The author has been fortunate enough to practice the data analysis needs of practitioners at different levels in the football industry in the past period of time, and to discuss and communicate with the analysis departments and data suppliers of clubs at different levels in many European countries. This series of articles will share some superficial experiences and insights to explore the best practices for data analysis to support decision-making in the field of football or competitive sports.
No matter which industry it serves, excellent data analysis needs to set a clear purpose, put forward reasonable assumptions, conduct analysis with clear logic based on sufficient and credible information, and finally obtain actionable results, thereby bringing added value to stakeholders.
In this workflow, the most difficult and important part is how to ask the right questions in the right form, at the right time, and clarify the purpose of the analysis. In actual work, 80% or more of the time is spent on collecting and organizing information. After completing these two steps, the analysis itself is a relatively simple part.
The coaching staff, team boards, agents and media that I have personally served can exactly represent the minimum requirements for data type, granularity, accuracy and cost among the four different levels of requirements when issuing an analysis report with acceptable quality.
Coaching team
The coaching team needs GPS/GNSS tracking data, inertial acceleration (deceleration) data, external (such as running) and internal (such as heartbeat) physical-performance data data of players in games and training. They are the group with the highest requirements for data granularity and accuracy. Coaching staffs almost exclusively need micro-level tracking data to uncover hard-to-observe details and provide decision support when it comes to strength and conditioning training.
However, for the coaching staff who are not responsible for transfers, they may be the group that needs data the least. First of all, the cost of collecting and processing GPS data is high, which is difficult for teams with limited budgets to afford. Secondly, game tracking data requires a certain understanding of football principles and considerable data processing capabilities in order to draw meaningful insights, which is sometimes not as straightforward as watching videos directly.
This type of data is mostly collected through the Pod on the back of the vest worn by the players. Generally speaking, except for the acquisition frequency of inertial acceleration which is 100Hz, the other several data acquisition frequencies are 10Hz. After nearly 20 years of product iteration and strict verification standards, some of the industry's top GPS/GNSS equipment can collect position information with an accuracy of less than 10 centimeters, and the error of most indicators is within 5%. However, there are some indicators that may have an error of 5%-15% due to unknown technical defects in the equipment, and the coaching staff needs to be informed. After all, in training, a 5%-15% difference in intensity is likely to lead to injuries.
event data statistics (such as the number of passes, the number of confrontations, the number of passes, etc.) are not mentioned here, because the event data statistics can only reflect superficial phenomena and have limited application value. When misused, it may even mislead the coaching staff and lack preparation for strategies that are not commonly used by opponents. In addition, at the macro level, event data has insufficient added value compared to video analysis, and is mostly for verification and reporting purposes.
sporting director/Board of Directors
In recent years, with the explosion of computer vision technology, the combination of video tracking and event data with position coordinates has gradually become popular among top league teams. Currently, all teams of La Liga and Premier League have applications. In the latest version of Skillcorner data, compared with GPS tracking, the accuracy of capturing player position information in non-intensive positions has been reduced to 0.4%, which is a significant improvement from the previous 3%.
Although video tracking is limited by the broadcast perspective, about 10% to 20% of events are usually lost in each game. Its limitations discourage coaching staff who are qualified to use GPS/GNSS tracking. However, the remaining 80% to 90% of events that combine on-ball actions and off-ball movements are enough for sporting director to understand player physical fitness, speed and specific abilities through data at a relatively low cost, and establish a scouting database for further screening.
After having video tracking and events with location information, there are different schools and methodologies for different clubs and institutions on how to model player abilities. But my personal principle is that after having this kind of data, we can translate football into the data as much as possible, and construct indicators that can explain and prove from the principles of football that the players can meet the specific tactical requirements of the game model of the team coach (such as the center retreating to his own half to receive the ball and passing the ball to the designated area, making a short pass from the frontcourt to the right and then passing the ball to the left, etc.), which can be regarded as a qualified and responsible indicator for sporting director and the board of directors' decision-making.
Broker
The now popular event data is a powerful assistant for brokers. Although event data has limitations, and some suppliers' key indicators (such as wyscout's dribble) have vague definition problems, it is cheap and easy to use for agents to track and classify player performance. Agents who have the conditions can also purchase a more clearly defined and re-labeled event data through a data processor and produce a report to make the players they recommend more credible.
For agents, it is important to be able to quickly recommend a suitable list of players based on the club's capabilities and budget needs for further analysis by the club.
However, if the club is unfortunately in a situation where it has to find someone frantically on the transfer deadline, neither sporting director nor the agent will have time for even half an hour of detailed analysis. At this time, the simple and flexible event data every 90 minutes, coupled with the simple qualitative analysis (or delivery) brought by the agent's understanding of his players, is enough to decide the player's fate in a few dozen seconds on the phone. Grasping at straws is a symptom of club management problems and a desperate situation that clubs need to avoid, and agents can often take advantage of it.
Media
For the media, they do not care about the exact definition, type, granularity, precision and other details of the data. They only care about how to tell a good story, whether the visualization is beautiful, and the lower the data cost, the better.
The media and the first three levels no longer belong to the same world. The player who has taken the most wickets is an interesting story that entertains the public, but it is almost meaningless to coaches, teams and agents; player ratings can trigger a lot of discussion after the game, but if anyone really tries to use a single score or a single indicator to evaluate players, it does not rule out the case of good luck, but most of them will only suffer the consequences and make others laugh.
When comparing event data indicators, the media often lacks an examination of the context and accurate definition of data, which is the hardest hit area when it comes to fighting against related event data. First of all, Opta’s definition of confrontation against the 50-5 kick-off is very loose, which means that the intensity of different confrontation events may be completely different. In addition, the two sides of the confrontation, the area where the confrontation occurs, the time when the confrontation occurs, and the form of the confrontation cannot be measured by a single number of confrontations. Ignoring context, such behavior can only be called comparison, not analysis.
Sadly, due to the extensive influence of the media, its misuse and even abuse of data can mislead the public at least, and at worst affect the understanding of some decision-makers who have limited knowledge of football. In such a media environment, how to explain to such customers that a forward who cannot score goals and has many offensive event data indicators hanging at zero is actually the person the team needs most. I am afraid it will be an unsolvable problem until the customer pays tuition and goes bankrupt.
Limitations of data quality in practice
In actual operations, due to various reasons such as budget, technical limitations or data collection environment, it often happens that data quality or quantity cannot meet the minimum standards. However, these problems are not unique to data analytics but are prevalent in any field that relies on data-driven decision-making.
A significant challenge is data integrity. Data collection in football matches is often limited by the coverage and accuracy of technical equipment. For example, video tracking technology may cause data loss or inaccuracy for some players at specific moments due to camera angle and field of view limitations.
Another key issue is data consistency. The standards and methods of data collection may differ in different competitions, different stadiums, and even different countries and leagues. This inconsistency affects the comparison and integration of data. For example, different vendors may have different definitions and criteria for certain events (such as pass success rate, confrontation, Dribble, xG).
This is not so much a question of data quality as it is a question of fostering consensus. I hope that as many people as possible can understand that with reasonable analysis methods, although the data is not 100% correct, it is reliable; even if the data is not perfect, it still has reference value.
Therefore, as a data analyst, you must maintain sensitivity to data quality, be aware of the limitations of the data you handle, and understand the causes of these limitations with the most critical eye. You must be able to clarify the limitations when reporting, and issue verification reports or videos when necessary, and then continuously optimize analysis methods and processes to cope with various challenges in reality and provide the most accurate and reliable analysis results possible. Only by working with this attitude can trust be built in collaboration.