That was right in the corner, nobody’s saving that.”
“They had no right to save that.”
“You would have expected them to save that nine times out of ten.”
Some commentator lines are repeated so often, it can feel like they sit somewhere in a fan’s subconscious every time they see a shot on goal during a live game.
However, how accurate are such subjective cliches around the ability of a player to finish a high-quality chance, or a goalkeeper’s ability to stop them?
Following the successful launch of Expected Goals (xG) over a decade ago, Opta introduced another model, Expected Goals on Target (xGOT), also known as post-shot xG. xGOT leverages Opta’s historical database of shot data to identify the likelihood of a goal being scored based on the trajectory of a shot, its end location in the goalmouth and many other contextual factors.
Now, the OptaAI team have overseen a range of enhancements to xGOT, making the model even more powerful for commentators, studio analysts and dedicated performance analysts working for clubs.
This enhanced xGOT is now also split into separate models for men’s and women’s football, using data from 32 men’s competitions and 31 women’s competitions collected by Opta. It has been trained on over 300,000 shots on-target from recent historical competitions. There is also another separate model dedicated to penalty kicks, which has been trained on over 20,000 penalties.
So what does these xGOT models now take into account to deliver their real-time insights? Well, before we get into that, here is a very quick reminder of how xGOT is different from xG.
Expected Goals on Target Explained
Opta’s xG model measures the quality of a chance based on the location of the shot and dozens of on-field contextual factors taken from the moment a shot is taken. This means a clear sight of goal, from close range, is going to generate a much higher xG value than a speculative effort from way outside the box.
Whilst having a better understanding of what we would expect the average player to score, based on shot location, provides insight onto the quality of a chance, what we also know is that different players can execute the same chance very differently. A shot from this chance heading into the top corner is far more likely to result in a goal than a shot that is hit straight down the middle of the goal.
This is where xGOT comes in to measure that next level of context.
xGOT measures the likelihood of that on-target shot resulting in a goal, based on the combination of the underlying chance quality (xG) and information relating to the execution of the shot, including the shot end location within the goalmouth.
So if we think of xG being the quality of the chance based on the shot location, xGOT records the chance quality post-shot – it gives more credit to shots that end up in the corners than shots that go straight down the middle of the goal. As the name suggests, the model is used only for on-target shots given that, if you don’t get your shot on target, there’s a 0% chance that it will result in a goal.
Here is a good example of xGOT in action from last season’s Premier League.
Tottenham’s James Maddison stood over a direct free kick against Aston Villa, where the location of the free kick had an xG value of 0.12. Maddison’s execution of the free kick, around Villa’s wall and inside Emi Martinez’s near post recorded an xGOT of 0.73. Essentially, this was a difficult chance, measured by the xG value, executed to a very high quality, reflected in the much higher xGOT value.
You can watch Maddison’s execution of this free kick here (UK only).
So, how has xGOT been enhanced?
Without going too deep into the detailed intricacies of the model, xGOT’s enhancements can be summarised in four key points:
#1 Separate models for men’s and women’s football
Since the 2023 FIFA Women’s World Cup, all xG values from chances in women’s matches captured by Opta have been generated using a dedicated model trained on shots from historical women’s matches. More information on our women’s xG model is available here.
In the latest iteration of xGOT, we now also have separate models for men’s and women’s matches. The men’s model has been trained on data from 32 men’s competitions and the women’s model had been trained on 31 competitions.
Using xGOT values from last year’s top-flight women’s leagues in England, France, Germany, Italy and Spain, we can identify players whose shooting consistently improved the quality of the chances they had, often by striking the ball cleanly and placing their shots into the corners.
Using the Shooting Goals Added metric, which measures the total a player’s aggregated xGOT for the season exceeded their xG, we can see that from open play, TSG Hoffenheim striker Selina Cerci improved the quality of her pre-shot chances by 4.6 goals. This indicates that, from her open play chances, she was both getting her shots on target and hitting them in good locations where it was difficult to save for the goalkeeper.
#2 Deeper depth of goalkeeper data
As well as the underlying chance quality (xG) and trajectory of the shot, xGOT now incorporates detailed information about how the goalkeeper’s position influences the probability of a goal being scored. This is in relation to both their distance to the shot trajectory but also their position relative to the shot location and goalmouth at the time of the shot.
This information leads to intuitive outcomes in the xGOT value – for example a goalkeeper closer to the trajectory of the shot is more likely to make a save. Similarly, a goalkeeper standing in the left corner of the goal is less likely to save a shot heading into the bottom right, as illustrated by the goalmouth graphic below which displays a good underlying chance, 14 metres from goal with an initial xG of 0.26, where the xGOT value is 0.95.
#3 Shot execution insights
xGOT now accounts for extra qualifiers relating to execution of the shot itself. For example, if a shot was deflected or mishit, this will now impact the xGOT value attributed to the shot, as will any swerve applied to the shot.
#4 A dedicated penalty model
Penalty shots are different in their nature to shots from open play, given that the player taking the shot has an uncontested, clear path to goal and the goalkeeper has to have a foot on the goal line as the kick is taken.
As a result, a separate xGOT model has been developed trained exclusively on a historical dataset of penalty kicks.
Across last season’s big-five men’s European leagues, Bayern Munich striker Harry Kane generated the joint highest xGOT per shot (0.86) of players who attempted at least 5 penalties, scoring 9 out of 9 spot kicks for the Bundesliga champions.
So, who were other standout performers last season based on xGOT?
When we apply the enhanced model to every shot from last season’s English Premier League campaign, we can see that Everton’s Jordan Pickford prevented more goals with his saves than any other goalkeeper (6). As illustrated by the goalmouth graphic below, Pickford was expected to concede 50 goals (50 xGOT conceded) but only conceded 44 goals throughout the campaign.
Another metric which applies xGOT to illustrate the shot stopping ability of goalkeepers is Goals Prevented Rate. This rate standardises the xGOT metric based on the number of shots each goalkeeper faces, so that goalkeepers who faced a larger volume of shots can be measured more fairly alongside keepers with a stronger defence in front of them.
For example, in La Liga last season Mario Dmitrovic, who played in goal for relegated Leganés’, prevented more goals (4.75) than Real Madrid’s Thibaut Courtois (2.8). However when we normalise for their volume of shots faced, we can see that both goalkeepers prevented goals as well as each other, conceding 1.1 goals.
From an attacker’s perspective, we can obtain insights into how effective a player is at executing a shot with a relatively low underlying chance quality, based on the initial xG, getting away a shot on target with a much higher xGOT. Another good example is Omar Marmoush’s goal for Man City against Bournemouth last year, which had an xG of 0.02 but the xGOT was 0.63.
How can I find out more about xGOT and its applications?
Like all OptaAI metrics, these can be utilised via a range of feeds and product integrations, empowering you to share powerful insights into player performance in front of goal and a goalkeeper’s shot stopping prowess. More details on these solutions can be found via Stats Perform’s Product Finder.
If you are a data scientist working in-house at a club, our Pro Solutions team are also available to walk you through the model’s development and full list of features. They can be contacted on prosolutions@statsperform.com
For a full list of Opta model explainers, visit the Metric Explainer page on Opta Analyst.
