Oscar best picture changed from single vote to full preferential ranking | Hacker News Reader

Oscar best picture changed from single vote to full preferential ranking

carpetbagger.blogs.nytimes.com

37 pointshuangm26 comments

When reading about voting systems, I'm always reminded of Arrow's impossibility theorem, which everybody should be aware of:

http://en.wikipedia.org/wiki/Arrow%27s_impossibility_theorem

In short, no "fair" voting system can satisfy these three criteria:

- If every voter prefers X over Y, then the group prefers X over Y.

- If every voter prefers X over Y, then adding Z to the slate won't change the group's preference of X over Y.

- There is no dictator.

Never knew about this - thanks for the most interesting read.

Is the 2nd factor a problem with the voting system or with the psychology of the voters though?

This is a desirable property of a voting system. If we prefer X over Y (X > Y), then adding Z can give: Z > X > Y, X > Z > Y, or X > Y > Z. Adding Z ought not change the ordering of X and Y.

And to relate specifically to Instant Runoff Voting, IRV fails to satisfy monotonicity. It also fails to satisfy the Condorcet Criterion (a commonly held desirable property that the person with the majority #1 votes should be ranked 1st).

There exist `better' ranking systems such as ones involving markov chains (http://www10.org/cdrom/papers/577/), but I'd imagine that the Oscar Committee requires their voting/ranking scheme to be palatable to the general public. Using IRV is a good step in the direction of `better' voting, while still being explainable in a few sentences.

Slight nitpick: the Condorcet Criterion doesn't actually specify that the one with the majority #1 votes is first ranked. It explicitly states that the candidate that, when compared pairwise with each other candidate, wins most frequently should win overall.

http://en.wikipedia.org/wiki/Condorcet_criterion

That can't be correct as written because if I could magically extract everyone's scalar utility function from their minds, a mixture of utility functions would satisfy those three criteria.

You need an additional requirement, "The citizen's true ordering of preferences is the only data presented to the algorithm." In other words the algorithm can't access the scalar utility function, only the ordering, and every similar ordering has to result in the same social ordering - no mixture of utility functions will satisfy this, obviously. If you want to prevent "strategic voting" that would impose other requirements as well.

Impossibility theorems often aren't as impossible as they look, and indeed Arrow's Impossibility Theorem makes a good illustration of that, especially as the evasive answer is pretty obvious if you walk around thinking of expected utility maximizers all day long. Doesn't work for non-mind-reading applications because of the strategic voting problem, but in terms of whether it's possible for a society to aggregate the true utilities of its citizens, no sweat.

Of course, there are other specifics to the formal definition of Arrow's theorem. It's limited to situations where people express preference rankings -- as if on a ballot they fill out.

It's not meant to describe situations where their perfectly-weighted preferences can be measured -- as if by some mind-reading utility-function-extraction-and-summation process. (Can two persons' Utils be meaningfully compared or summed?)

If you could magically extract everyone's scalar utility function, then it wouldn't be a voting system.

The requirement that you state is in the definition. It's just implicit in the word "voting" rather than explicitly stated. Arrow's Theorem does rely on an explicit definition of voting, which is that voters make discrete choices about their candidates subject to certain criteria. Some theorem (possibly part of Arrow's Theorem) says that it's not possible to eliminate the potential for strategic voting in a 3+ party system without negating other desirable qualities like those listed above.

Looks like they're using instant runoff voting: http://en.wikipedia.org/wiki/Instant_runoff_voting

Boo. IRV sucks (probably as much or more than plurality). Approval voting FTW. See http://zesty.ca/voting/sim/. Noice how Approval and Condorcet voting produce the same pictures (aside from Approval having a little fuzzier edges).

Boo. Approval sucks. Schulze method for the long haul. Majority criterion is fairly important, yeah? Approval (aka restricted range voting) fails that.

On a side note, I've been working on a JSON web service to calculate election winners in various systems. Feedback would be sweet. http://vote.cognitivesandbox.com

IMHO, Schulze is too complicated for the average voter. People need to understand the voting system. I don't deny that it's theoretically better.

I heartily disagree that people need to understand the voting system. Ideally, people should just be thinking about ranking the candidates. If the way that the votes are counted will affect the way they vote, that means that strategic voting is a prominent part of the system.

Granted, strategic voting is theoretically unavoidable without introducing other degeneracies. But on a quantitative scale, its effect and its frequency are smaller under some voting systems than others.

http://en.wikipedia.org/wiki/Arrows_impossibility_theorem http://en.wikipedia.org/wiki/Gibbard–Satterthwaite_theorem

People generally don't understand how their household appliances work. They only care about understanding the interface (e.g. open fridge door, put stuff in, stuff stays cool).

Interesting visualizations, but do the problems mentioned have any relevance for real elections? What happens if you move from two dimensions down to one, or up to many dimensions? While some of the pictures look ugly, does any IRV scenario actually result in a disastrous outcome -- ie a candidate with views far away from those of the electorate being able to win?

Approval voting seems like a worse idea in practice than in theory. I can easily imagine scenarios in which an unknown (and quite possibly incompetent or extremist) candidate with an attractive-sounding party name could garner enough votes to come out ahead (something like this recently happened with the Family First Party in several Australian elections).

That is a really fantastic visualization. It's amazing how much IRV produces inconsistent results.

The important question is: how likely is this to change the results? In recent decades there's been a trend for the Best Picture Oscar to wind up going to films which may not be all that good but which are seen as "Oscar-worthy" due to weighty subject matter or otherwise "seeming smart". Historical dramas have made up something like twenty out of the last thirty years' winners, including such so-so films as Braveheart. No comedy has won since Shakespeare in Love (which was acceptably smart-seeming because it had Shakespeare in it) and before that it was Annie Hall (which, incidentally, beat Star Wars).

I wonder if preferential ranking will improve the quality of winners by better reflecting the true preferences of the voters rather than the preferences the voters seem to think they ought to have.

I wonder how long this will last. "Fair" voting approaches such as preferential ranking attempt to address the challenge of producing an outcome that is agreeable to everyone, which is subtly different than recognizing the "best". This will favor a compromise result over one that is challenging, thought-provoking, or "artistic" in other meaningful ways.

What kind of voting system would result in correctly recognizing "the best" in your opinion? And how does "the best" film differ from the compromise candidate?

In my opinion, the "best" work also scores highly across the criteria. In my experience, the voting body increasingly ignores criteria as it gets larger. Perhaps the quality of the results would improve if the voting body had to rank the criteria to provide weights, then rank their preferences for each criteria. This informs the voting body and the results might surprise (and even please) everyone. But that's complicated.

Having designed and run preferential ranking systems for selecting multiple winners (unlike the Oscars), I've seen that polarizing choices tend to get eliminated regardless of artistic merit. Sometimes, we address this by letting the results of the larger voting body inform the decision of a smaller panel, who are allowed to vary the algorithm and play with other variables before making the final determination (in ranking subjective things like art, fairness in voting is not always the main priority).

The problem with the compromise candidate is that it is often not anyone's first preference, therefore nobody considers it the best. This is especially true when the number of candidates is large.

As an experiment, I'd like to ask voters to select only the best and worst single candidates, then apply a straightforward preferential ranking algorithm against those as if they were the top two choices. I wouldn't be surprised if this would narrow the field to works with the most artistic merit.

I've been convinced that range voting it the best -- better than IRV and approval voting. It's intuitive also. See http://en.wikipedia.org/wiki/Range_voting

Range voting is one of the few systems that requires strategic voting with only two candidates. This isn't a good thing. :/

  Demographic A (10 voters): Candidate X valued at 40, Y at 60
  Demographic B (3 voters): Candidate X valued at 100, Y at 0

  Candidate X gets 700 points
  Candidate Y gets 600 points

This scenario demonstrates how susceptible range voting is to tactical voting.

I always wondered what HN would look like if we used preferential ranking, and almost started to work on something earlier this month.

How does that work when we get different stories at different times? You might be able to come up with a system in which everyone votes on the day's top stories, and then they're ranked, but that would be tedious and would not clearly produce a better result than what we have now.

Not saying any of this is necessarily a good idea, but to explore an interesting vein of ideas:

What if the front page weren't continuously updated, but only updated after fixed balloting periods (say, 6-12 hours long)? Voting would only happen when all stories for a period were in. Each voter would see a different randomized order-of-presentation.

Yes, it'd completely change the character of the site. No immediate upvote feedback for being first to submit a 'hot' story. Stories would be guaranteed to lag other 'breaking' news sites by some number of hours.

But these changes might be good things: reducing both the incentive to submit, and the rewards for constant reading, of frothy/gossipy linkbait.

A site seeking impressions over all else wouldn't want this approach -- but HN isn't ad-driven. 'Slow HN' might be a better HN.

Reply on news.ycombinator.com