WITH Germany and Spain out and Russia still in at the World Cup, it is natural to wonder whether this year’s edition of the quadrennial football tournament has delivered an unusually high number of upset victories. After all, the Cup has repeatedly failed to follow the expected script in previous years. In 2014 Spain, the defending champion, suffered an early departure, alongside highly-rated England, Portugal and Italy. Back in 2002, Brazil had only the 13th-strongest team at the start of the tournament according to the Elo system, a statistical measure of performance. During 2001, they had endured a streak of just one win in ten matches. Nonetheless, they managed to win their fifth title. Alongside such unlikely past events, how does the 2018 tournament compare?

To find out, we built a simple statistical model to predict the outcome of each international contest since 1998. It consists of three steps. First, it analyses the historical relationship between teams’ Elo ratings —which account for both won-lost records and scoring—and their offensive output, in order to determine the probability of each team scoring each possible number of goals. Then, the model picks one goal total at random from this distribution for each team in each group-stage match, and records whether the result is a win, loss, or draw for each side. It repeats this process for the knockout-stage games, until one country is crowned the champion. Finally, it conducts these simulations thousands of times, and records the percentage of them in which each team advances to each round.

For example, before any World Cup matches had been played between 1998 and 2014, the model gave the ultimate champions a probability of victory between 2% and 25%—a testament to the inherent unpredictability of the tournament. For 2018, the model thinks Brazil is the most likely to lift the trophy, giving the team a 29% chance of victory in the final.

Though we don’t yet know the winner of this year’s World Cup, we can assess the odds the model has assigned to different events along the way. The probability that Germany would advance to the knockout stage, for example, was 92% before the tournament was played, making the chance of their early collapse just 8%. That was indeed the most unlikely elimination in the group stage in recent World Cup history. 

The model deems Germany’s elimination slightly more shocking than the two other big surprises of this tournament: that Sweden and Russia would make it to the quarter-finals (with probabilities of 10% and 8%, respectively). But if Russia were to win its quarter-final match against Croatia, that would be truly astonishing. At the start of the tournament the model gave this a 3% probability, but it is now given a one-in-three chance of occurring.

Overall, though, and despite Russia’s surprising success, the 2018 World Cup has so far been quite predictable when compared to past tournaments. Just two teams from 2018 are present on the list of the 30 most unlikely World Cup outcomes since 1998. At the other end of the spectrum, the 2002 tournament generated six of the ten most unlikely events in recent World Cup history—including Brazil overcoming one-in-50 odds of reaching and winning the final, and South Korea advancing to the semi-finals. Matches from 2010 make up six spots on the list of most unlikely outcomes, and the 2014 Cup generated seven. Indeed, if the 2018 World Cup continues as it has so far, the most unprecedented thing about it will be just how unsurprising it has been.