##What’s new in 2015?
There are two main changes to how the model works this year, and one notable change to the contest.
Firstly, I’m no longer using betting odds to determine song quality. Although this was a really appealing idea, it didn’t really work very well last year. The resulting predictions were both wrong (not necessarily a problem) and far too certain (definitely a problem). This was particularly a problem in the semi-finals, where the model’s confidence seemed entirely misplaced. I tried some tweaks to reduce this problem, but the best solution seemed to be to remove it completely.
The second change, introduced partly to compensate for the first, is that I’m now using information about each song’s audio properties, which I obtained from The Echo Nest. This means that the model now includes interactions like “Armenia likes dance music” and “Austria hates sad songs”. The effects of these preferences seem to be small, but they’re interesting and data-driven, so I’ve included them.
The change to the contest is the somewhat controversial inclusion of Australia as an entrant. Eurovision has been popular in Australia for many years, and for the 60th edition, it was decided to allow them to enter. From a modelling perspective, this is a bit annoying, because we have very little historical data on Australia.
I say “very little”, rather than “no data”, because the Australian broadcaster SBS has been running an internet poll for the last five years, which means that we do have some idea of how the Australian public is likely to vote. In general, they don’t seem to have very many strong preferences. Their only strong relationship seems to be with Ireland, which they rate at +1.7, roughly equal to the positive relationship between Malta and the UK, or between Finland and Norway. It’s fairly likely that this is related to the large Irish immigrant population in Australia.
In the other direction, however, we have no information. We might imagine that Australia would get a boost from English-speaking countries like the UK and Ireland, but we have no data to back this up. As such, these effects are not included in the model.
We’ve also got the usual shuffling round of entrants. Perennial powerhouse Ukraine have withdrawn, due to the ongoing conflict. On the other hand, Cyprus, the Czech Republic and Serbia have rejoined the contest. Serbia have historically been one of the stronger countries, but without most of their posse of former Yugoslav vote donors, it’s quite likely that they won’t be able to reach those heights again. Both Cyprus and the Czech Republic have fairly dismal histories in the contest, and the only notable thing about them is that 12 points from Cyprus to Greece is one of the most predictable things about Eurovision.
So, the model likes Sweden (as always), and they’re currently also the gambler’s favourite, running at odds of 2.94 on Betfair, for an implied win percentage of 34%. The gamblers are also quite positive about Italy (odds of 4.2, or 24%), but the model doesn’t rate them very highly. Italy are relatively new to the contest, and the model has had high opinions of them in the past, but they performed fairly abysmally last year, so their chances have been adjusted downwards to about 1%.
The third favourite from the gamblers is newcomer Australia (odds of 11.0, or 9%). The model has them rated somewhere towards the middle of the pack (2.5%). I’d be inclined to trust the gamblers more than the model here, as the model is operating on very little information.
Overall, it looks good for Scandinavia, with Sweden, Norway, Finland and Denmark all in the top ten most likely to win. Of the automatic qualifiers, Austria are best placed to win, but they’re also the most likely to collapse completely - the model gives them a 21% chance of achieving nul points.
##Semi-final 1 predictions
Looking at the first semi-final, the model rates this as the stronger of the two, with a 49% chance of producing the winner (versus 35% from the other semi-final and 16% from the automatic qualifiers). The gamblers, on the other hand, have it pegged as the weaker of the two, but the market is quite illiquid. In any case, there are quite a few of the usual strong contenders here.
We can probably expect the usual safe qualification for Greece and Armenia (98% each), Russia (91%), Romania (90%), and Serbia (80%). Beyond that, Moldova (70%) will probably benefit from sharing a semi with Romania. Denmark (66%) will probably also qualify despite a Scandi-weak semifinal. The list is filled out with Georgia (65%), Hungary (64%) and Albania (55%). If one of these fails to make it, the most likely substitute is Finland (51%).