IRV -- Puzzling Findings: Council Race, Seat # 2

Aspen, Colorado just had its first experiment with an IRV election. There was plenty of warning beforehand that it wasn’t a good idea to be playing with matches without some adult supervision and serious thought However, in the rush to adopt IRV methods just weeks before the May election, the City got careless. We now have singed hair and blistered fingers as the details of the IRV black box are beginning to belatedly spill out. We managed to get through the election, with no big surprises in two races, but created a perverse puzzlement in one Council race. While there is much more to be analyzed, so far 1) three weeks post election, the City restated the official mayor’s race totals after the election was certified, as more votes were found for the winner. The missing votes had been delivered to the loser (that’s me), by using Cambridge, Massachusetts rules rather than using Aspen’s IRV rules, and 2) that “it will never happen in real life” non-monotonicity bugaboo produced a poster child for why IRV can greatly disenfranchise voters and candidates.

Aspen voters have not yet been informed of the second problem through the press or letters to the editor, but will be aware soon, when the worksheet proving the bizarre answer is ready for the public record. In the meantime, I’m sharing with some friends wise to the wily nature of IRV, the result that occurred here.

Michael Behrendt, Council candidate, got defeated by 75 of his own supporters, doing their best to support him by ranking him #1 on their ballot. Turns out that he lost to candidate Torre by just 43 of 2103 votes cast in IRV tabulation terms. However, two independent analysts have calculated that if Behrendt had only had the foresight to ask 75 of his loyal supporters to rank him #2 and change their lower rank for candidate Jack Johnson to #1, AHEAD of Behrendt, Behrendt would have won. Little did those voters know that they were costing their friend Michael the election by voting FOR him as the number #1 candidate.

Whether you were a Michael, Jack or Torre supporter it has to be a bit disconcerting to know that the order in which you voted for your favorite might have been hurting him instead of helping him get elected. Aspen’s flavor of IRV contains mysterious anomalies for us non-mathematicians, which, as demonstrated, can happen in real life, in our local elections. This “Michael Behrendt effect,” I’ll call it, is one of the side effects of “non-monotonicity”. Apparently, the larger the field of candidates, the greater the probability of puzzling outcomes from seemingly minor choices in ranking the candidates. Voters can’t simply depend on the normal voting logic we’ve known since kindergarten .

Did the Council and IRV Task Force know of such possibilities when adopting IRV? The risks were well documented, but in their rush to adopt a IRV system, non-monotonicity passed off as “rare”, or acceptable as “no system is perfect.” The fact that 2 incumbents (also candidates in the May election) appointed themselves to the IRV Task Force and then voted as Council members on the method their taskforce had recommended , might have had a bit to do with why such anomalies were not thoroughly discussed, although the public raised the issues repeatedly. The lack of independence was rather astounding! Were the voters properly informed as to the risk and the complexity? Ask Michael’s supporters, who voted for him first instead of second, thinking that they were definitely helping him get elected. In fact, City Hall assured us in the public hearings to adopt the IRV procedures that it could not happen, despite warnings from mathematicians.

While it is a shame that Michael was defeated in such a perverse way, his situation will bring the hot spotlight to IRV in Aspen. Michael is a well known long time local citizen, a Council member in the 1970’s, and a small lodge owner. He is loved by the entire community and quite active in civic affairs. The fact that the puzzling system defied logic and defeated such well-respected candidate will get additional attention. No doubt he will become the poster child for Monotone Violation in IRV!

I will be posting a review of alternative election results, including the “Michael Behrendt effect,” showing some what-if scenarios on www.TheRedAnt.com . I have also posted Kathy Dopp’s Instant Runoff Voting Flaws paper http://electionmathematics.org/ucvAnalysis/US/RCV-IRV/InstantRunoffVotingFlaws.pdf, to offer evidence to the skeptics.

For some less technical background arguments in Aspen against IRV see:

http://theredant.squarespace.com/red-ant-blog/2009/3/1/caleb-kleppner-and-his-kleptomaniacal-machineissue-30.html

(who knew?? That The Red Ant would be prophetic about that black box.)

and

http://theredant.squarespace.com/red-ant-blog/2009/2/16/dont-know-much-about-algebra-issue-29.html

Comments

[Rob Richie]

[POSTED FROM EMAIL RECEIVED FROM ROB RICHIE OF FAIR VOTE]
Hi, Marilyn,

Just to make sure you know, the situation in fact no different than any runoff system that eliminates candidates -- except that traditional runoffs are easier to game because you can change your choices between rounds.

Just to be clear on what I mean, let's use a race for mayor as an example with a runoff. Suppose I'm a backer of the frontrunner, who is facing two strong opponents. Let's assume that one of the opponents (let's say "Joe") would be easier to defeat in the runoff than the other ("Jill"). I could say "okay, I think my favorite is going to go to a runoff, and I think he/she would do better against Joe than against Jill. So I'm going to switch my first round vote to Joe." Joe makes the runoff, I then change my vote to my favorite and he/she wins the election.

But... suppose I and the other "switch-to-Joe-in-first-round" voters instead had voted sincerely for our favorite in the first round. Doing so leads to Jill making the runoff instead of Joe. In the runoff, Jill indeed does beat my favorite. So voting for my favorite in the first round caused his/her defeat.

A well-known example of how this can work in two-round systems occurred last year, when Rush Limbaugh urged McCain backers in open primary states to vote for Hillary Clinton, as he thought she would be easier for McCain to defeat in the fall.

Note, however, that this is a complicated and risky strategy to try with traditional runoffs. You often can't be sure your candidate cannot win on the first round, you can't be sure that they're safely in the runoff and you can't be sure who's really the tougher runoff opponent. But, here's the key point: it's ___much __ harder to try anything tactical with IRV because you can't change your vote in the runoff. Anyone ranking Joe first in my example below would have their ballot count for Joe in the instant runoff. Those Limbaugh-inspired voters would have to have their ballot keep counting for Hillary (e.g., they woudl have ranked her first). That's why none of this ever as any real-life impact in the thousands of significant IRV elections that have taken place, but you do see it more often with your old runoff system.

Let me know if you have any questions about it. You can see our analysis of monotonicity here:
http://fairvote.org/?page=2261

Rob