Explaining the Concept of Expected Goals

Every year, we see broadcasters looking to introduce new and innovative ways of improving their coverage of football.

One ‘new’ initiative which has been introduced to the Premier League’s TV coverage this year in England, both on Sky and on the BBC, has been the Expected Goals total following each game. However the Expected Goals concept is anything but a new phenomenon, having been used extensively by both performance and recruitment staff at clubs for over five years now.

But given that Expected Goals has now gone into the mainstream, we felt it would be a good time to revisit the basic fundamentals of what makes up the Expected Goals total for a game, and explain to people who are new to the concept why they are important for measuring performance.

What Are Expected Goals?

"You will never get a better chance to win a match than that.

"He did not only have part of the goal to aim for, but he had the entire net - and he put it wide. Unbelievable."

This is a paraphrased version of a former Premier League manager's post-match reaction to one of his strikers missing a close-range opportunity to score a winning goal. The quote goes a long way to getting right to the heart of what Expected Goals is all about - some goal attempts, based on the location of where they occur, should have a better chance of being scored than others.

Our colleagues at OptaPro now possess an extensive database of shot information from matches played across different leagues and competitions – totalling over 300,000 historical shots – and what they have done with that is evaluated a wide array of different, but very specific characteristics across this entire database, to measure the quality of a single chance to determine the probability of it resulting in a goal.

These factors include basic fundamentals, such as the distance from goal, the angle towards the goal and whether the attempt was a shot, header or from a rebound, but it also takes into account more nuanced factors. This includes how the chance was created, based on the passage of play in the build-up or whether the chance came directly from a set-piece. They also makes adjustments for individual competitions too, based on the competition’s characteristics.

What we end up with is a total value for each individual shot in a game, so for example if a shot with a specific set of characteristics is likely to be scored one time in every 10, it will be worth 0.10 Expected Goals (xG).

In contrast, the average xG value for a penalty kick is 0.79, which reflects that the chance of scoring a goal from a penalty is far more likely. So in effect, what we are doing with this metric is defining what makes a good chance as opposed to an average one.

The values for each attempt at goal during a game are then added together and this brings us our final Expected Goals total for that team during the game.

A more reliable way of quantifying who was the better team

How many times have you heard someone say ‘we didn’t deserve to lose, we were the better team’, then cite the amount of possession and attempts on goal they had to back up their viewpoint?

Whilst this conclusion if often entirely correct, the one problem with citing attempts on goal is that it isn’t taking into account the quality of the chances created in the game.

One of the benefits of having an xG model is that there is a clear measurement in place to determine the quality of scoring opportunities and adds additional context to a player or team’s shots that goes beyond basic shot totals.

Equally, Expected Goals is typically a more consistent measure of performance than actual goals. Whereas goals are relatively rare events that come and go in stretches, according to OptaPro research a team or player’s xG output tends to fluctuate much less from match-to-match. Of course the goals that are actually scored are the ones that win points, but xG gives us more context for evaluating a team’s performance.

Understanding a player’s underlying performance

By comparing the goals that a player scored during a full season with the chances available to him through xG, we can better understand what is driving that player’s performances. If the xG numbers are significantly below the player’s goal output it may be a sign of an unsustainable run or at least merits further study to understand why this over-performance - compared to the average at least - is happening.

We can also tell a lot about a player’s shot selection through xG. By analysing a player’s average xG per shot we can understand whether he is taking high quality shots or lots of shots from areas from which he is unlikely to score.

Two good examples of this can be found in OptaPro’s Premier League totals from last year. As we can see from the table below, Romelu Lukaku massively outperformed his season xG tally, which shows that he converted more opportunities which had a lower probability of resulting in a goal.

In addition, if we look at Jamie Vardy, we can see that he only had 53 non-penalty shots in 35 League matches, but his average xG output per shot was 0.19, higher than any other player listed. This suggests his attempts at goal were either of a higher quality, or he was refusing to shoot when there was a lower probability of him scoring.

Goal output can mask a team’s real performance

Using historical evidence, we can see that Expected Goals projections can help us unmask the real performance of a team, who may be over-performing or under-performing based on the actual number of goals they are scoring.

A good example of this is Arsenal’s start to the 2015-16 Premier League season. In their first six games they only scored five goals, which was an average of 0.83 goals per game. This is a worryingly low number for a team that was expecting to challenge for the title.

However, during this period Arsenal had over 12 Expected Goals, an average of 2.11 Expected Goals per game.

By the end of the season, Arsenal had scored 65 goals at an average of 1.71 goals per game, which sat closely alongside their Expected Goals output from the start of the season.

If we had only analysed the goals that Arsenal had scored in their first few matches of the season, we would have never expected them to finish the season with so many goals. However, by looking at their xG we can obtain a much clearer insight into how Arsenal were actually playing during those early weeks.

Other matrixes which build on Expected Goals

Alongside Expected Goals, OptaPro have also built a number of other reports which build on the Expected Goals concept.

One of these calculates a revised total of Expected Goals based on the quality of each shot once it leaves a player's foot or head, whilst another, called 'Keeping Goals Prevented', measures how many goal-bound opportunities are either beating or being saved by goalkeepers during the season, based on their difficulty. This information was reviewed at length by Sam Green on the OptaPro Blog a few years ago, which can be reviewed here.

In the coming months we will be publishing more Blogs, looking at other areas where data collected from a match is being used to help us understand performance, to provide context to support the subjective assessment of players within a club’s scouting operation.

We hope you enjoy reading them.

Some of the examples used in this Blog has been taken from an earlier Blog of the same title, published by Sam Gregory. To read this version, click here.

by Andy Cooper & Sam Gregory PR Manager/Data Scientist

Published 27 November 2017