2017, eh? What a year. At least we have Eurovision for a brief respite from the horrors of international politics!
Ha, only kidding. The big news this year is that the Russian entrant, Yuliya Samoylova, has been banned from entering Ukraine, where the contest is being held, because she violated Ukrainian law by touring the Russian-occupied region of Crimea in 2015. Russia have pulled out of the contest for this year, and will be holding their own party in Sevastopol, to which the Ukrainians are not invited.
The lack of Russia will probably benefit the other former Soviet countries (ironically including Ukraine) - Russia usually acts as a magnet for their votes, which will be spread around more equitably this year. Semi-regular entrants Bosnia-Herzegovina are also missing this year, for financial reasons.
Returning after brief absences, we have Portugal with some old-fashioned crooning, and Romania with this delightful yodel-rap number:
Last year’s completely bananas changes to the scoring system are being kept, most likely because they’re precision engineered for maximum drama. For the full skinny on these, see last year’s run-down.
The model is mathematically the same as last year, but I’ve moved from a custom-built Gibbs sampler to using Hamiltonian Monte Carlo with Stan. This should make it a lot easier to experiment with different models. However, I spent all my time getting the new model working, so I didn’t actually do that this year. Oh well.
We have five terms in the model:
- How much the voting country “likes” the performing country
- How strong a typical song from the performing country is
- How strong an individual song is
- How “populist” a song is (i.e. how big a difference there is between televote and jury scores)
- Random noise
I’ve fit these to the data going back to 2009, when national juries were reintroduced to the contest. For some years, I only have “combined” voting data for televote and jury scores, so I dropped the populism terms for these years.
One hypothesis I suggested last time was that the new voting system might advantage “extreme” songs (i.e. those with either very positive or very negative populism scores). That effect does show up in the model results, but it’s very small, less than five percent. So we’re unlikely to see a shift towards more varied entries as a result. Of course, the real world may not behave in the same way as the model, or the EBU might rewrite the rules completely again, so take this with a pinch of salt.
The big question is, who’s going to win? The model is, as always, incredibly bullish about Sweden, with a probability of about 13%. For comparison, they’re currently about 7% implied probability on Betfair. By comparison, the overwhelming bookies’ favourite is Italy (model says 6%, about 55% implied probability on Betfair), although those odds have been lengthening over the last while, and the betting markets for Eurovision tend to skew towards western countries.
Portugal are also strongly favoured by the betting (17% implied probability). I’m not sure what’s going on here - the song is a bit meh, and it hasn’t even registered on fan votes like the OGAE. The model gives them less than 1% chance of winning, which might be a bit harsh, but they very much do not have good historical form - their last top ten finish was in 1996.
Other countries looking strong in the model are Australia, Ukraine and Azerbaijan. Australia are still a bit of a wild card - they’ve only been around for two years, so data is sparse, but they’ve come in the top five both times, so it would be foolish to count them out. Ukraine are the hosts - it’s unusual, but not unheard of for the host country to win. I haven’t been able to find evidence for a home field advantage or disadvantage in Eurovision. Finally Azerbaijan are probably being over-rated by the model, because of their run of high finishes from 2009-2013. Since then they seem to have stopped trying so hard, but the model doesn’t have a time-dependent component, so it hasn’t noticed this.
Semi-final 1 predictions
This is by far the stronger of the two semifinals: Sweden, Australia and Azerbaijan are all competing, and Greece and Armenia are both typically very strong (although less reliably so). On the other hand, things drop off quite quickly from there, so we’ll probably see a few countries coasting in on “bloc” votes (looking at you, Cyprus).
The model prediction here is easy qualification for Sweden (87%), Australia (81%) and Azerbaijan (80%), as well as most likely Greece (79%) and Armenia (70%). The next tranche of countries are Georgia (61%), Iceland (60%) and Albania (56%). At this point we’re getting into coinflip territory, but the best bets for the last two spots are probably Belgium (54%) and Cyprus (54%). Portugal, so beloved of the betting public, are at 48%, so slightly more likely than not to go out at this stage.