Invisible Action: 5 Metrics to Capture Off Ball Value
The dominant paradigm of football data analysis is events data. Events are defined in relation to the ball — did the player pass the ball? did the player shoot? did the player tackle the ball?
In many ways defining our analysis by reference to the ball makes a lot of sense. The aim of the game is to get the ball into the opponents net and to keep it away from your own. In another sense, however, on ball events data misses 99% of what happens on the pitch. There is one ball and 22 players. At any given moment, 21 players are acting “off ball”. For any fundamental principle of the game, there’s a Cruyff quote. Here’s what he said on off ball action:
“When you play a match, it is statistically proven that players actually have the ball 3 minutes on average. So, the most important thing is: what do you do during those 87 minutes when you do not have the ball. That is what determines whether you’re a good player or not.”— Johan Cruyff
Tracking data offers access to data where every player is on the pitch at any given moment. Without the correct methodology, using this off ball data could lead to granular analysis of each individual movement on the pitch. We need to synthesise both tracking data and events data into metrics that can be summarised, quantified, and statistically analysed.
The Athletic recently published an article detailing how tracking data is being used to enrich data analysis of the game, in conjunction with Stats Perform. The article highlights the immense potential of tracking data, and many of the metrics discussed echo my own thoughts on how to make the most of tracking data. Serving as a useful summary of the state of tracking data in the industry, this article is referenced throughout. Stats Perform’s research, however, still largely centres analysis on the ball. The purpose of this article is, in contrast, to look specifically at what players are doing without the ball.
With this in mind, I propose 5 possible “off ball” metrics that could be used to measure elements of off ball play. Each metric can be distilled into the sort of summary data you’d find on FBref, allowing for comparative analysis and profiling.
Gravitational Pull
Some players attract defenders. By dragging defenders out of position, players with high Gravitational Pull open up space. The proposed Gravitational Pull metric credits this sort of off ball movement by recognising how these runs can facilitate progression and generate opportunities.
In the situation above, the pass is only made possible by the presence of the player on the right wing. Without that player attracting the opposition away from the pass recipient, the pass cannot be made. This player deserves credit for the move despite not getting on the ball.
We can use tracking data to spot occasions where opponents move towards a player, opening up a passing channel.
The value of a player’s gravitational pulls can be defined in relation to the on ball action the pull enables. We can measure this simply using OBV or xT, two metrics which provide a value for each event that happens on the pitch. Such a metric will capture the progression the pass allows for, and will credit the player that made the off ball movement for the opportunity provided.
A dataset summarising the Gravitational Pull metric might look something like this:
Space Creation m²
Whilst the Gravitational Pull metric aims to capture a player’s ability to attract opponents, the Space Creation metric captures a player’s ability to repel opponents. Space creation might look like shaking off a marker or running into a pocket of space in an advanced position. Space is one of the most valuable resources on the pitch, so we should credit players that create space, whether or not that space is directly utilised in the particular passage of play.
Stats Perform look at the space an opponent is in when receiving a pass (how close the surrounding opponents are) to determine the degree of pressure the pass recipient is under. The Space Creation metric differs in that we are interested in all off ball space creation, regardless of whether that player is passed the ball.
In the diagram below, the blue player makes an off ball run up the pitch. This run improves the amount of space that the player can receive the ball into (denoted by the grey space).
When quantifying this sort of action, we can choose to count the number of occasions a player moves into space. We can also choose to weight these space creating actions by the area of the pitch they occurred in (space creation in the final third is more valuable than in less advanced areas of the pitch). We could also value these actions based on the area of space created in m². This allows us to quantify how much space was created.
A dataset summarising the Space Creation metric might look something like this:
Shot Opportunities
It’s City vs Bournemouth, Foden is through on goal, but instead of passing to an unmarked, centrally positioned Haaland, he chooses to shoot from wide. The shot doesn’t go in. In the end, this moment had little impact on the result, City cruised to victory. Nonetheless, Foden’s decision making came under heat. This proposed metric is about crediting Haaland for his good positioning on goal, despite not being fed the ball.
Stats Perform use tracking data to create pass prediction models that assess a player’s on ball decision making.
“What passing option does a player choose? What are the risks and the rewards of that decision?”
The major difference between this sort of model and the Shot Opportunity metric is about who the metric intends to credit. The pass model looks at a player’s passing options to assess how the player on the ball performed. The Shot Opportunity metric would use the data to assess a player’s ability to get into good positions, regardless of whether that position is chosen by the passer. The player being assessed is the ‘potential pass recipient’, not the passer. The method described in the article would be used to assess Foden, I want to assess Haaland, who was in a great position but didn’t get to touch the ball.
The player that is ready to receive the ball in front of goal can be credited with the xG the shot they could have taken would have been worth. For Haaland, that would have been an exceptionally high xG opportunity. This sort of metric would be particularly useful in helping identify strikers that do all the right things but might not have the teammates that can back them up. That striker might be excellent at finding shot opportunities, but is not getting the service they require to rack up an impressive goal tally. Such a striker would also have an unimpressive xG accumulation as they are not given the ball to shoot with.
I imagined that this metric could be used specifically for players that found themselves in situations with a significant goal scoring opportunity. Research at the Barça Innovation Hub explores the prospect of using tracking data to assess opportunity creation across the pitch. The off ball positioning would be credited by its “off ball expected threat”. This excites me and I recommend anyone curious to read their article addressing the topic.
For now, here is how Shot Opportunities might look as a dataset:
Players Bypassed/Passing Channels Blocked
Stats Perform explore how opposition position is valuable context when ascertaining how important a pass was. In their article they make reference to two Thiago passes that are the same length. One of the passes cuts several opposing players out of the game, whilst the other doesn’t. Looking at by standing opponents can reveal a lot about the value of a pass.
“The second output is a greater understanding of ‘line-breaking passes’ — i.e. how many players does the ball eliminate with a single pass?”
On the face of it, this sounds like an augmentation of a typical on ball metric: we are using off ball positioning to help judge an on ball action (a pass). Most metrics have two sides to them: goals scored is one side of goals conceded. Likewise, there are two ways of looking at this data. Is the side in possession good at progressing the ball past opponents? Is the side out of possession easily bypassed? For each bypass, a player has been bypassed.
I feel that assessing midfielders through this lens could be particularly interesting. Perhaps some midfielders are great at line breaking passes, but don’t hold the line themselves. For others it would be vice versa. The really special players, though, would rate highly both as bypassing and as impassable.
The dataset might look like this:
Defensive Line Position
Stats Perform use off ball data to analyse team shape, with technology that “can detect the most common shapes adopted during a match” and quantifies these as a percentage. They focus on looking at how formation changes in and out of possession.
Using a similar methodology, I am interested in how we might learn more about a side’s defensive line through looking at their positioning in possession. Currently very trendy in football parlance is the idea of the high line, where a side’s defence are positioned high up the pitch. This improves their capacity for involvement in attack, but makes the side vulnerable to being caught on the break.
Tracking data can help us quantify a side’s defensive line without relying on the eye test. Much like how Stats Perform quantify formations, we want to ascertain the percentage of time defensive players spend in zones of the pitch during possession. The zones can be defined by whatever granularity deemed fit. For the sake of this example, I have opted to split the pitch into 5 zones, as shown below.
The dataset might look something like this:
All sorts of insights are possible from such data. We might find that the right side is positioned higher than the left, creating asymmetry and potential vulnerability.
Concluding Thoughts
I firmly believe that the use of tracking data to quantify off ball value is the future of football analytics. To implement metrics like those I have suggested above seems difficult, but more so than technical expertise, I believe that what this area of innovation needs is minds at work looking for ways to make the most of these rich datasets.
