IMO

My Oscar picks hinge on Avatar winning Best Picture. Somehow everyone else in my pool picked The Hurt Locker. But I would still be very happy to see Hurt Locker take the prize. I have seen 8/10 of the nominees and loved every one of them.

If one watches Julie and Julia (best actress nominee, Meryl Streep) followed by Food Inc. (best documentary nominee), will one want to eat food?

Instant Runoff Voting will replace Plurality voting for the category of Best Picture. “This year, for the first time since 1944, there are ten Best Picture nominees. And that means the return of the “preferential voting” system also known as Instant Runoff Voting (IRV).” [chicagoist]

IRV and the Spoiler Effect

Instant runoff voting has many nice properties. Some of which may be even more desirable as the number of candidates (nominees) increases. The chicagoist cites IRV proponents, FairVote, and claims that IRV avoids “spoiler effect”. This is misleading. IRV can sidestep spoilers in some special cases but is not immune to them.

The spoiler effect is technically referred to as a violation of Independence of Irrelevant Alternatives (IIA). Your preference of A over B should be independent of the existence of choice C, where choice C is the “spoiler”.

Many claim that Green Party candidate Ralph Nader “spoiled” the 2000 election for Al Gore. It is quite likely that George Bush won despite the fact that a majority of voters preferred Al Gore, and that Bush would have lost if Nader was not on the ballot at all.

The chicagoist article puts it another way, pointing out that “a movie that gets the most “first choice” votes could lose out to a movie that gets more.” This is true. For simplicity, let’s consider only the three nominees with the best odds of winning - Avatar (A), The Hurt Locker (HL) and Up In the Air (UitA). Imagine the following list of ballots (where [#] is the number of ballots of that type).

1      A        HL     UitA
2      HL      A       HL
3      UitA   UitA   A
[#]    [5]      [4]     [2]

Avatar is the winner under the old plurality system (with 5/11 first place votes). However, The Hurt Locker is the winner under IRV (UitA is eliminated with only 2/11 first place votes. Then HL wins with 6/11 votes). IRV seems justified here, given the fact that a majority of voters preferred HL to A. Avatar wins under plurality only because of the existence of UitA (a violation of IIA, a “spoiler effect”).

Despite success in a few special cases, instant runoff voting is not always independent of irrelevant alternatives. Counter examples are not difficult to find. FairVote seems to know this. Their website is careful to only say that IRV “assures the so-called “spoiler effect” will not result in undemocratic outcomes”. This weaker, more ambiguous statement is completely dependent on the definition of “undemocratic”.

IRV and monotonicity

What should not be overlooked is the fact that IRV has a very serious flaw. Let us first define a very basic and desirable property of any voting system.

A candidate X should not be harmed if X is raised on some ballots without changing the orders of the other candidates.

This is the property of monotonicity. We can quickly understand why one would want their voting system to satisfy monotonicity. It would be foolish for the winner of Best Picture to have to give up their Oscar because some number of voters raised the winner on their ballot without changing the orders of the other nominees. Put simply, the winner should not become a loser after receiving even more favorable votes.

Instant Runoff Voting fails monotonicity!

I will of course supply an example. Again, lets confine ourselves to three nominees. Using all ten nominees does not change anything. It only complicates the example. Initial ballots are cast as follows.

1      HL      A        A       UitA   UitA
2      UitA   UitA    HL      A       HL
3      A        HL      UitA   HL      A
[#]    [6]      [2]      [3]     [4]      [2]

After all 17 ballots are counted, IRV declares The Hurt Locker the winner of Best Picture. Details: No nominee had more than 50% of 17 first place votes. Avatar is eliminated for having the fewest first place votes (5). Then HL is the winner with 9/17 first place votes.

Now suppose two voters change their ballots, moving previous winner, The Hurt Locker, above UitA and keeping everything else the same. The resulting ballots are displayed bellow.

1      HL      A        A       UitA   HL
2      UitA   UitA    HL      A       UtiA
3      A        HL      UitA   HL      A
[#]    [6]      [2]      [3]     [4]      [2]

Now Avatar is declared the winner! Details: No nominee had more than 50% of 17 first place votes. UitA is eliminated for having the fewest first place votes (4). Then Avatar is the winner with 9/17 first place votes.

An absolute riot would ensue. The media would be scrambling to understand and explain voting theory and IRV would be given a second look. You may say, “that example is very contrived and unlikely”. You may be right. It is certainly contrived (for ease of explanation). Likelihood is difficult to measure, but it has happened.

Past failures of IRV

Recently Andy Montroll lost of the 2009 Burlington Vermont 2009 mayor election that he could have won if 753 of voters had lowered him on their preference lists. This crazy election was also a real-life counter example of a spoiler in an IRV run election. Montroll would have handily won a head to head vote against any of the other 3 candidates. He finished 3rd in the IRV results.

FairVote declared the Burlington IRV run election a success.

IRV at the Oscars

Unfortunately full ballots are not published in the case of Oscar voting. We will probably never know whether a loser of the Academy Award for Best Picture was victim of a IRV’s susceptibility to spoilers or non-monotonicity.