Expected Goals: A Battle Between Variability and Reliability
There are infinite possibilities that can follow any touch of the ball in a football game, from the more likely ones like goals, assists, or loss of possession or to the extremely unlikely such as the ball shooting out into space, hitting an alien ship, angering the captain of said ship, leading to the alien blowing up planet earth, destroying life as we know it. There are also uncountable variables impacting a player’s shot.
The player’s mindset on that particular day, how loud the fans are cheering, the quality of the pitch, how tight the player tied their shoelaces, how important the match is to the team, how many minutes are left on the clock, and so on and so forth. If the sun shines too brightly in Martin Ødegaard’s eyes, he might not aim a pass as precisely as he usually would.
One of the things that make football so enticing is the lack of predictability. There are outcomes that you can expect from football but you can never take anything for granted. That being said, as the sport develops further, so have the tools we use to understand the game.
Statistics are increasingly becoming important not only to analysts and pundits but to the average fan as well, with the internet making betting and games such as fantasy leagues incredibly popular.
One of these tools is a statistical measure called expected goals (xG). This metric was devised by Sam Green in 2012. The measure does not aim to predict the number of goals scored by a team or player. It analyses, based on datasets of previous shots taken from similar angles and locations, the likelihood that a shot will result in a goal. Similar calculations can be made for how likely a pass is to result in an assist (known as expected assists or xA).
How is xG calculated?
Assume a player is standing in their own half when they attempt to score. Now assume the player is standing in their opponent’s penalty box. Intuitively, it’s easy to understand why the second situation is more likely to result in a goal. Other important factors to keep in mind are what part of the player’s body are they using to take the shot (header, right foot, left foot, etc.), what angle they are shooting from, etc.
Different analysts have different regression models for how they calculate xG, xA, and other related measures. The process of model construction is the key to the reliability of the data. This comes from reducing the margin for error and carefully choosing the variables plugged into the model.
Model Construction
The accuracy of a model can be understood from how well actual data matches the expected data calculated from a large sample size. In simpler terms, as you increase the number of shots included in the sample size, you increase the accuracy of the model.
Opta doesn’t take into account the positions of defenders while calculating xG and xA, while other providers like Stratabet may do so. Each shot that a player takes is affected differently by different variables. A model calculating xG will not only rely on a large sample size to help reduce the margin of error but will also rely on only taking into account relevant variables.
Adding unnecessary variables to the model will only complicate it and make the results less meaningful as well as harder to calculate.
As it stands, xG, when put in context and used in tandem with other statistics can provide astute insights.
How is xG used?
When used in conjunction with the actual number of goals a player or team scored, xG can provide convincing evidence to analyse whether the team/player is performing as expected, better than expected or underperforming.
For example, if a team’s xG in a particular season was 20 goals and they actually scored 25 goals, that does not mean that the xG was inaccurate. Instead, it means that they managed to convert chances that were less likely to be converted, indicating that the team is performing well. Similar principles can be applied to any individual player and to other metrics such as xA.
Hypothetical Datasets
As the professional game evolves, so do the rules governing it. Making changes to models calculating xG could possibly help understand the impact these rules have. For example, it might be telling to analyse the impact of VAR decisions on the number of expected goals in a season.
Looking a little further into the past, goal-line technology is also a change that impacted the definition of what constitutes a goal.
Even as our understanding of the sport deepens, aided by technological innovations, football will always remain largely unpredictable, a trait that makes football so compelling. It can be argued that footballing technique is an art in itself, and tools like xG can help us quantify exactly how much skill was required to score what the naked eye sees as almost impossible goals.