Performance analysis in elite football: all in the game?

The beauty of football is hidden in its complexity. To be able to perform on an elite level, excellent physical, technical, and tactical skills are required (Impellizzeri and Marcora 2009). These skills need to be expressed in the context of the game to beat the opponent. Traditionally, these skills are assessed using isolated lab or field tests for the purposes of talent identification (Huijgen et al. 2014) or evaluation of training interventions (Brink et al. 2010). Whether outcomes of these isolated tests truly reflect match performance is under constant debate. With the development and use of sensor technology in matches, the question arises if match analysis itself can serve as an alternative for isolated tests. Historically, lab tests were developed to assess physical capacity of football players (Reilly 1995). These test protocols were often derived from endurance sport and executed on a bicycle ergometer or a treadmill. Incremental and continuous protocols were used to assess physical capacity. The main advantage of these lab tests is standardization. However, these tests lack sportspecificity and do not mimic the intermittent character of the game. Next, the type of activity is straight running, while decelerations, accelerations, and pivoting on turf are important characteristics of football. Finally, lab tests are often time-consuming and usually limited to one player per test. To tackle these disadvantages, field tests were developed, like Yo-Yo intermittent recovery test, Interval Shuttle Run Test, and 30-15 fitness test (Krustrup et al. 2003; Lemmink et al. 2004 ; Buchheit et al. 2008). Concurrent validity was assessed with comparison to traditional lab tests and construct validity to discriminate elite from sub-elite players. This supported the use of field tests in an applied setting. Since the development of these tests 15 years ago, they have found their way from professional football to amateur level across the globe. Similar to the development of tests to assess physical capacities, isolated tests were developed for technical skills, such as the Loughborough passing test (Ali et al. 2007). Tactical skills were traditionally captured by observation, but tests for decision-making skills are now also integrated into computer and tablet software or virtual reality environments. These isolated lab and field tests in the physical, technical, and tactical domain try to simulate the match as closely as possible. However, with the development of sensor technology available during matches we could also turn this around. So the question is, can we use outcomes from match analysis to assess physical, technical, tactical skills? From matches, one could derive physical performance based on distance covered, number of sprints, acceleration and deceleration profiles, and directional changes. It is expected that over time,more andmore relevant physical performance indicators will be developed to track both external and internal load during match-play. Inertial measurement units have the potential to determine the mechanical load of the lower extremities but also capture technical performance indicators like ball control, passing and kicking (Blair et al. 2018). Lab-on-a-chip development in healthcare can find its way to the sport field and quantify internal load based on skin temperature and sweat loss. Finally, tactical performance indicators, like passing efficiency, in combination with spatiotemporal features, such as space control and putting pressure on the opponent, may lead to more advanced measures in this domain (Rein and Memmert 2016). However, it is known that large match-to-match variation exists as a result of contextual factors like playing strategy and formation, strength of the opponent, score line, and environmental conditions (Kempton et al. 2015; Carling et al. 2016). This variability tends to be highest for high intensity activities such as sprinting and lowest for global measures like total distance covered (Kempton et al. 2015). To control for some of the contextual factors, one could think of standardized (small sided) games (Rowel et al. 2018). It is known that these games are closely related to the actual match, but the strategy, playing formation, and set-up can now be standardized (Olthof et al. 2018). Besides, advanced data science methods, like machine learning, can help to control for known external factors over a series of matches (Rein and Memmert 2016). In sum, information from matches can help practitioners and scientists to better understand the game. Identifying talents and evaluation of training interventions based on match performance may have its limitations, but the challenge for football scientists is to resolve this with new sensor technology and advanced data analysis techniques. This is not limited to physical, technical, and tactical aspects of the game but could also include mental aspects. Recently, face recognition has been used to determine emotions in sports like tennis (Kovalchik and Reid 2018). Furthermore, driver fatigue was detected based on eye state (Lin et al. 2015). These developments will likely find their way into football. If we acknowledge that all information is already in the game, we may better understand football in all its complexity.