Measuring football – Miasanrot visits the NESSIS conference
Hanns Braun of Munich, an athlete and also footballer of FC Bayern, won the silver medal in the men’s 400m heat at the Olympic summer games in 1912 with a time of 48.3s, beaten only just by the American Charles Reidpath. Uruguay started the final of the 1930 football World Cup in a 2-3-5 formation, which was the prevailing formation at the time, and beat Argentina 4-2. Data has always played a very big part in sports because it allow insights into, and appreciation of, sports events that one did not witness first hand, even if they happened a long time ago.
Data analytics helps to better understand the flow of games, find patterns in actions, be more effective at coaching, and more accurate at recruiting. In the US, complicated mathematical and statistical analysis has long found its way into sports, familiar to a broader audience perhaps from the “Moneyball” phenomenon of the Oakland Athletics in the late 1990s.
Data driven analysis of sport competitions has led to a number of innovations in sport, such as the best way to use a relief pitcher in baseball, when to go for a fourth down in American football, and how to most effectively use three point throws in basketball. Not least the success of teams that rely heavily on statistics and analysis – for example the teams from Philadelphia and the New England Patriots – has been fuelling the hype.
Nevertheless, it would be imprudent to believe that data analytics were a fast track to success. It is much rather about decreasing the likelihood of mistakes and increasing the chances of finding a winning strategy.
Around 250 people attended the conference, many undergraduate and graduate students, post-docs, and professors among them. Another big contingent were analysts and representatives of big American professional sports teams, and the sports media. European football was officially represented by William Spearman, a PhD in physics and the “Lead Data Scientist” of Liverpool FC.
The conference showed how closely science and sports are interrelated in the USA. Whereas it was not too long ago that Ralf Rangnick was still ridiculed as a “professor” in Germany, it is completely normal for a man like Barry Nalebuff, a professor of game theory at Yale, to be an integral part of the community and hold the keynote address at the conference. The talks at the conference were generally quite academic in content and terminology, which would have made it difficult to fully appreciate them without a background in mathematics or statistics.
Opta event data and related analyses like completed passes, pressing intensity, expected goals and so on are box standard in analytics circles. Tracking data, however, is the latest fashion in the community. They have been used in the major US team sports for a number of years already, and now they find their way into association football. The idea is to minutely track every player with a chip to gain information about their absolute and relative positioning on the pitch throughout a match. Here is an example:
The gif from Ron Jurko shows a NFL running game. The blue team has the ball and plays from right to left in trying to reach the end zone. The running back in possession is marked in black. The special gimmick of this analysis is that Ron has highlighted in red the various probabilities where the play could end. The more intense the colour, the more likely it is that the play will end there.
Liverpool is currently trying to transfer this kind of analysis to association football. William Spearman explained that their efforts have been about trying to identify the probability of scoring or conceding a goal depending on the positioning of their own players and the opponent’s players in different areas of the pitch. He gave this fictitious example: if Coutinho has the ball in zone 14 and no opponent is closer than 90cm to him, and there are three opposition players and two teammates placed between him and the goal, then he has an 11% probability of scoring if he takes a shot.
These complicated analyses are studiously observed by Liverpool’s coaching team including Jürgen Klopp, who has a keen interest in data driven analysis.
The Houston Rockets changed NBA basketball forever with their radical approach to focus almost exclusively on three point throws. There has not been anything as ground-breaking in association football yet. Instead, the analytics departments’ efforts have been aimed at improving on the details of the game, up to such seemingly minute aspects as the perfect formation for an opponent’s goal kick.
Liverpool hired Thomas Gronnemark, a Danish throw-in coach before the season. On average, teams remain in possession only 50% of the time from a contested throw-in. Liverpool’s figure even was considerably worse last season. Gronnemark managed to raise this quota to 68%, a substantial improvement and the second best value in Europe. The first place is taken by FC Midtjylland, Gronnemark’s previous team and one of the forerunners and primary advocates of football analytics.
Gronnemark does not simply train the throw-in itself but short liberating movements and even entire plays. Liverpool’s opening goal in the Champions League against Salzburg, for instance, resulted from a throw-in in their own half: it took the team 7 passes and 9 seconds to outplay Salzburg’s throw-in pressing and score through Sadio Mané.
Tyler Heaps, Manager Analytics of the USMNT, illustrated several applications for heat maps. First, he analysed 10,000 penalty kicks of his team to find out where the shots went at which rate of success. In doing so, he and his team were able to identify several sweet spots where no goalkeeper had ever been able to stop a shot. Then he did the same for the penalty kicks against his team. The USWNT was able to profit from both these analyses during the recent World Cup in France. Alyssa Naeher parried a penalty against England in the semi-finals and Meghan Rapinoe scored a penalty for the 1-0 goal in the final against the Netherlands.
While Heaps’s analysis was reminiscent of Lehmann’s famous note at the 2006 World Cup, some of the other developments he presented at the conference appeared like real innovations: while corner kicks are usually defended well initially, resulting second balls are yet an understudied phenomenon. So Heaps’s analysis has concentrated on trying to find out where the second ball after a corner most probably gets depending on how the corner kick is taken in order to position the players accordingly. As a result, corners are not necessarily taken with the aim of directly converting from them any longer, but so as to maximize the probability that they end up in an interesting area of the pitch. This idea is easily transferrable to crosses, which might breathe new life into what is often a rather uninspired crossing game at present.
To conclude, our visit to NESSIS allowed us to gain a unique insight into the world of sports analytics and will surely not have been our last excursion into the matter.