Nowadays players are not correctly reflected in statistical models since it looks as if they could pass the ball in any direction no matter of their body orientation, as though they were omnipresent. If we introduce orientation, the models will be more realistic and will give added value to the reports drawn up for the coaching staff.
We shouldn’t forget that most coaches spend a lot of time training their players in orientation, so that they know at all times how to face the different attacking and defending situations they might find themselves on the field.
—What stage are we at in developing these new variables?
—Great changes are taking place in the analysis of data and the creation of predictive models. The specialization in football nowadays is making technologies progress rapidly, allowing for the development of new statistical models. There have been reliable tracking devices for positional variables for around four or five years, and now we’re breaking into the field of computer vision.
This opens up a new era in which we can combine the different data points and create more robust prediction models. If we had talked about this five years ago, no-one would have imagined we’d already be doing it today.
—Fascinating! What uses and applications can the analysis of orientation give us?
—There are some really interesting applications, not just for the coach but for different members of the coaching staff. For example, during the training sessions we can see what direction each player is running in (forwards, backwards or sideways), and according to the type of movement, this affects certain muscles or joints. Thanks to orientation we can personalize each player’s workload more effectively, helping to create more realistic programs for the player and his position.
Another practical example is that it helps to improve the pitch control statistical model. Right now, position and speed variables are taken into consideration for this model. If a player is running backwards, the model detects that he is generating a space behind him because he’s moving backwards. In fact this isn’t so, the space he’s controlling is in front of him. Thus, if we incorporate this variable, we improve the model, making it more realistic with respect to what is happening on the field.
—And for the players?
—There are events which become more significant with orientation, like when you receive the ball. When you’re in possession of the ball you have two options—you can direct the play or pass it on. By knowing the orientation, you can determine which action is most appropriate between the attacker and the defense. If no-one is within your field of vision, in theory it’s better to move forward with the ball. Likewise, players can also lose the ball because they’re not as well oriented. We can improve these situations in the future by using customized reports that analyze these different player circumstances.
We can also incorporate orientation in the pass probability model. The player who makes a pass can’t pass in all directions, and the player who receives it gets a pass to his foot if he’s facing the passer, and into the gap if his orientation is different. Right now, only passes to the foot are considered. With this new variable, we can see the likelihood of making different kinds of passes which are not considered today.
—How is this data represented?
—It’s important to summarize all this information visually and intuitively so that everyone involved can understand it. It’s very easy to get lost with so much data, but the key is being able to represent it simply enough to explain the main concept being conveyed. As Javier Fernández and Raúl Peláez remarked in Barça Sports Analytics last year, this part of the process is vital, if not all the work carried out is useless because it can’t be understood. We’ll see in the future how UI/UX designers are also incorporated into data analysis teams to help shape them.