1. Wow, that is a terrible article. Mr. Zeitchik apparently took no effort whatsoever to understand IRV or the problem it’s meant to solve. Just look at that last sentence: “a movie could pull a Bush vs. Gore — win best picture despite not getting the most first-place votes.”

   The irony that he’s using an example of the sort of problem IRV solves to criticize IRV is incredible.

2. In large agreement with Drew.

   I also find it amazing that the author compares IRV calculation to understanding cold fusion, when he actually does a good job explaining the process and relating it to the use at hand.

   Somehow the author thinks IRV creates the kind of electoral college debacle which caused Bush to win with less votes than Gore (the third time this has happened). In a strict sense, IRV does allow the person with less first choice votes pull ahead when they have support from lower choices. However, IRV does this in an appropriate way. The alternative is to have a winner with less (often much less) than 50% of the vote. This has occurred in about a third of US presidential elections over the last century. Obviously, IRV does what IRV does best, which is preventing vote splitting among the leading first choice candidates.

   Normally, I would throw some love to Condorcet here, but I actually like IRV in this situation. I don’t see the presence of the kind of ordered polarity that Condorcet deals with so well here. With this application, I see more of a cluster of awesome movies which would otherwise have their vote split and lose to a “good” movie. I could see range voting having some possible application here. However, voters might feel pressured to give a 10 to a movie thought to be in contention with another hyped movie and thereby vote insincerely. IRV, on the other hand, is very good at resisting insincere voting in this situation.

   Go IRV. Boo LA Times writer.

3. To #1: IRV *can* “pull a Bush v. Gore”, and to #2, it doesn’t solve the vote-splitting problem.

   I’d go into detail, but I just wrote on precisely this issue at my blog:

   http://leastevil.blogspot.com/2010/02/oscar-update.html

4. The vote splitting issue that I see IRV addressing is between the leading first choice candidates only. In lineups that tend to have a specific ordering, there’s a subtle and real vote splitting issue that occurs which comes from IRV’s monotonicity problems where comporosmise candidates are taken out.

   If it’s your suggestion that the movies will have a tendency for order this vote splitting could be a concern. That is, one group orders A>B>C>D>E>F>G>H>I>J and another group says J>I>H>G>F>E>D>C>B>A, then I can see the issue of monotonicity coming up. However, I’m not sure that is present here and as a result I don’t see vote splitting generating a large concern. If this ordering is actually created after several rounds, this could still be a problem. It’s a shame the academy doesn’t reveal the ballots. I’d be curious if IRV’s unsavory behavior managed to creep in.

   You make an interesting point in your blog about IRV favoring fanatics. This otherwise unappealing characteristic may actually be useful here depending on what one wants from the winner.

5. Aaron Hamlin,

   First of all, monotonicity has nothing to do with the vote splitting problem. These are two completely separate issues.

   Second, IRV does not fix the vote splitting problem, even among front-runners. Here’s a simple example IRV election that is MONOTONIC and exhibits a vote splitting effect.

   33% X>Y>Z
   32% Y>X>Z
   35% Z>Y>X

   The IRV winner is X. But if Z was not a contender, vote would look like:

   33% X>Y
   67% Y>X

   By being in the race, Z splits the vote with Y, causing Y to lose to X. Hence the bottom row of voters has an incentive to insincerely top-rank Y to get their second choice instead of their third.

6. I like your example, Clay. The ZYX voting block does actually hurt themselves by ranking Z first. Looks like they should have ranked Y first. The monotonicity issue isn’t present here because Z is not the one hurt by its raised ranking, so you’re correct. Thanks for correcting me.

   I can see this coming into play with the Oscars, a sort of bunching effect in the later rounds with the most popular movies. However, this goes back to the idea Dale raised in his blog. Are the drawbacks of IRV and its vote splitting acceptable for its tendency to select core support winners? Maybe this is something we want in this election.

   Clearly, this example bruises IRV. Y was even the Condorcet winner here. Perhaps a better question is, do you know how frequently IRV fails to produce a Condorcet winner? All systems have their failings, but it’s their frequency that’s important. I’ve finally found the answer to how frequently there is no clear winner in a Condorcet election: Collective Decisions and Voting, Tideman pg 119.

   In the same book Tideman mentions that from a real world sample it is rare that IRV fails to select the dominant winner (pg 194-195). He doesn’t do a probability function from all possible voting sets given different voters and candidates. However, one would guess that movie voting is likely to be close more frequently, which would make this failure more common. The fact that we have to guess at this for the Oscar scenario reinforces the frustration of the academy not releasing the ballots.

7. Aaron,

   Two of your comments caught my attention.

   1) “Are the drawbacks of IRV and its vote splitting acceptable for its tendency to select core support winners?”
   2) “All systems have their failings, but it’s their frequency that’s important.”

   Ah! Exactly! But it’s a bit more complex. It’s not just the frequency of failures that is important, but their _severity_. So suppose you could calculate how far a voting method is from being “perfect”, by summing up the frequency of its failures multiplied by the size of their effect?

   That is exactly what Bayesian Regret calculations are!
   http://scorevoting.net/BayRegDum.html
   http://scorevoting.net/UniqBest.html

   Only, they measure ALL failures simultaneously — even failures of criteria that have never been invented. And they use the REAL weight of those failures.

   To contrast with Tideman, his analysis involves arbitrarily select criteria, and giving them whatever weights (“importances”) he deems correct. Nevermind how OFTEN they happen with different voting methods, or how severely they skew the result.

   Warren Smith, the math guy behind ScoreVoting.net, does a pretty good job excoriating Tideman here.
   http://scorevoting.net/TidemanRev.html

   When you actually look at Bayesian Regret values, its clear that IRV is a pretty poorly performing system.

   As for your question about IRV’s tendency to elect Condorcet winners, see table 14 here:
   http://scorevoting.net/RandElect.html

8. Just wanted to say thanks for the links and thanks for your work in helping to develop such an informative site. After reading more and more about range/score voting, I find myself becoming an advocate for that system. This is after moving from IRV to Condorcet. For single winner elections I think Range is now my favorite. I know what to advocate for once I get out of law school (I went specifically for election law).

   Right now I prefer STV for multi winner elections and would like to bring it back to Cincinnati. I’m starting to look into RRV. I understand the argument for having a similar single winner system as a spring board for multi winner. Trying to keep an open mind. Though, I don’t like asset voting, but a creative idea (even if rediscovered).

9. Aaron,

   Good to hear. Sounds like you have a very open impartial way of thinking. I was a pretty staunch advocate of Condorcet myself when I first encountered Score Voting. One thing that helped to make the switch was this page, which says that SV may actually be a better Condorcet method than real Condorcet methods.

   http://scorevoting.net/AppCW.html

   Asset Voting is a little weird, but it has a huge advantage in terms of simplicity, and precinct summability — as opposed to STV and RRV for instance. It was originally proposed by an Oxford mathematician named Lewis Carroll (who wrote a certain story about a girl named Alice).

This site uses Akismet to reduce spam. Learn how your comment data is processed.