Understanding Eurovision tele-voting through modelling

Abstract

The Eurovision Song Contest has been held since 1956, providing European countries an opportunity to vote on other countries' songs. Voting has long been controversial, however, with accusations of unfairness and collusion. In this paper we model the tele-voting by members of the public in the most recent ( 2021) competition. Our goal is to determine to what extent a simple model with known factors can explain the results. Our model is based on five factors, three of which apply to some countries only:  Quality-based (Q) voting  Expatriate (α) voting (people in country A with ancestry from country B voting for country B)  Nordic (β) voting (some countries only)  Singer-based (γ) voting (some countries only)  Covid-based (δ) voting (some countries only)We simulated the voting for a sample of 15 countries. Table 1 summarises our results. Our main measure of performance was the mean absolute error in simulating actual tele-voting scores from the event. The mean absolute error was 20.7, with the best match to real data being Norway, and the worst being France.The performance of the model suggests that no large additional factors contribute to the voting outcomes, although a better fit to the real data would be obtained by explicitly modelling differences in taste between countries and by having a better model of how likely expatriate voters are to vote for their "home country."Mean 20