In this post I want to talk about player roles. For the present purposes, I prefer to think in terms of role rather than style because role conveys something rather more fluid than style. Of course, every player has his own style, which is derived from his individual mix of skills and preferences, and his own strengths and weaknesses. But he is also required to perform a job for the team, so the coach’s system and tactical plans will have a strong effect on what he actually does, how often he does it – and where on the pitch he does it. The role can be considered as style, inflected by team context.
I am also going to distinguish between roles and abilities. Roles are descriptive, abilities are evaluative. Roles define what is done, abilities define how well it is done. For example, shooting is an aspect of role-scoring is an aspect of ability.
One way to specify roles is simply to categorize players: a player can be a striker, a second striker, a left-sided attacking midfielder, and so on. Such categories have evolved over the years, and will doubtless continue to evolve with the game. We don’t speak of inside rights or outside lefts anymore, and the role of a sweeper is a relatively modern invention. Nevertheless, the basic roles are well-known and widely used and have proved their usefulness over the years. A more recent development has been to develop new categories of players based on statistical analysis, but a commonly agreed set of statistical categories has not yet emerged.
However, we can also define a player’s role by the particular pattern of activities he displays, without necessarily trying to categorize him. (Although categorizing the patterns always remains an option.)
The Structure of Player Activity
To identify the patterns of player activity I use a technique called Principal Components Analysis (PCA). PCA is an experimental method that finds clusters of variables in a dataset that are closely related to one another. In PCA, each cluster of related variables is associated with its own component or as I call it here, its own dimension.
I conducted a PCA on a large dataset of OPTA KPIs. The analysis revealed eight dimensions as shown in the gallery below. The items in the panels are OPTA KPIs, and the numbers beside each panel item are the so-called ‘factor loadings’. The factor loading of a KPI shows how strongly it is associated with its dimension. (It’s actually the regression coefficient of the measurement on the KPI.) One or two KPIs belong to more than one dimension – that’s OK. And a small number of KPIs have a negative relationship with their dimension; these are shown in red.
Scoring the Dimensions
Having identified the dimensions, the next step is to calculate player scores for each. For this, we need ‘factor score coefficients’ (not the loadings described above, but the regression coefficients of the KPIs on the dimensions). And then the KPIs must be scaled so that each one contributes to the score in an appropriate manner. If we used raw numbers, frequent actions like passing would dominate and infrequent actions like shooting would be under-represented in the score. And finally, we scale the dimension scores so that a score of zero on a dimension represents the average level of activity.
Now we have got dimension scores we can construct a role profile for an individual player or a group of players.
Or you could pick some cool casinos for the Brits and hammer out that maths and thinking:
- Non Gamstop Casinos for UK with No Deposit Bonus
- £5 Deposit Casinos not on Gamstop
- Casinos 300% Welcome Bonus
- Casinos 400% Welcome Bonus
Validating the Dimensions
All very well in theory, but does it work in practice? At the very least we would want to see different and credible profiles for different types of players. The chart below compares the profiles of Defenders, Midfielders and Forwards.
Here we see that:
- Defenders score
- well above average on Defensive Play
- below average on Creating Chances, Attacking the Goal and Moving Forward.
- Midfielders score
- close to average on most dimensions,
- slightly below average on Defensive Play
- slightly above average on Regaining Possession and Creating Chances.
- Forwards score
- above average on Attacking the Goal, Moving Forward and Creating Chances,
- below average on Defensive Play, Regaining Possession and Passing.
These profiles certainly seem credible and give us some confidence the dimensions are meaningful. But perhaps the most useful application of this profiling system is comparing individual players.
A tale of Two Defenders: Terry vs Zabaleta
The role profiles of Terry and Zabaleta are quite distinct. Terry is well above average on Passing and Defensive Play, and well below average on Moving Forward and Creating Chances. Zabaleta also rates highly on Defensive Play, while his well-timed interceptions and crunching tackles are reflected in his high scores on Regaining Possession and Hard Play.
By the way did you know that Terry plays in casinos? If he came to our site I would suggest he choose a casino from sites not on Gamstop.
A Tale of Three Forwards: Carroll, Suarez and Defoe
The profiles for these three forwards are pretty different from one another. Carroll, a classic center forward, is strong on attacking the Goal – but not much else. Defoe looks like a low-involvement player, with few notable characteristics, while Suarez looks like the complete attacking player, equally strong on Moving Forward, Creating Chances and Attacking the Goal.
A Tale of Three Midfielders: Fabregas, Lallana and Song
Cesc Fabregas’s role has evolved over the years, but his overall profile is that of a creative midfielder, with salient scores on Passing the ball and Creating Chances. Lallana is also above average on Creating Chances, but as different from Fabregas, his secondary strength is Moving Forward rather than Passing. Song’s main strengths are Passing and Regaining Possession, backed up by Hard Play; this clearly reflects his role as a defensive midfielder.
The Bottom Line
The eight activity dimensions I describe here seem quite a good way of differentiating players. They give more nuanced information than the traditional categories, without getting bogged down in the detail of dozens of KPIs. Of course, they only reflect the OPTA domain and give a complete picture. they should therefore be supplemented by tracking information. But because they provide simplified and comprehensive coverage of the OPTA domain, and are easy to understand, they have potential as a recruitment tool. They could also be used for example to track how a player evolves over time, or to see how his role changes when he transfers to another club or when a new manager arrives with a new system.
In future posts, I hope to look at some of these ideas, and I also want to look at how these dimensions can be modified to quantify player abilities and how to extend them with some tracking metrics. I sometimes think all this information can help in putting together the right bet, if you decide to do so you may want to consider these betting sites not on Gamstop.