• Attempting to quantify the value of a wide range of player actions using only event data has gained traction recently, with VAEP, xT, g+, and PV some of the most well known models.

• All of these models return a single point estimate of “value” which seems sub-optimal given we know so much vital information hasn’t been considered by the model.

• We also have no way of knowing how much trust we can place on any conclusions we are drawing without making an attempt to quantify the uncertainty inherent in our models.

• We can define two types of uncertainty, esoteric and aleatoric, and explore some techniques to add this information to our models. Established and new metrics can both advance from a solid theoretical grounding to something that can actually be used as part of a real-life decision making process.

Go straight to the interactive applet

The “curse of dimensionality” is a somewhat dramatic phrase you might not necessarily associate with the less than high-octane world of modelling 22 people kicking an inflatable piece of leather around for an hour and a half. It is however something that the entire field of data science has to contend with.

Whilst the public analytics movement has barely scratched the surface in terms of realising the potential of event data, the use of tracking data in tandem with events undoubtedly offers additional context to the analysis in many situations. This is the first in a series of articles which will take a practical look at how tracking data can be applied to pre- or post-game analysis, with the aim of gaining actionable insight on our own team or an opponent.

• A step-by-step guide demonstrating how to generate a finishing ability metric from scratch including confidence interval

• An xG model built using only tracking data freeze frames is demonstrated, producing encouraging results compared to a current state-of-art model

• Distributing a given xG total into few, high quality shots will not maximise points return for some teams in certain match situations – in fact the opposite is true