Eurovision, a major actuarial exercise?

Historically, Australians love a good bandwagon which allows us to be patriotic, invent drinking games, and watch the UK getting crushed. For these reasons, the Eurovision Song Contest has been a huge hit in Australia over the last 34 years. Bizarrely enough, Australia has been invited to participate in the contest since last year, despite still being well within the Southern Hemisphere.

This year’s winner came from the Ukrainian entry 1944, about the Stalin-era deportation of Crimean Tatars into Central Asia – narrowly beating out runner-up Australia, represented by the now-world-famous Dami Im, who wowed the international audience with her song Sound of Silence.

So how does the voting work? The contest starts with 42 countries, with only 26 qualifying for the Grand Final. Each of the 42 participating countries submit two sets of ranks for the finalists (not including themselves); one set of points is taken from their local 5-member jury, and the other comes from their public vote.  The 84 sets of points are then combined in order to crown the winner of this annual glitterstorm.

Every year, the voting system comes under scrutiny of being perverted by political favouritism, jury corruption and poor judgement. With the power ogranular data released by the European Broadcasting Union (10,716 independent data points from this year), generally good but potentially spurious statistics, and quite frankly, a lot of spare time, I hope to address some of these claims.

So, get your wind machine and a table of Critical Values of your favourite statistical distribution: it’s time to begin.

1. The juries dragged public-favourites Russia down.

Amongst public voters, Russia was the clear winner – 30 of the 41 countries’ televoters placed Russia in their Top 3 (from 26 finalists), and every country’s public vote awarded Russia at least 3 points. In contrast, juries were far less favourable: more than half of the juries awarded Russia 0 points.

But, the worst jury treatment for Russia came from four of the national juries who slapped Russia in their bottom 5 despite favourable public votes. Three of these juries happen to also be from the three European countries most recently invaded¹ by Russia:

Ranking out of 25 given to Russia by each recently invaded country’s Jury and Public.

The fourth jury to downvote Russia was Lithuania, with four of the five jurors ranking Russia in last place, which seems like a bit of history repeating.

Ranking of Russia (out of 25 finalists) over the last two years by Lithuania.

2. Europe has been enlisted in the Dami Army.

Australian representative Dami Im, who charted well around Europe, would have won based on last year’s voting system. Points are given only the top ten acts ranked by each cohort. In the former system, a “cohort” was a country (with jury and public votes aggregated), but this year, the public votes and jury votes of each country were treated as independent cohorts. As Ukraine’s song was divisive between the juries and the public, it would have received less points under the former system.

Number of cohorts giving Australia, Ukraine and Russia points, segmented by Jury and Public votes (41 Jury Cohorts, 41 Public Cohorts).

Despite this, Dami still managed break into the some of the strong voting blocs, securing Top 2 rankings from countries in most historical blocs: the Balkans, the Warsaw Pact, Scandinavia, the Baltics, the former Soviet Union, and the Mediterranean.

3. Statistically, the Australian public is not the European public.

Why bother with basic geography when you have p-values? At the 95% level of significance, when all 42 countries’ public votes were combined, the only individual public vote statistically uncorrelated with the rest was Australia (r=0.36; Critical Value of r=0.39). Below are the four weakest correlations and a fact about each country.

From the five Australian jurors, most were significantly correlated with the votes of the Australian public, with Australian Idol runner-up and cult hero Shannon Noll being the least correlated with the Australian public vote (r=0.25; not significant).

4. The public ranks male contestants based on their looks.

Who doesn’t love fit-for-purpose (invented) statistics? This is captured by a statistic called the IBOGL Index (Influenced By Others’ Good Looks, and pronounced “eye-boggle”), which measures the tendency of a voting cohort to rank performers on their looks. This is calculated by stratifying a cohort’s voting ranks by gender, and then using a Spearman Rank Coefficient for each segment’s votes against a proxy for the performer’s attractiveness. I’ve used results from a fan poll³ called “Eurovision’s Next Top Model” as the baseline for this – fortunately, it carries a decent sample size of 66,000 votes, and is pre-segmented by gender.

Number of cohorts with statistically significantly IBOGL indices (Critical Value of IBOGL Spearman Coefficient is 0.53 at the 95% level of confidence).

Almost half of all voting countries’ publics exhibited statistically significant IBOGL indices for ranking male contestants, indicating that these public cohorts were ranking male contestants based on attractiveness. Zero public voting cohorts judged female performers in the same way, which might speak to the general demographic of Eurovision fans. Juries hardly exhibited this type of voting pattern for male or female performers.

5. Serbia was the biggest beneficiary of bloc voting this year.

Serbia’s 18^th place result was its worst in a Eurovision Final ever, but this was boosted by strong jury ranks and unanimous first place rankings from all former Yugoslav republics:

Rank given to Serbia by each voting cohort

During the first semi-final, the public voters of Greece and Cyprus mutually ranked each other in first place. Only Cyprus managed to qualify for the final, where again, they received top points from the Greek public and no one else, leaving Cyprus in 21^st place overall.

6. The Swedish jury somewhat didn’t agree with their public.

Sweden’s massive impact on the music industry globally was visible in this year’s competition, with 10 other countries’ songs being written or produced by Swedes (on top of Sweden’s entry). The below table shows the Swedish public and jury votes for three of these countries – Czech Republic, Azerbaijan and Malta.

Essentially, the Swedish public’s bottom 3 were ranked in the top 8 by their jury.

After stripping these three anomalies from the Swedish votes, the correlation coefficient between the Swedish jury and Swedish public significantly jumps from 0.14 to 0.51 (Critical Value is 0.39), which would then make the two cohorts significantly correlated at the 95% level of confidence.

7. The Czech Jury statistically significantly ranked performances the opposite of their public 

That’s right – if you reversed the Czech jury votes, they would significantly predict the Czech public vote at the 95% level of confidence (r = –0.41).

But, as the Czech public was significantly positively correlated with the rest of Europe (r=0.62), this indicates that the jury was statistically unaligned with most other voting cohorts. The table below shows that the Czech public’s top and bottom three are somewhat aligned to the final place given by all countries, but the Czech jury votes are in a very different direction.

8. The juries – not the public voters – downvote non-English songs.

Only Bosnia and Herzegovina, Macedonia, and Austria performed with entirely non-English songs. In each semi-final, whereby only the Top 10 of the 18 semi-finalists qualify for the Grand Final, juries showed a clear aversion to these three songs:

Ranking (out of 18) segmented by public and jury votes during the Semi-Finals only. Green highlighting indicates a Top 10 (qualifying) rank.

Based on the public votes alone, all non-English songs would have qualified; based on the jury votes alone, none would qualify. Only Austria – who strangely enough, sang in French – managed to qualify for the Grand Final once the votes were combined.

Furthermore, these songs were ranked higher by juries from countries where the language has some recognised minority or official presence:

• Bosnia and Herzegovina sang in Bosnian: a recognised minority language in Croatia and Montenegro. Both of these countries contributed 10 of the 26 jury points given to Bosnia and Herzegovina in Semi-Final 1.

• Macedonia sang in Macedonian: a recognised minority language in Serbia and Albania. Both these countries contributed 24 of the 34 jury points given to Macedonia in Semi-Final 2.

• In the Grand Final, sole qualifiers Austria received a strong 8^th rank from the public; far from the harsh jury rank of 24^th (out of 26). Interestingly, from the few jury points received, the juries from all three officially French-speaking participating countries – France, Switzerland and Belgium – ranked Austria’s French-language song strongly:

French-Official countries’ ranking of Austria’s song: green highlighting indicates jury members with significant French-language media presence²:

The above table is a breakdown of each juror’s vote, and individual jurors who have a significant French-language media presence² are marked in green, who collectively appear to rank the song more favourably.

9. The juries continue to downvote Poland.

It’s hard to develop a statistic or index to explain why this trend has continued for the last three years, so here’s a table to show the difference between public and jury ranking of Poland:

10. Eurovision might just be a major actuarial exercise.

At this point, you might be mistaken into thinking that a European Call Option is the phone number to vote for another country’s song. Fear not: for next year, I’m calling on the Actuaries Institute of Australia to form a Specialist Sub-Committee and commission a Research Paper into optimising Australia’s entry. Strobe light flashes should follow a standard Brownian Motion process; ideal on-stage costume changes could be optimised by a Markov Chain; and the words “love” and “peace” arrive in song lyrics according to a Poisson Process. But no matter how many times a Stochastic Stiletto Simulation is run, there is one limiting factor that can’t be avoided…Australia still isn’t in Europe.

CPD: Actuaries Institute Members can claim two CPD points for every hour of reading articles on Actuaries Digital.