of 2007, 9 of them either elected non-Condorcet winner or exhibited “no show paradox” (cannot tell which), and 9/150=6%; and see http://www.rangevoting.org/IrvParadoxProbabilities.html for computer simulations in three probability models determining probabilities for various IRV phenomena. The “elects non-Condorcet winner” (aka “thwarted majority,” paradox “Y”) chance is 3.4%, 3.0%, and 19.4% in the three models and 4.7%, 8.5%, 22.2% in same three models but now only for elections where IRV elected somebody differing from plain plurality voting (so that IRV “had an effect”). So it looks highly unlikely to me Eberhard’s “less than 1%” claim can be correct.
Re Eberhard’s claim 3, I do not know what “other” voting methods she had in mind.
But if you check http://www.rangevoting.org/Burlington.html you will see that (a) using the official 2009 IRV ballots but now input into other voting methods, most elect MONTROLL, and (b) an analysis is presented that approval & score voting would have elected him too. Indeed IRV and plain plurality are the only methods from the set considered there, which elected anybody other than MONTROLL.
This probably has something to do with why Burlington repealed IRV soon
after this election. I don’t actually see how the 4-step process — (i) IRV-advocates tell lies about IRV (documented in above link) causing Burlington to enact it, (ii) IRV fails dramatically, (iii) IRV-advocates falsely claim said failed election was a “great success” (iv) Burlington repeals IRV — constitutes “electoral reform momentum.”
It seems to me genuine reform momentum would be to tell the truth about election methods, therefore a good one is enacted, and then since it is good it tends to work well, causing it not to be repealed.
Re Eberhard claim 5, what I personally mean by “spoiler” in voting theory, is somebody who did not win, but if they had dropped out of the race, that would have changed the winner. Both plain plurality voting and IRV can and do exhibit spoilers.
Score and approval voting cannot exhibit them directly but can indirectly. That is,
if nonwinning candidate “Sally Spoiler” is erased from all IRV ballots (or score-voting ballots) in an election but everything else the voters wrote on those ballots is unaltered, then that with IRV and plain plurality and Borda voting can change the winner. But with score or approval the winner cannot change. That was “direct.”
But if Sally were delisted from the ballot and voters were asked to re-vote,
then they might vote inequivalently about non-Sallys versus when Sally was on ballot.
(“Indirect.”) In that case the winner can change even with approval and score voting.
Re claim 6, It is distressing (to me anyhow) that some IRV voters get a preference counted, while others get some preference they stated on their ballot, ignored, by the IRV process. Which ones are counted and which ones are ignored can seem pretty arbitrary. http://www.rangevoting.org/IgnoreExec.html
Re more candidates, I agree I’d like there to be more candidates and more choices,
and (more importantly) more choices who actually can get elected in the real world.
As opposed to (which I consider unhealthy) 2-party or 1-party domination. However, with IRV that pretty much does not happen. In Australia, the country which has used IRV more than the rest of the world combined (80+ years electing House with it),
their House is 2-party dominated, and during the 3 House-election cycles 2001, 2004, 2007 Australia elected a grand total of zero third-party House members for the
450 seat-races. See http://www.rangevoting.org/AustralianPol.html#indepsimpler .
So IRV will *not* change the USA’s massive 2-party domination.
So if Eberhard wants more choices she is foolish to advocate IRV instead of a system with a good chance to break duopoly. And indeed, not coincidentally, the Australian 3rd parties all want to get rid of IRV for House-elections, and so does the Australian public (in the latter case even if forced to go back to plain plurality voting).
To see the evidence for those facts, see
Again, this evidence suggests to me that IRV is not “electoral reform momentum” but
unfortunately rather a “stupid mistake that is killing desperately needed actual reform.”
Warren D. Smith (PhD)
warren.wds AT gmail.com
Thank you for your continued efforts, Warren. We desperately more people to look at the exhaustive data you’ve compiled and see that Range Voting is clearly the way to move forward without the mathematical failures of systems that masquerade as being fair but break down to the same 2 party domination as plurality.
Meeting “Later-no-harm” generates selfish, socially harmful results.
To “successfully” meet that criterion, a voting method must Not elect a consensus candidate because a particular voter (who is part of the consensus, btw) happens to prefer a different candidate, one that is almost necessarily more polarizing.
Is a voting method that prefers extremism over consensus (by “successfully” satisfying LNH) really a good one? Isn’t that what got us into this mess in the first place?
> Wrong. IRV counts votes in rounds, just like a primary and a general.
Kristin is missing the point here. IRV can truly ignore certain preferences indicated by voters. In Burlington for instance, the IRV tabulation mechanism was able to register all the voters who preferred the Progressive to the Democrat, or the Republican to the Democrat. But it did not register that ANY of the Progressive voters favored the Democrat to the Republican, or that ANY of the Republican voters favored the Democrat to the Progressive. Hence the Democrat was eliminated even though he was preferred by a large majority to every single rival.
Let me repeat: IRV completely ignored those preferences. The Democrat was eliminated before those preferences could even be counted.
Kristin’s description of the Spoiler Effect is also problematic. According to Wikipedia:
“The spoiler effect is the effect of vote splitting between candidates or ballot questions with similar ideologies.”
But “similarity” is subjective, and so it’s no surprise that the latter half of that Wikipedia quote is disputed.
The Independence of clones criterion is certainly not relevant though, I can say this much. According to this criterion, “clones” are candidates who always stick together in a preference ordering. Ralph Nader and Al Gore were thus not clones, because there were indeed people who preferred Gore over Bush over Nader. Yet Kristin agrees that Nader is “a classic example” of a spoiler.
But let’s move beyond mere semantics. The practical issue people are generally lamenting when they talk of “spoilers” is twofold:
1. The spoiler has little hope of winning, but could change the outcome from X to Y by running (where X is the Spoiler’s most similar frontrunner). Thus people who prefer X to Y see this as a risk, and wish for the spoiler to stay out of the race. Having lots of choices should be a good thing, but it becomes a risk.
2. People who prefer Spoiler over X over Y may feel tactical pressure to vote for X, in the hopes that they can at least avoid getting Y.
Issue #1 is about Independence of Irrelevant Alternatives, which Score Voting and Approval Voting pass but IRV fails.
Issue #2 is about the Favorite Betrayal Criterion. Which Score Voting and Approval Voting pass but IRV fails.
Both these facts were on display in Burlington. Republican voters who favored the Democrat to the Progressive were punished (got a worse outcome) for voting Republican. If they had insincerely ranked the Democrat in first place, then they’d have gotten their 2nd choice instead of their 3rd.
Kristin makes the common fallacy of assuming IRV fixes the spoiler problem because the Green would typically be eliminated, allowing for the transfer of most of those votes to the Democrat. But this is a common myth. IRV <a href="https://asitoughttobe.com/2010/07/18/score-voting/"maintains two-party domination.
Lastly, I’m baffled by Kristin’s invocation of gender (“it’s often a he”). Kristin further diminishes “voting theorists”, sort of characterizing them as out of touch eggheads with inadequate grasp of the real world. But voting methods are in fact mathematical. They are algorithms. You input ballot data and get a result. Human behavior plays an important role in shaping those inputs, but we find that in many respects that’s a relatively small role compared to the intrinsic mathematical properties at play. For instance, Warren Smith’s Bayesian Regret measures found that Score Voting and Approval Voting can outperform IRV even if we assume 100% worst case tactical behavior for the former and 100% honest angels for the latter.
Just like climate science, this is a highly technical area with a lot of math and jargon. Your comments seem to belittle this. It somewhat reminds me of climate skeptics who assert things like, “We just had a record snowfall—what are you on about with your pocket protector and scientific jargon?!”
Warren D Smith
Looks like Kristin Eberhard is correct about gender (“it’s often a he”) based on
all the comments — all by male names! Personally I have noticed on the internet in
discussions of voting methods and related topics in fora on those topics, that there
indeed is a heavy male domination, let us say 10:1 M:F ratio as a rough estimate (I have not tried to determine it precisely). I have no idea why that is. It perhaps is related to male domination of politics in the USA (and many other countries) in general. In any event I am glad to see some women such as Kristin appearing. Also, in France, where there have a score voting advocacy and science group also (“vote de valeur”) it is possible the women are dominant, or anyhow a lot of the people doing that research there were women. For the French website see
and for my own webpage about a number of government-sponsored French studies
(many by women) comparing different voting systems in action for the French 2007 election see:
(There are English-language versions of some of those French papers.)
I have wondered if the topic actually attracts mostly men, or if men predominantly self-select to comment and critique. Interesting that more women are involved in France–thanks for the links!
Why did you bring up gender in the first place?
A simple empirical observation, as Waren notes: the people putting forth the theories above are 10:1 M:F. English’s annoying gendered pronouns make gender-neutral writing awkward at best, so given the observed gender imbalance, I decided using the male pronoun throughout was reasonable.
It has been fascinating how angry many male Voting Theorists are about the choice of pronouns! Had you really never noticed, as Warren has, how male-dominated the group is? Had you noticed but are just angry that a woman would point that out in public?
As a male voting reform advocate (but not a math-centric Voting Theorist per se), I agree completely with the observation of male dominance here and it makes me very uncomfortable. I think there’s a lot of reasonable hypotheses here, several of which point to real problems correlated with the sexism in our society.
In this particular case, the problem is not *at all* that you simply point out the gender issue. If you wrote an independent article talking about the gender imbalance in voting theory online discussions, that would be totally appropriate. We don’t live in a post-gender or gender-equal world, we live in a patriarchal, male-dominated, sexist world. And it makes sense to grapple with this and point it out in specific contexts.
The ONLY problem in this case is that you brought it up in an article that tells readers to reject the “Theorists” and not listen to them. In that case, saying “it’s usually a ‘he'” puts someone like me (a non-angry, working to be respectful, feminist-as-he-can-be voting reformer male) in a position where the readers are instructed to lump me in the class of “angry, unreasonable, male-dominated Voting Theorists to ignore” in the sense that when all the replies to your article come from men, the “usually a ‘he'” bit serves to tell the reader to ignore the substance of all the replies.
Your reaction to Clay is consistent with this whole issue. You have no basis on his particular personal case to question that he’s angry about a woman pointing out the male dominance. We can be pretty certain based on all the evidence, that Clay would be perfectly happy with an article written by one of the women activists/theorists supporting Score Voting where it lamented the male-dominance.
Sexism in online and offline is so real and rampant that any attempt to downplay it is misguided. But at the same time that we need to call it out and address it, we shouldn’t use it (or any of the other intellectually dishonest things you put in the article here) to silence or reject the views from people you disagree with on topics that have nothing to do with gender.
> A simple empirical observation
But how is it _relevant_? If you noticed most commenters had typically black names, would you say, “Yes, the commenter is usually black.”
What was the point of injecting identity into a conversation about voting methods? I mean, other than to mimic the “Bernie bros” kind of tribal politics.
I’m having the same problem with minimum wage: Wendy Cunningham did a good review of literature; all the good recent literature on MW employment effects seems to be written by men…all the BAD MW literature appears to be written by men…I found one woman who was a co-author on one paper…. Women in voting theory appear to be the same kind of fascinated Wikipedians as the folks at FairVote and RangeVoting, rather than people who actually understand social choice theory. This is odd: I follow a lot of economists, and a large collection of women with Ph.D.s and highly-respected research is unavoidable (Claudia Sahm? Janet Yellen? Lisa D. Cook?).
I’ve been working on a college course covering basic democratic legitimacy, social choice, and voting theory. Also going to build a high school civics course on that. Destroying IRV is basically the first lesson; score voting gets run through the wringer with Baujard’s laughable research (he concludes voters don’t vote strategically in a play election with no stake—exit polls—so they desire to vote honestly but can’t vote honestly because plurality-runoff sucks, because apparently he has no clue how humans work) and Feddersen’s excellent experiment (his team offered monetary rewards to voters, varying the size of the voting group; they voted for what’s best for the group when the voting group was large, but voted strategically in self-interest when the group was low and pivot priority was high). Borda gets absolutely destroyed because lol Borda (see: Kiribati, very first election using Borda, manipulated by collusion between the government and one party to get rid of another party’s popular candidate by adding non-campaigning candidates).
I have an entire lesson on manipulating elections by exploiting flaws in voting rules, and one on dissecting propaganda. Some rules are less-manipulable than others, as you need infeasible amounts of information about other ballots, or the conditions by which they can be manipulated are extremely narrow. I’m investigating Geller-STV, which is much nicer than regular STV in theory…and I’m having trouble manipulating it. I’ve gotten it to work within 1 vote of a tie, but it’s proportional and I can’t get much out of manipulating it anyway; there’s a failure mode where a solid coalition attempting to manipulate it ends up giving another coalition their full vote (the only feasible way to manipulate it is to list another coalition’s candidate rather high in your preferences, which if mishandled can cause fragmentation of a solid coalition attempting to vote strategically in an uncoordinated manner). Some systems can be strategically manipulated only if coalitions band together to do so, caveat: one coalition is better off, but all their co-conspirators are worse off, and they know this going into the election, so why would they conspire?
When social choice theory and voting theory are well-known, I’m sure we’ll see lots of women and men and blacks and asians and everyone in the field doing good research. More information for me to consume.
My observations from all of this have been simple.
First, ranked pairs is easy to explain (I literally break IRV, explain the Condorcet winner, explain cycles and the Smith set, and demonstrate ranked pairs in 2-3 minutes when canvassing door to door; it takes another 90 seconds to explain Meek-STV). It’s got good properties, including that it’s easy to hand-count, can be counted at the polling center and summed, and is more stable than IRV (i.e. as more ballots come in, it should be probabilistically consistent, although none of these systems have consistency).
Second, Geller-STV is promising. STV and IRV aren’t resistant to manipulation because they only take the first non-eliminated candidate into account, as many theorists have said today; they’re resistant because they’re proportional. I know this because Geller used BORDA SCORES to do the elimination, and Borda is one of the worst rules for manipulability. Besides that it’s surprisingly hard to manipulate Geller-STV, you’d only be able to change the winner from a candidate some other coalition prefers to some OTHER candidate that coalition prefers (in the same way a Condorcet system allows non-mutual-majority voters to change which of the mutual majority set is elected). There appear to be mathematical reasons Geller resists manipulation, too.
Third, I have a way to allow equal rankings in Geller-STV, particularly when built on Meek’s method (Meek-Geller). Quota-reaching candidates are elected by vote count, yes; but if we only elect the single quota-reaching candidate with the highest Borda score—candidates are eliminated based on lowest Borda score, so we’re pretending all but one winner is “eliminated” temporarily—and then recompute transfers and Borda scores, we dodge the simultaneous vote problem. Borda score ties can be broken by just taking the one with the higher vote count; and if there aren’t ties for the highest Borda score among simultaneous quota winners, then the elimination of one will cause a transfer to the other equal-ranked and so change (lower) their Borda scores and vote counts.
Equal rankings are important: not voting conveys information. When you have 15 voters voting for ice cream, 7 want chocolate, 6 want strawberry, and 2 are equally satisfied with either chocolate or strawberry, the two abstainers leave the rest to select chocolate. 9 satisfied, 6 not. If the two abstainers are forced to vote and vote randomly for strawberry, you have 8 strawberry, 7 chocolate. 8 are satisfied, 7 are not—an objectively worse result. (To understand precisely why requires understanding public reason and public discourse, but it should be crudely obvious without that.)
This too is a strength of Ranked Pairs: it allows equal ranking. Equal ranking of A and B means not voting in the pairwise A vs. B election.
Schulze is complicated. Smith/IRV requires centrally hand-counting all votes and so counting is more complicated and expensive, and can’t be started until all ballots are reported (this already plagues STV, and can’t be worked around), and can’t allow equal rankings except of unranked candidates. The merits of Smith/Minimax are debatable; Minimax plain is unworkable, as it may elect the Condorcet loser.
IRV vs. score/approval/lollypop eating contest is not the only possible discussion here, and I believe IRV vs. some reasonable Condorcet method is a valid and much more agreeable debate.
Re: “this voting theorist is usually male.” Sure, they are, but I’m female and so are a number of other women I know working for Star Voting/SRV in Oregon and Portland. I’d love to encourage more women and minorities in politics and I think a great way to do so would be to engage in an open minded and respectful conversation on the issues. This article came across to me as very divisive and finger-pointing. I think that I am “this voting theorist” that you urge others to discount and ignore. :/
The framing of the article here is concerning to me and it seems like your points are slanted in such a way to confuse the issues. IRV and Star Voting are the top two contenders in Portland right now, where you are writing. Lumping Star Voting/SRV in with Approval and plain Score (which nobody here is advocating as far as I know,) is misleading and seems to intentionally skew conclusions in your favor.
Kristen, you repeatedly bring up Condorcet Criterion and then break down why it’s not a perfect criterion. But what you leave out is that the point of Condorcet as well as other methods like Voter Satisfaction Efficiency (VSE) and Bayesian Regret are to measure Accuracy of an election or voting system. You don’t say so outright, but you have offhand discounted Accuracy as an important criteria for you. Any way you slice it and by every metric that I know of IRV is only marginally more accurate than Plurality and the scenario where it fails the most is specifically where we Plurality fails. Elections where there are 3 or more viable candidates.
The reason that such a tiny percentage of elections these days have 3 or more viable candidates is BECAUSE Plurality (and IRV in Oz) is wildly inaccurate in this scenario and so there is huge push back to prevent minor candidates from running and to prevent voters from supporting them, even when they would like to. That’s the problem and a functional election system would absolutely lead to much more diversity of both candidates and voters. Let’s plan for that and get Star Voting or another system that is highly accurate with any number of candidates!
@Sara, what, precisely, do you mean by “star voting,” and how is it different from score voting?
Sara is talking about this.
The reason most elections only feature two strong candidates from the major parties is because of plurality voting. Plurality voting causes and produces the strategy of organizing into two broad party coalitions, one on the left and one on the right.
With a different voting system, this strategy would evaporate and we would see more true multiparty democracy, with new political parties and a realignment of our political landscape. In this future scenario, there will be a lot of races where there are three or four or more candidates from parties that have a decent chance of winning. Therefore we need a voting system that will be robust and not break down when several strong candidates are running in an election.
The Later No Harm criterion is over-rated and mostly mitigated with Star Voting (Score Runoff Voting) with the runoff creating the incentive to score at least your second favorite choice highly so they will make it into the runoff over others.
The spoiler and favorite betrayal criterion are more important.
I completely agree with you that plurality in single-winner districts leads to two party domination, and we need more parties to have a more functioning government!
Unfortunately, no single-winner system (not IRV, not Star) will break the two-party duopoly so long as we have single-winner districts. We need some form or proportional representation.
Amen to the point about PR. Hearing really honest thoughtful arguments about that from IRV proponents changed my mind on both PR generally and on STV being positive. I used to think that because STV was like IRV, it mean they were both lousy. I now believe that only IRV (single-winner) has the level of real problems worth worrying about. STV would be positive enough to be worth supporting, and I even understand now why STV proponents are biased toward IRV as a stepping-stone (even though I’m not convinced there).
People can and do change their minds when presented with good arguments. We all grow and improve when we work to be open-minded and intellectually honest. That’s why I was so appalled at this article because it’s directly a call for closed-mindedness and intellectual dishonesty.
John R Moser
Absolutely. I am seeking to REMOVE parties from our system. You get a ballot, it has 7 candidates on it, next to their names 3 say “Democrat” 2 say “republican” there’s a Green and a Libertarian…just for information. No, they don’t have a special place reserved for them; they all petition, they all get on the primary ballot, we all have a single transferable vote primary between all candidates, and the voters choose their nominees—yes, the voters, not the parties (the voters registered to a party being the members of the party).
A party’s central committee appoints the replacement when a seat is vacated? No. We’ll elect a board of appointments by STV. When there’s a vacancy, they will do the appointing. They can nominate and then vote by a Condorcet method, or they can vote to approve by 3/5 or 5/7 super majority.
I don’t care much for party labels. The parties have multiple factions within. We don’t need more parties; we need elections that represent voters. If that means voters are selecting the sort of moderate-liberal Democrat or the left-progressive Democrat instead of the “Liberal Party” or “Progressive Party” candidate, then so be it—the name is just for show. Using STV as a non-partisan nominating primary gets you several candidates from the various parties based on voter support for candidates, rather than several candidates based on various party labels.
If it were always valid to say that practice trumps theory, the Wright Brothers would have been wasting their time tinkering with heavier-than-air aircraft.
The Wright brothers didn’t just theorize about how an airplane might hypothetically work. They built one. And flew it. In the real world. In fact, their first paper design (the Kite) didn’t work in real world conditions quite like it had worked on paper, so they use their real-world experience to revise and improve it in ways they could not have done if it had just stayed in their heads and on paper. I have the utmost respect for their methods of bringing theory into practice.
Here’s a lesson from history that you should really take to heart. When your apparently favored system of Single Transferable Vote was invented in 1819, it was designed based on mathematical principles—pure theory. It wasn’t used in public elections until 1855, when it was proposed for Danish elections by Carl Andræ, a Danish politician and mathematician. One can imagine he faced similar pushback about such an abstract and theoretical new idea. But he apparently had a mathematician’s brain that allowed him to see the potential in spite of all that.
Modern “Voting Theorists” like myself have a massive advantage over those 19th century reformers, as we’ve seen groundbreaking new work in social choice theory (e.g. Arrow’s Theorem and the Gibbard-Satterthwaite Theorem). We also have the ability to perform sophisticated computerized electoral simulations. And the internet enables researchers across the globe to cooperate and build on each other’s work. So our proposals are certainly less risky and more vetted, by an order of magnitude, vs. the initial adoption of STV.
We’ve also utilized a great deal of empirical data. For example:
So when you assert that the systems we promote are “untested”, that’s totally out of line with fact.
And unlike with airplanes, a substantial amount of voting method behavior simply cannot be empirically tested, because we cannot tell the future and read human minds. This is why Bayesian Regret calculations must use simulated people instead of real people.
You’re literally asking people to not adopt good voting systems in the real world because they haven’t yet been adopted in the real world. You don’t see the flaw here?
No, I am not at all saying we shouldn’t try new things.
What I am saying is comparing how you imagine an untested method will perform in the real world to how another method has *actually* is not a fair comparison. I’m merely suggesting we look at what we actually know, and separate that from motivated speculation.
Kristen, I will continue to believe (because I’ve seen evidence in many cases from you and others) that you people hold intellectually honorable and also dishonest views at the same time.
You seem to have an *honest* assertion of skepticism about concepts that haven’t had real-world testing. No matter what anyone else says, until we have real-world Score or Star elections, we cannot actually know that we’ve accounted for all the issues in the theory, it *needs* that level of empirical support.
Your article here goes beyond acknowledging the unknowns given the small amount of empirical evidence. It goes into advocating for intellectually dishonest thinking where you call the thing you disagree with “speculation” while not taking anything close to that level of skepticism in viewing your favored view. More specifically, this is a strawman in that nobody of the IRV critics is suggesting anything other than looking at what we actually know and separating that from motivated speculation.
I mentioned Rapaport’s Rules before. You should try them (as should everyone, I too could do better):
1. Attempt to re-express the other person’s position so clearly, vividly, and fairly that they say, ‘Thanks, I wish I’d thought of putting it that way.’
2. List any points of agreement (especially if they are not matters of general or widespread agreement)
3. Mention anything you have learned from them
4. Only now are you permitted to say so much as a word of rebuttal or criticism.
Right. You’re asking people to not adopt good voting systems in the real world because they haven’t yet been adopted in the real world [and therefore we can’t predict how they’ll behave]. Same difference.
Kristin appears to ignore that we use score / range voting all the time in rating systems in the real world. It’s not “untested” at all. What’s anomalous is simply that we haven’t been using it for elections, mainly because the 2 parties in dominance want to maintain that dominance which a plurality system as well as IRV (which breaks down to 2 party dominance) both allows.
Star Voting has been tested by theory, by simulations, by mathematicians, by live elections during real public events like Portland Forward’s “Spring Thing”, and is currently being used by private groups and organizations. Test it yourself. I invite you. It takes about 5 minutes to build your ballot following these simple instructions and a more user friendly interface is in the works. Enjoy. Everyone else has so far.
You will be interested to note that the results we’ve gotten back have had a high ratio of mid range votes and most people don’t seem to bullet vote after all. Even if they did, the system is still more accurate than IRV and for some people that might be a legitimately honest, expressive vote. Some people just see things in black and white. Others choose to be more open minded.
There is one important benefit to approval voting that IRV doesn’t offer that I hope you consider. Approval voting allows voters to easily form powerful voting blocs around issues that IRV can’t.
As an example, let’s say that you are a college slacker that only cares about legalizing pot. If you want to vote to legalize, you could just vote for all the candidates endorsed by “Citizens for Legalizing Pot.” You don’t have to worry about who is electable, you can just vote for all of them and be one step closer to legalization. If you also care about protecting net neutrality, you can just vote for all the candidates mutually endorsed by “Citizens for Net Neutrality” and “Citizens for Legalizing Pot”. Or if you are more generous with your vote you can vote for all candidates endorsed by either of them. Politicians will have to pay attention to the important issues to earn endorsements from the advocacy groups supporting those issues.
This is something that can’t happen with IRV because order of elimination really matters and it is not safe to put your favorite first. If an advocacy group just ranked their endorsements based on a candidates strength of support for their particular issue, they could be helping to eliminate more electable allies in the early rounds. With approval voting they could offer direct support to any candidate who has supported their issue without having to hedge their bets.
IRV is a poor voting system. This has been proven both by theory and real-world experience. I agree that inertia is good, and infighting over “the best” method is bad, but it doesn’t follow that we should continue pushing a bad voting system just to maintain that inertia.
Most people don’t support IRV because it’s a good, proven, system. They support it because it’s the only reform they’ve heard of. It has a lot more marketing power behind it than the better systems you’ve mentioned.
Maine made a mistake in adopting IRV. It will probably work out ok for them, most of the time, but when other states or nations reform their elections, they have a chance to choose a system that actually works. Why are you fighting so hard in favor of mediocrity?
> But Score, Score Runoff, and Approval Voting all also fail the Condorcet criterion
Yes, because Condorcet criterion only exists in the flawed world of ranked-choice voting, which ignores strength of preference. The real goal should be to elect the Utilitarian Winner, the one who best represents the voters and makes them happiest with the result of the election. Condorcet systems *typically* elect this candidate, but not always. IRV rarely does.
Kristin Eberhard asked (on September 2, 2017 at 6:02 pm):
I have wondered if the topic actually attracts mostly men, or if men predominantly self-select to comment and critique?
i dunno, maybe power-lust and testosterone have something in common.
Anyway, I haven’t read through every comment, but I did a few. Has anyone suggested that rather than argue about what method satisfies Condorcet the best, why not just advocate a Ranked-Choice Voting method decided by a Condorcet-compliant method, likely Ranked-Pairs or Schulze ?
IRV is flawed because in the case where at least 3 candidates are ostensibly viable for the single winner office, it is in real danger of failing to elect the Condorcet winner and the CW can be identified. Then if fails in its resistance to the Spoiler effect and in its promise to voters that they can sincerely mark their favorite candidate as their favorite, without risk of helping to elect their least-favorite candidate. That is what Burlington Vermont experienced in 2009. And I was a resident and voter (and knew all three candidates).
For those who say that the Condorcet winner is not always the “best” candidate, that’s like saying that the democratically-elected candidate is not always the “best” candidate. I presume they mean “best” in terms of merit. Like the populist demagogue who gets elected has less merit than the candidate he/she defeated. (I can’t imagine a current example… :-0 .
Anyway, if you say that there is some situation where a CW exists but some other candidate should be elected, then you are saying that we should elect a candidate when more voters marked on their ballots that they preferred some one else that this candidate. And a *specific* someone else, not just the complement of the plurality. Who would advocate that we elect the person who gets the fewest votes in a 2-choice election?
Like any election system and election, sometimes a douchebag gets elected. No one says that Condorcet is immune or even resistant to douchebags. All Condocet does is, whenever possible (which should be most times), that the candidate who is elected is preferred by the voters to every other candidate.
Why advocate for IRV instead of Condorcet? Let’s advocate for Condorcet instead. Whenever a pairwise champion can be identified, Condorcet elects that candidate.
Robert, I agree that Condorcet is a good indicator of who should win the election. In a ranked choice election the Condorcet winner absolutely should win because they were the preferred over the others judging from the ballots cast.
The only time when there is a clear Condorcet winner but another candidate should win instead would be a close tie in an election using a more expressive ballot like a score ballot. A score ballot (or star ballot) shows a voters preferences, just like RCV, but can also show how much each candidate is liked or disliked. All this extra information (and the fact that no ballot data is disregarded like happens in IRV,) means that there actually could be another winner with slightly more support than the Condorcet winner. It’s an edge case and absolutely not a valid defense of what happened in Burlington.
I think we need to get back on task here. The reason we’re talking about Condorcet is because we are trying to talk about accuracy. Does the voting system pick the best winner? Any analysis of voting systems that ignores this fundamental question is way off track.
Condorcet methods *typically* choose the correct candidate, but not always, since they discard information about strength of preference. Plus they’re terribly complicated, both to tally and to vote under, so why promote them at all? Ranked ballots are old and busted; arguments about which methods meet which mathematical criteria will outlast the sun.
Cardinal methods are simple, transparent, and more effective than any ranked method. How do we get people to realize this?
Rather late to this game but here are my 2d.
1) “IRV maintains the duopoly”. Of course it does, only proportional representation breaks the single seat duopoly.
2) All voting methodologies pass and fail various voting criterias. Voting theorists often neglect to consider the tactical voting opportunities different voting criterion compliance/non-compliance offer. For instance “the independence of clones” any voting method that fails this are incredibly open to tactical voting.
I really wanted to read this article but I couldn’t make it past the initial remark about gender. It’s really too bad.
Interesting comments. The people here seem to understand the various nuances in quite some depth.
For the record, I am pro-Scoring and anti-IRV, but there is something in particular about Scoring-type systems that bothers me. One of the main selling-points of scoring ballots is that they let voters express a more fine-grained, honest, and accurate strength of preference. If the voters are highly knowledgeable about all of the options and they are honest, then that sort of ballot seems mathematically optimal. However, empirically most voters seem mostly ignorant and dishonest, and a scoring ballot would seem to exacerbate those problems relative to less expressive ballots. For example, let’s say a maximally ignorant and dishonest voter would assign the highest rating to a random candidate and the lowest rating to all other candidates, effectively mimicking a Plurality ballot. If all other voters are trying to be honest, then the ignorant, dishonest voter’s ballot would seem to carry more weight due to the nature of scoring. As a toy example, if there were two voters, and one was maximally dishonest and the other maximally honest, then a Scoring system would never elect anyone other than who the dishonest voter gave the highest rating to. At least with a Plurality system, a 100% honest voter has the chance to offset a 100% dishonest voter, but with a vanilla Scoring system they have no chance to do so. The people who are 100% honest and who try to give as accurate a score as possible on a linear scale would be at a disadvantage to people who dishonestly / unknowingly / ignorantly compress their scores into narrower bands.
This leads me to speculate whether a ranked or approval ballot might actually be more expressive of true preferences than a scoring ballot (given the likelihood of high dishonesty), because ranked and approval ballots limit how dishonest someone can actually be (relative to 100% honest voters) while still providing good enough expressivity to 100% honest voters. On ranked and approval ballots, 100% honest voters can fully offset 100% dishonest voters, and that makes those ballots highly appealing in my eyes.
Given all that, wouldn’t it be great if we leveled the playing field by forcing everyone’s initial ratings to be in narrower bands, eg, a person could say “these options are acceptable” and “these options are not acceptable”, and then within those bands we allow them to rank or score the options? In other words, use a scoring ballot, but make the counting rules like this (arbitrarily using a score of 1 to 4):
1) Every score that is a 3 or 4 is considered “acceptable”. Every score that is 1 or 2 is considered “unacceptable”.
2) All but the two highest “acceptable” vote getters are eliminated.
3) The winner of the two finalists is determined by either ranked preference, average score, run-off, or some other system with good properties.
This system is similar to STAR, but it is different in one important way: it does not give an advantage for extreme dishonesty while still allowing people to honestly express a preference within the “acceptable” and “unacceptable” bands without having to worry about the spoiler effect. I haven’t really analyzed this sort of system since I just came up with it while thinking about my concerns with pure scoring-based systems, but it seems to address the problems that I worry about with most other voting systems.
Anyway, I didn’t mean to go off on a tangent. I just wanted to put out in the universe what I see as a serious flaw with plain vanilla scoring.
I guess I would call the system I described “2-Tier Approval”, which could then be generalized to “N-Tier Approval” using a scoring ballot with a larger range. The more I think about it, the more I like it ^_^
On ignorance, you might be right. But that is a flaw that effects every voting system. In every voting system, people are ignorant and don’t do their research on candidates. But, in any system, those who do their research on candidates will be be able to express their preferences much better than those who don’t actually do their research to determine how candidates compare.
As another commenter pointed out, even with dishonest strategic voting, score voting does much better in terms of voter satisfaction and regret. To quote Clay Shentrup “Warren Smith’s Bayesian Regret measures found that Score Voting and Approval Voting can outperform IRV even if we assume 100% worst case tactical behavior for the former and 100% honest angels for the latter.”
Additionally, STAR voting (also known as score runoff voting) disincentives strategic voting quite a bit. In STAR, the two candidates with the highest average scores go to a second round in which the candidate who is preferred by a majority of voters wins.
John R Moser
Bayesian regret is the Drake Equation of voting theory: it supposes a bunch of immeasurable things and then imagines what those variables might be, so as to show a certain outcome.
Bucklin Voting with tied votes allowed seems to fit your description of N-Tier Approval. If the 1st place votes don’t produce a majority winner, the 2nd place votes are added in, and so on until majority is achieved.
Regarding “banded” voting, have you looked at Majority Judgment? It only cares about the median vote, so that votes at the very top & bottom of the scale are weighted no more strongly than votes one notch above or below the median.
I have two main problems with IRV.
1: non-monotonicity. See Voting Simulation Visualizations ( zesty.ca/voting/sim ). The bizarre behavior shown in those graphs might be rare in actual practice, but to me IRV is only a lesser evil, not a top preference.
2: bad advocates. Way too many people who support IRV claim (either from ignorance or malice) that IRV is the best option out of the EXACTLY TWO voting methods known to exist (three if you include dictatorship). They behave exactly like the folks who insist that since Mercator maps are terrible for classroom use, then Peters maps MUST replace them. I oppose false dichotomy on general principle.
John R Moser
The bizarre behavior required for IRV to be non-monotonic is negligible.
Ranked pairs is monotonic, Smith-efficient, and simple enough. IRV is just crap, but not because it’s non-monotonic.
This is a very shameful article. Why is Sightline working against democratic voting reforms that put power in the hands of the people? You’re tools of the establishment, trying to perpetuate our polarized two-party system by adopting fake reforms like IRV?
Please take a look at this paper:
The Failure of Instant Runoff Voting to accomplish the very purpose for which it was adopted: An object lesson in Burlington Vermont
This is now, again, a hot topic in Vermont with the re-emergence of a push for RCV from FairVote and VPIRG.