( – promoted by odum)
(cross-posted from Integral Psychosis)
It’s with a somewhat heavy heart that I write this post. A number of times I have written extolling the virtues of IRV (Instant Runoff Voting), including directly after the recent 2009 mayoral election in Burlington, VT. In fact, my previous post (on my own blog) went as far as to declare IRV in Burlington’s 2009 mayoral election to be a success while simultaneously pointing out that a paradox- a failure of the system to be consistent, free and fair- occurred. I want to be very, very clear as I move forward here: plurality voting (i.e., whoever gets the most votes wins) is a drastically flawed system that only works when there are only two choices on the ballot. This is because of something called Arrow’s Impossibility Theorem, which is simply a mathematical concept which states (recognizes) that whenever more than two choices appear on a ballot the standards for a consistent, free and fair vote (the Fairness Criteria– developed and agreed upon by political scientists and mathematicians) cannot be met. In the past I have extolled IRV as being the best alternative because it generally is capable of creating paradoxes or violating the Fairness Criterion a reasonably low percent of the time. In my opinion, there may still be some truth to this, and it is entirely true that IRV out-performs plurality voting. However, what seems to be even more true is that there are other methods for voting which greatly out perform IRV, and at the very least a modified form of IRV is necessary to over-come its most glaring flaws.
Unfortunately, the 2009 Burlington mayoral race demonstrates pretty much everything that can go wrong with IRV voting, and the simple fact that it did so should give us all pause and make us re-consider our acceptance of this voting method. Again, I want to be as clear as I can: IRV is drastically better than plurality voting (the method we’re accustomed to using when electing the President or members of Congress). That IRV in this instance suffered from a number of paradoxes and most certainly proved itself to not be consistent, free and fair should in no way whatsoever be construed as a reason to revert to plurality; likewise, the other options on the table for possible ways to hold an election all carry with them their own strengths and weaknesses. Again, Nobel winning mathematician Arrow proved for us decades ago that no system is going to be perfect, and even those few since him who claim to have found or proven an “impossibility” (a voting system that doesn’t violate the Fairness Criterion) have not done so to the degree that they have gained widespread acceptance for their findings. This is, in the end, a terribly complicated issue and there are no easy answers. There is, on the other hand, at least one important question on the table: was Bob Kiss preferred by the majority of voters in Burlington this past Town Meeting Day? A careful examination of the ballots suggests the answer is no, absolutely not. And because this is the answer the ballots themselves give us, it’s important we re-examine IRV because Kiss is the candidate it handed over as the “clear” winner.
As for the 2009 mayoral race in Burlington, I’d like to draw your attention to the facts as pointed out by UVM political scientist Anthony Gierzynski and Temple University mathematician Warren D. Smith (*and please note, whatever political agendas or personalization of this issue may exist between either of these two men and IRV or ardent IRV supporters like Terry Bouricius is not of my concern whatsoever- I am interested only in the facts as the actual ballots and a thoughtful, scientific analysis of them reveals):
–As I had already been pointed out in a previous post on my site, Montroll was the “Beats-All winner” (aka the “Condorcet winner”) as he would have beaten both Wright (56% to 44%) and Kiss (54% to 46%) in head-to-head races, demonstrating that he was the preferred candidate by the majority of voters. This is called a “Thwarted-Majorities Paradox”.
–Despite claims that IRV eliminates the “spoiler” effect, the truth is that Wright was a spoiler to Montroll. If Wright had not run, Montroll would have won (which, incidentally, would have been preferable for the majority of Wright supporters who listed Montroll as their second choice 1513 to 495 for Kiss as their second choice). Anyone who voted for Wright without fear that doing so would cause Kiss to win was simply wrong, because if those Wright supporters had strategically voted Montroll (like if Nadar voters had strategically voted Gore in Florida 2000) Kiss wouldn’t have won. This is called “Favorite-Betrayal” (i.e., when voting for your favorite harms either that candidate or another one you would have otherwise preferred).
–This election also produced a “No-Show Paradox”, meaning that if 753 Wright supporters who preferred Montroll over Kiss had simply stayed home on Town Meeting Day they would have gotten a more desirable (in their view) outcome- Montroll would have beaten Kiss. For these folks, the election results would have been more to their liking if they didn’t bother to vote!
–The software used to tabulate the votes eliminated Smith (1306 votes), Simpson (35 votes), and “write-in” (36 votes) in the first round of counting because it deemed them all to be mathematically (“inevitably”) losers of the election. Yet the software provider has made no clarification (that myself or the other listed authors of this analysis can ascertain) as to what criteria defines this: using the information and “logic” that has been provided, we could very well say that Montroll (and even Wright) would have “inevitably” lost and eliminated them in the first round as well.
–Most problematic of all (and for me, once I realized this my entire opinion of IRV began to rapidly shift): this election featured non-monotonicity. If 753 people who voted Wright as their first choice (specifically, all 495 who voted Wright>Kiss>Montroll plus 258 of the 1289 people who voted Wright only) had instead voted for Kiss then Wright would have been eliminated instead of Montroll and Montroll would have beaten Kiss in the final round of IRV (4067 to 3755). Put another way, Kiss won, but if 753 Wright voters had instead voted for Kiss, then Kiss would have lost and Montroll would have won (Monotonicity is one of the four Fairness Criteria and it states that “if x wins an election, and in a do-over or re-election the only changes made are in favor of x, then x should still win”). Simply put: if things change in your favor, but those changes negatively impact your final outcome, that is not consistent nor fair.
In the end, there are a whole slew of voting methods to consider. As I said, they each have their strengths and weaknesses. In the past I have been a strong proponent of IRV. Unfortunately, this election should give us all pause. Consider voting methods that would have chosen Montroll (simply using the preferences indicated by the ballots cast by the voters themselves): Nanson-Baldwin, Black, Raynaud, Schulze-beatpaths, Simpson-Kramer minmax, BTR-IRV, Tideman-ranked-pairs, WBS-IRV, Copeland, Heitzig-River, Arrow-Raynaud, Borda, Dodgson, Bucklin and (probably) Range and Approval. Voting methods that would have elected Wright: plurality. Voting methods that would elect Kiss: IRV.
Another way we can look at this all: all of the above voting methods are unanimous in one thing: Wright was the least favorite of the top three choices (i.e., he was the “worst choice” or the least popular among them). If this is true, then at the very least the election should come down to Kiss versus Montroll, and we know the voters preferred Montroll 54% to 46% over Kiss.
Again, none of this should be mis-construed: in a plurality voting system, Wright would have won the election despite being the “Lose-To-All loser” (i.e., he would lose head-to-head against either Montroll or Kiss), which would have been a far greater dis-service to consistency, fairness, and democracy. As well, I am not particularly extrapolating from any of this a recommendation about what system should replace IRV. Simply put, the jury is still out. We do know, however, that plurality is pretty much the worst possible method for voting we could use. As well, IRV doesn’t necessarily fail on this level every time- but the fact that it has, no matter what the statistical probability, is deeply concerning. We also know that, according to the consensus of the political science and mathematical communities no known system completely satisfies the most basic and simple criteria for consistent, free and fair elections (and though there are a number of individuals or small sects which claim otherwise, these claims have thus far not been accepted by the wider academic community and therefore should be considered with care and a thorough examination of the facts; the public has generally demonstrated to have a weak stomach for having its democracy rest on the back of an unknown system). Still, the claims of IRV have proven dubiously inconsistent at best,and possibly far worse than that. Because of this the public needs to carefully, thoughtfully engage in a conversation about how best to move forward. Range voting, for instance, may be a worthwhile step forward.
In an effort to pre-empt at least some of the critics who will come out against these conclusions: I absolutely favored Kiss to win and if I had lived in Burlington I would not only have voted for him, but probably would have volunteered for his campaign. Likewise, the voters have approved IRV as the method they want to use and it is absolutely true that IRV produced Kiss as the winner, and I do not think any of this analysis calls into question the fact that he won the election and is mayor (Wright’s re-count, even if carried all the way through, would have reproduced Kiss as winner every time using IRV, which is the method Burlington currently uses, by law). However, analysis of the ballots strongly suggests that Montroll was the most preferred candidate by the majority of the voters and the fact that IRV (or any other system) fails to elect him is troubling; again, not because I personally would prefer it (just the opposite, actually) but because the democratic will of the people would have. Sound logic does not make value judgments, except against that which proves inconsistent. The results from the 2009 mayoral race in Burlington were inconsistent (the candidate most preferred by the voters failed to win), unfair (voters intending one thing and preferring yet another got neither) and unfree (some voters would have gotten results more to their liking if they voted insincerely, or not at all).