Whenever you watch any sport you quickly hear the commentators talking about concepts such as IQ and decision making in regards to player.
But despite the boom of event and tracking data we don’t really have great metrics to capture or understand good decision-making. This isn’t just because our data isn’t good enough, it’s because the concept of an optimal decision is so context specific in free flowing sports that it quickly becomes a very difficult disentanglement problem.
So let’s try to define decision making is first:
An optimal decision is the one out of your possible options that best helps your team achieve their goals
Possible options is a function of two things:
What do you perceive as an option?
Simply put, a teammate you can’t see or don’t know is there is not an option to pass to
What do you think your team can execute?
A very difficult pass with your off-foot that you know you have 0% chance of executing and therefore pass up could be a good decision
Obviously goals usually differ in sports depending on time and score but for simplicity let’s say the goal is to maximize goal difference.
As a result of all of the above, decisions are really hard to measure due to how context-sensitive they are. An example that may further clarify this concept:
- Busquets missiles a very fast pass to Messi – he knows Messi can control it. If this pass were to someone else, it would be a worse decision
Some output statistics hint at decision-making – things like the rate at which you turn good positions into shots or shot-assists, passing success rate, key passes to turnover ratio, but nothing not tremendously flawed and context specific.
So in the absence of good tracking data, where else can we turn to attempt to quantify decision making? Below are a couple of ideas I had when thinking about this problem.
Freeze-frame shot taking
Using StatsBomb’s FF data to quantify what was a “bad shot” relative to options.
Perhaps build a pass difficulty assumption and take a look at scenario’s where a pass and first-time shot from teammate would’ve been higher expected value for the team.
Next action success rate
If it was a good decision to pass the ball, it follows that what happened was more likely to be a successful event even when adjusting for it’s difficulty – and the quality of the subsequent decision.
Receiver’s pass % above expected (or above personal baseline even).
Turnovers that lead to counter-attacks
Quantifying bad turnovers as one’s that lead to counter-attacks or quick ball progression against your team.
I might take a crack at trying to code some of them up next weekend!
Fikha's Blog 