[EM] Continuous bias

[Juho Laatu]

On 16.5.2011, at 15.30, Kristofer Munsterhjelm wrote:

> Juho Laatu wrote:
>> The final of the Eurovision Song Contest of this year was held last
>> saturday. In the vote all countries give points to the songs of all
>> other countries (that made it to the final). The voting traditions
>> are a bit biased. Countries tend to give high points to their
>> neighbours or otherwise similar countries. Countries are not allowed
>> to vote for themselves, but minorities living or working in some
>> country may have considerable impact since they may have sympathies
>> also towards some other country. All this means that in addition to
>> voting for good songs people vote also for their best friends.
>> Eurovision Song Contest is a friendly competition though, and a major
>> carnival, and people don't worry too much about this kind of (well
>> known) voting patterns. Maybe they are just part of the fun and even
>> one essential part of the competition. But as a person interested in
>> voting I started wondering if this kind of voting patterns could be
>> fixed or eliminated.
> (...)
>> Would this approach maybe be useful and practical somewhere? What
>> other approaches there are to eliminate this kind of systematical
>> bias?
> 
> There's a problem with this sort of blind compensation, because the method itself can't know whether the bias is because a country is consistently good or because the other countries consistently favor that country.

If the country is consistently good, then all countries should give it lots of points. In that case the factors will remain low for all countries. They will get higher only if someone gives (continuously) more points to some song than others do. The given vote is thus compared to the average result of that song only (not to the average points of all the other songs, which is a constant number).

> 
> Say, for instance, that country X somehow gets very good at making Eurovision songs, so it wins a lot more often than would be expected by chance. Then your compensation scheme would make it harder for X to win; X is punished, ratchet effect style, for being good. It gets even more blurry when you consider that the countries reward each other according to "popularity" - perhaps the people of the Eastern European countries like the kind of music they themselves make, for instance, so that the "bias" is indirect rather than direct?

Let's say some country makes good songs, and it will get 12, 10 or 8 points from most countries every year. It gets maybe 10 points on average from all the other countries. In that situation a country that gives it 12 points gets a factor of 1.2 which is very low. So, support of good songs will not be punished (or only very lightly). On the other hand giving 12 points to a country that gets on average 0.5 points several years in a row yields a high factor (24). Voting for bad songs is thus a more likely way to gain high factors (for that country with bad songs).

It is true that the method to some extent punishes Eastern European countries for liking "eastern style" songs. Is not the intention of the method to punish for sincere musical opinions. Probably that factor is however not high if Eastern European countries support each others as a (reasonably) unified group. Within a group is is also not possible to give all countries 12 points in the Borda like method of the Eurovision Song Contest. Note also that the assumption that the Eastern European countries support their own songs more than the Western European countries do already implies that the Western European countries must prefer their own songs. They will thus be equally punished, which makes the method neutral again. The next problem is what happens if different "blocks" are of different size. In the case that the size of some (sincere musical) blocks is two, they will be punished more. But still they would (usually) be punished less than in the case of strategic support between the two countries since in the strategic (/friendly) case the quality of the songs would have no impact on the points given to each others. (The factors will be low if their points vary according to the quality of the song.)

The method thus relies on that it is not a common case that one country always likes the songs of another country (good or bad from and good or bad from the point of view of all the countries). Even if that happens, this probably does not have much impact on who wins. It would be however good not to unnecessarily reduce the points of any country. But it is not easy to separate strategic voting from sincere constant and song quality independent to some country.

In the Eurovision Song Contest countries tend to produce songs that are liked in all the participating countries (also this fact has been criticized). The Eurovision Song Contest thus does not probably suffer too much from this phenomenon. But there might be other elections where the grouping effect and candidates that are intended to please only a subset of the voting countries are more common and therefore more problematic.

I note that in the Eurovision Song Contest the Eastern European countries (that started participating the competition later, or that are numerous since they were formed from old larger countries (and that now have therefore "multiple votes")) have often been referred to as good examples of countries that often vote strategically. In my quick analysis the highest factors that I got did however not come from that direction. Actually, in one of my calculations the highest factors came from relationships Andorra --> Spain, Ireland --> United Kingdom and Monaco --> France, i.e. from the Western Europe. (Eastern Europe had its share of high factors too, but lower.)

> 
> I think the proper way to do this, if getting rid of bias were to be important, would be to make a video (or audio) recording of each country's song and then play it without saying what country it is. The countries then rate based on that alone, and the country names are revealed afterwards. However, there are many ways to "smuggle" information through audio and particularly video, so it would only weaken the effect. Besides, it would affect the circus aspect of the Eurovision Song Contest, and would be nearly impossible since the ESC has multiple rounds.

That would be a good approach for many elections. In the framework of the Eurovision Song Contest there are some additional problems here since in many countries the song of that country is elected in a similar public competition and public election. But if we discuss at general theoretical level, there are other elections where this approach would work very well.

> 
> Alternatively, one could use strategy-resistant methods: median ratings if cardinal, or something like the "IRV until there's a CW" method if ordinal, so that the actual effect of this kind of bias is weakened further. Borda is very manipulable, and I expect the Eurovision variant isn't far off Borda level, either.

Strategy resistance is a bit different case than resistance against continuous bias. For example if we used some IRV like method and we assume that to be strategy free enough, then we would still like to avoid one country ranking some other country always first. Ranking one's friend (even with bad songs) always first improves the chances of that country to win, and I wanted to eliminate that (or at least find methods that could be used to do that).

Juho


> 
> (I can't really see Eurovision doing the Condorcet-IRV method though: "Let's see if there's a pairwise champion among those who remain! No? Oh well, too bad, Germany: you're out!". :p)
> 
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20110516/4a968a02/attachment-0004.htm>