It was possibly the most underwhelming manner in which it could have been achieved, but after a thirty year wait for a top-flight league title, it was unlikely that any Liverpool fans cared. Premier League Champions 2019/20.
This most recent success caps a remarkable run of form for the English football team. In two years, they’ve picked up a Champions League title after making two consecutive finals along with a Club World Cup title picked up earlier this season.
What catapulted Liverpool FC from a solid top six side in the Premier League to one of the best teams in world football? Well, there are a number of factors: their manager, Jürgen Klopp, for one; their owners, Fenway Sports Group, for another.
But there is another aspect of their approach to the game of football that has often been vaunted as the secret to their meteoric rise: the use of data.
Everyone has at least seen the film. Angry Brad Pitt and his unlikely Oakland A’s narrowly missing out on a World Series.
The premise of Moneyball is simple. Try and find value in sports players by finding outliers whose qualities are not recognised by people within the sport. How do you do that? Assign metrics to actions. Trawl through the data. Pick up on people who aren’t picked up on through the naked eye.
In practice, the process is much more complex. Liverpool’s data analysis department is headed up by Ian Graham, holder of a Cambridge PhD in polymer physics… whatever that is. His department is made up of scientists and mathematicians any of whom could have gone on to work in the best universities around the world.
The fundamental question that this department faces, however obliquely, is: how do you make data meaningful? How do you take information from gameplay scenarios and use it to make suggestions about how to improve your play? How do you use data to assess how well a player at another team will fit in with your team.
For the most part, the solutions to these questions are arrived at by building models into which the data gathered from games can be plugged which can then assign value to events. From here, you can then compare players to assess their usefulness across a number of different metrics.
This is all very well and good. But what does it look like in practice. One of the best ways of illustrating this is to talk about “expected goals” (often shortened to xG in football analytics).
The idea of expected goals translates well into the Rocket League scenario. Imagine two situations. In the first, you’re stuck in the corner with a shot at goal. The angle is narrow but you can make it. There is no one guarding the goal. In the second, you’re right in front of goal. But there is someone goalkeeping.
Which of these situations is more likely to end in a goal being scored? In the first one, the angle is against your favour but the lack of opponents is in your favour. In the second, the opposite is the case.
This is the question that expected goals looks to measure with a degree of objectivity. By taking the data from thousands of examples of football games, an xG model will assign a value between 0 (impossible to score) and 1 (impossible to miss) so as to assess the likelihood of a chance being scored.
Depending on the model, variables such as shot location, position of opponents, height of the ball, part of the body the shot is hit with are all taken into account.
And the result is a number which translates into a percentage. In our first scenario, you might assign a figure of 0.1xG. That means 10% of the time you might expect a goal to be scored.
For the second, you might see a higher figure, 0.5 xG for example. You would expect that chance to be scored 50% of the time.
It All Adds Up
Of course, in individual instances, expected goal values aren’t that helpful. If you’re a regular player of Rocket League, you should have a fairly good eye for judging chance quality. Where it becomes helpful is in assessing chances over a long period of time.
Let’s imagine we looked at all of Fairy Peak and Kaydop’s chances from last season. This would be a large number of chances for both and it would be impossible to assess these numbers individually.
If we had an xG model for Rocket League, we could generate the percentages for all of these chances and we could work out the quality of the chances across that time.
Imagine that, hypothetically, Fairy Peak’s accumulated xG total is 65.3 (that’s the sum of the xG for all the chances he created in the season). That would mean that, given the quality of the chances he created, you would expect him to score 65.3 goals.
If he actually has only 55 goals, then you might worry that he’s under-performing in front of goal.
But then imagine we look at the average chance quality of these shots and we find that the average of his chances is 0.1 xG. This might indicate that he’s taking more low percentage chances than he should.
Let’s assume that Kaydop has a total xG sum of 45.7 but he’s scored 58 goals. Of course, it’s good that he’s over-performing his numbers. But will that mean at some point he might regress and start scoring fewer goals as the numbers suggest he should?
What if his average shot quality is 0.6 xG? Does that mean he’s getting into good positions and scoring well as a result? Or does it suggest he needs to increase his shot volume and might score more as a result?
All these are potential ideas thrown up by the use of data in sport. This is just one way of assessing a player in football. There are countless other metrics that can have similar insights.
Where’s the Data?
Liverpool FC are unusual in the extent to which they use data.
Why might this be the case? Well, because it’s actually an expensive and time-consuming process to gather data.
How do you get from a game of football to an Excel spreadsheet? You gather the data. You have people watching games back on video, logging events and building a numerical picture of what happened. This is an incredibly laborious task.
One way to speed this process up might be to make the process automatic: using a computer process to analyse a video feed and generate a data picture of the game.
In football, one of the big developments is tracking data: tracking the players and the ball to be able to analyse positioning and patterns of play to work out if there are marginal gains to be found in the structural arrangement of players around the ball.
It’s in this sense that a lot of esports are at a material advantage to their less electronic brothers and sisters. For this reason: a computer game generates data immediately.
When you finish a Rocket League game, you can download a replay of the whole game with every single action in the game rendered in numbers. That is to say, in simply playing Rocket League, you are generating data which can then be analysed after the game.
When you take a shot in-game, the location of the ball, the shooter, the defending players and their speeds are all known. If you wanted to, you could scrape this data after the game and use it to build a picture of what was happening.
Over the course of the season, you could generate a whole database of information which could then be farmed for insight into how a player was performing through that time-frame.
Maybe we would find that Scrub Killa was arriving at aerials at a slightly slower speed than the average. That might encourage him to work on his approach in aerial play.
Maybe we would find that GarrettG was hitting aerials slightly lower than the average. That might get him working on arriving at the ball quicker in the air.
In all cases, you would be using the data produced by the game to improve your own gameplay and find out more about the meta of the game.
Have We Missed the Ball?
When it comes to Rocket League, then, it feels as though we are behind the curve when it comes to data analysis.
Yes, there are sites like Ballchasing.com which are starting to use data in creative ways. But the sport is already lagging behind other sports in this regard – a fact made event more damning by the ease with which data is available for Rocket League players.
As the sport continues to professionalise, the use of data analysis techniques will only become more and more important. Once a plateau is reached through traditional coaching approaches, it is inevitable that data-driven approaches will increase the marginal gains and carry the sport to the next level.
Let’s start a conversation about data in Rocket League analysis. This way, we will ensure that the game continues on its upward trajectory.