How was it for you?
For once, Tuesday night’s semi-final didn’t produce any major surprises, either from the perspective of the model, or for a human observer. The model’s score on qualifier prediction was a healthy 8/10. We missed out on predicting Estonia and Belgium, having Denmark and Moldova instead. None of these were particularly certain predictions - we gave Estonia about a one in three shot of qualifying, and Moldova about the same odds of getting knocked out. All in all, not a bad night for the model.
It’s possible that we could have done better here by taking into account the running order. Moldova performed first, which is generally held to be a bad position - many viewers will have forgotten about the entry by the time it comes to vote. The average position predicted for Moldova was about eighth place, so it wouldn’t have taken much to push them out of the top ten. Perhaps performing first cost them that qualification spot. On the other hand, this does little to explain why Belgium qualified after performing third.
Updated final predictions
Following qualification, Sweden are still top, but Greece has jumped to near the front of the queue. Things are probably too tight to call between them, Russia, Serbia and Sweden.
To be honest, I’m not really sure why Greece’s chances are rated so highly - Betfair have them trading at odds of 400 to 1. Greece are a bit bimodal in their performances - they either do really well or quite badly (relatively so - they’ve never failed to qualify). The most likely reason their chances are rated so highly is that the model has assigned a high probability to their having a decent song this year. That, plus a strong diaspora vote, would give them the win. Unfortunately for Greece, I’ve heard the song, and I doubt that’s going to happen.
Looking at Thursday’s second semi-final, there are a few obvious qualifiers. Azerbaijan have never failed to qualify, and can usually rely on votes from the rest of the former USSR. They also tend to get suspiciously high scores from countries with low populations, where buying votes might be more economical. This is reflected in the model by their having oddly high “like” scores from places such as Malta and San Marino. The model rates their qualification chances at 96%.
Norway (75%), Sweden (89%) and Iceland (90%) should all also qualify, as they’re likely to trade votes with each other. It typically takes around 50 points to qualify from a semifinal, so a pair of tens or twelves can get you well on your way. The model also has high hopes for Poland (79%), Malta (73%), Ireland (75%), and Lithuania (64%). Note that these four countries have a combined probability of overall victory which is well under 2%. They’re really just making up the numbers in the final.
Finally, the best guess is that two out of Cyprus (59%), Latvia (57%), Israel (57%), Portugal (55%) or Switzerland (54%) will qualify. Going by the numbers, the prediction has to be Cyprus and Latvia, but these are really in a statistical dead heat.