Sunday, January 27, 2008

How to vote in VFM - Interpolated Consensus,WTF?

Voter Funded Media, the contest that accompanied the AMS elections for the second year this year, is an idea that's meant to award media public funds by the will of the people, thus fostering better journalism, more informed voters, better elected leaders, and healthier democracies. The assumption is that media, as opposed to candidates or special interest groups, are able to engage larger audiences, since they are experts in communication. This, at UBC may or may not be the case, particularly considering the contest's mismanagement this year, but in any event, voting for this "race" of sorts is on now on WebVote until the 31st. There's an 8 thousand dollar prize pool that will be distributed among the various media according to votes. If you've enjoyed reading this blog, I encourage you to login and vote for us.

Before you do that though, you should find out about the voting system: "interpolated consensus". It's a tad complex, so just bear with me - and by the end, you'll know how to best allocate your votes! Alors, when you log into WebVote, you'll notice that for each media, you have the option of allocating to them $0, $500, $1000, $1500, or $2000. Lets do a simplified scenario: after everyone has voted, the votes are counted, and the median is determined. The median is a number which 50% of the numbers in a set are below, and 50% or the numbers in a set are above. It's the 50th percentile. The median will be one of the five amounts of money. If this is done for each media, you have an amount of money that each should receive. This is the "consensus" part of interpolated consensus. Taking the median, as opposed to the average, as a basis for awarding prizes is meant to discourage strategic voting - that is, it should encourage the voter to vote for the amount they actually believe that the contestant should receive. With averages, people are encouraged to engage in strategic voting (ie. voting above or below their real opinion) in order to "pull up" or "pull down" the average. If you use the median, on the other hand, the actual number you choose has no bearing on the amount that the media is awarded. All that matters is whether it is above, or below the median. Your vote will pull the median closer to your vote, whichever direction (up or down) that may be. You don't know. So if both 2000 and 1500 is above the 50th percentile of votes, they will both have the same effect on the outcome - by how much they are above the median doesn't matter.

Now, let's abandon our simplified scenario, and look at how it actually works. First, your five voting options represents a discontinuous set. That is, you're only allowed to award media in $500 intervals. To make the set more continuous, each vote for 500 is interpreted as 1/5 th of a vote for each 100-dollar interval between 300 and 700. Similarly, each vote for 1000 is interpreted as 1/5 of a vote for each 100 dollar interval between 800 and 1200. this is the "interpolation" part of interpolated consensus. If you didn't get that, just ignore it - it's a way of making the set of numbers more continuous. Second, we have to scrap thinking about the median (the 50th percentile) that we've been imagining. This is because taking the median of each media's votes and giving them that amount of money may not add up to the prize pool of 8000 dollars. In order for the system to actually allocate the prize pool, the percentile which will allocate exactly 8000 dollars is used - lets call this the pth percentile. This pthe percentile arbitrarily represents the "consensus" vote, and voting above or below it will change the prize for that media.

I have a few problems with this system. It seems to me that strategic voting is still possible: if you want to be sure that you'll have an upwards effect on a contest, always vote 2000. If you want to be sure to have a downwards effect, vote 0, or don't vote. If you want the media to get a specific amount, you should vote for that amount. The system will bring the pth percentile closer to the amount you chose, no matter if it's an up or down effect. The other problem I have is with using the pth percentile to determine how much money to give to each contestant, instead of using the median, scaled to $8000. It seems to me that it's quite likely to have a highly discontinuous set of votes with some media. This makes taking some percentile and awarding it highly arbitrary. It could jump from quite high to quite low as the result of a couple people that didn't vote (ie, voted 0) - or vice versa. If you're going to use a consensus system I think it makes much more sense to use a weighted median. I tend to think a voucher system with averages makes more sense to begin with, but that's just me. Thoughts?

For a more detailed explanation and simulation, go to VoterMedia.org