ANALYSIS OF GOALKEEPER PERFORMANCE IN THE ENGLISH PREMIER LEAGUE.
An analysis of player performance as at Gameweek 16 by Jeffrey O.
BACKGROUND
The English Premier League is made up of twenty football clubs that play each other twice in a season, once at home and once away. Each club plays 38 games, yielding 760 games in a season. There are a maximum of 114 points to be won. The point system for each game is as follows: 3 points for a win, 1 point for a draw, and 0 points are awarded if a team loses a game.
Performance is everything in this league, and, according to The Mirror UK, with over £160m awarded to the winner of this extremely competitive league, one can see why clubs in this league do everything to get ahead of their opponents in terms of recruitment, coaching, and performance, in order to win as many games as they can to meet their targets of winning the league, competing for the league, qualifying for continental football, staying midtable, or avoiding relegation.
There is a reason why coaches like Antonio Conte, Eddie Howe, Pep Guardiola, Jesse Marsch, Mikel Arteta, Patrick Viera, Jürgen Klopp, and Frank Lampard get FURIOUS for every misplaced pass, every big chance missed, every duel lost, and every little mistake. And why? You see coaches famously try to take digs at their opponents, play mind games to influence their own chances of winning by getting into the heads of their opponents. The stakes are so high and every little thing matters, from match preparation, fitness, tactics, subtle patterns to exploit in your opponents game, etc.
We’ve all at some point loved a Mourinho dig at Pep, or a Conte altercation with Tuchel, or a brawl in a derby game after a player made a very dangerous tackle, or a cheeky snub by a coach after a defeat, or a big win against a competitive rival.
Since the start of the season, five clubs have already replaced their managers as of gameweek 16, and given the expensive cost involved in terminating a manager’s contract when results are not delivered, clubs still do it anyway. It should tell you about the sheer desire of every club not to be left behind, not to lose too many games, not to finish lower than expected, or, in a simple phrase, to perform.
Using data from FBREF, this article will examine goalkeepers based on key performance metrics.
For this article, we will use data from FBREF.
Since 8 out of the 20 clubs in the league fielded more than one goalkeeper during the period we want to analyze, we will use the goalkeeper with the most appearances to represent these clubs, and where there is a tie, we will use the goalkeeper with better base statistics.
First off, we will start with the performances of goalkeepers by club. We will start off by checking the correlation between a club’s actual goals conceded and the expected number of goals they would have conceded. I will introduce a key performance metric known as "xG."
Excuse me, did I just hear you read out, “What the f@#k is xG, Jeff???!”
I know there is a lot of terminology in the game these days so I’m waving the white handkerchief, so let me explain :)
Imagine the football club you support is currently losing a game at home by 0–1. A pass goes through your striker, who has only one defender between him and the goalkeeper, and he manages to beat the defender… just barely, but then he blazes it over the bar, wasting what you apparently thought was “a golden” chance. Don’t worry, Darwin Núñez, my Liverpool friends told me not to tell you we were talking about you.
Or, better yet, you are at another game where your country is playing. The game is 1–1. One of the midfielders takes a very good shot from just outside the penalty area, but the opposition goalkeeper manages to make a very good save… but he spills it! Your striker is at hand to score from the rebound, but oh? What has he done? Your striker, who had a “very good” chance to bury it in the back of the net, smashes the ball against the post!
Or you are in extra time of a World Cup final, and the score is 3–3, thanks to a teammate’s hat trick. We are only SECONDS away from the dreaded penalty shootout. A very good pass is played to your striker, but the defender is there, and it looks like he is going to clear it… Oh yes, he misjudges the ball and misses his kick; your striker is through on goal…. He shoots it straight at the keeper!! Oh no, everyone is gasping.
Kolo Muani, frèro, s’il te plait, c’est pas toi!
He SHOULD HAVE scored that you say. Your uncle, an old man who needs a cane to walk, also says he would have scored that easily. Someone says it was a difficult chance because the ball came too fast or the angle was too tight, but now your dad, absolutely fuming, chips in and says he would have controlled that ball like prime Zidane, dribbled past the players between him and goal, and scored it easily. Even everyone who knows this is the funniest thing they have ever heard your dad say does not even realize it because of the shock of such a miss.
Everyone is fuming. Everyone is berating the player who missed that. Some people try to sympathize, but nobody seems to entertain that.
What do you classify as “a golden” chance? How do you even know how easy or difficult it is to convert a shot into a goal? Is being one on one with the goalkeeper enough to say that is a clear chance? Is a clear header on goal a golden opportunity? How about a shot from outside the penalty area?
Enter our friend xG!
xG stands for "expected goals." It is a model that measures the probability of a shot becoming a goal. We all see players take shots and headers on goal all the time during football games, but xG tells us the actual quality of that shot. The metric also tells us if that shot had a high probability of resulting in a goal or not. It measures shots based on other variables and produces a figure ranging from 0 to 1, with xG values closer to zero indicating a very low probability of a shot resulting in a goal and xG values closer to 1 indicating an extremely high probability of ending up as a goal.
For example, an xG of 0.2 indicates that that shot has a 20% probability of becoming a goal, or, in simpler terms, if you take that shot 10 times, you would be expected to score it twice, so basically a bad chance.
Trying to take a shot from outside the penalty area would have a lower xG value close to zero. But how close? To put this in some context, a penalty kick’s xG value is usually fixed between 0.73 and 0.80, depending on the xG model being used.
For example, according to FBREF, Di Maria’s goal in the World Cup Final game against France was worth 0.38 xG. Mbappe’s second goal of the game, for example, was worth only 0.12 xG, a very difficult chance, and had the lowest xG rating of all the 6 goals scored in the game. Lionel Messi's second goal, a rebound from Lloris' initial save, was the easiest opportunity of the game. It was worth 0.86 xG. Kolo Muani’s chance, for what it’s worth, was the highest xG chance that did not end as a goal. The xG worth of that shot was 0.30, which, objectively speaking, was quite a difficult chance.
How is xG calculated, I hear you ask? The model is fed by key variables such as: the body part the shot came off; the foot or head, was it a big chance, location of the shooter; distance and angle, what offensive action led to the goal; a dribble, a cross, or a setpiece, alongside other variables, and awards a score from 0 to 1.
As you can see, xG models do not only factor in the shot at goal itself and classify the chance as an easy or difficult chance, but they contextualize the shot to tell us if that shot was a quality shot or not.
There are quite a number of xG models available, each of which tries to factor in more metrics to actually qualify a shot at goal as a quality chance to score or not.
Back to this article now. For goalkeepers, measuring their performance based on the perspective of the shot taker is a bit unfair, as there is a primary factor that comes into play: the goalkeeper’s positioning after the shot is taken. Enter PSxG!
I know the terminology might make your head spin, but stay with me. PSxG is short for Post-Shot Expected Goals, and it is a model that measures the shot stopping performance of a goalkeeper.
After a shot is played, the keeper’s positioning relative to the ball immediately after it is struck primarily determines the keeper’s ability to stop the shot from going in the net.
The model factors some variables, such as the distance of the shot, shot power, the angle at which the shot came from, the placement of that shot, etc., to predict if a shot will go off target, 0, or if that shot is highly likely to be a goal, 1.
Usually, PSxG is higher than xG as goalkeepers are usually more protected by defenders or the angle. Even for penalties, keepers have a 0.99 PSxG value, indicating the extremely likely outcome will be a save as the keeper is near perfectly positioned relative to the ball’s location.
Back to our data. We will measure the shot-stopping ability of each club’s goalkeepers using the PSxG metric and the number of actual goals the club has conceded. I'll create two plots for this article. The very first one will be for all clubs in general, and the subsequent ones based on different metrics will be for individual goalkeepers.
I chose a scatterplot to visualize relationships between metrics being used for our comparisons and also because it was the best way to show the location of each club's or goalkeeper’s performance, relative to all other clubs, in the case of the first two plots, based on shot-stopping performance.
Fulham, Bournemouth, and Nottingham Forest have conceded the highest quality shots on goal. Bournemouth conceded 32 goals, the highest number of actual goals. They underperformed their PSxG, as they were expected to concede at least 26 goals. Nottingham Forest, the third club behind Fulham and Bournemouth in terms of highest quality shots faced, conceded 30 goals, the second highest number of actual goals conceded, when they were expected to concede around 26 goals too.
It is intuitive that the more quality shots a keeper faces, the higher the probability of conceding, and this is where Liverpool’s goalkeeping performance is quite impressive. When we analyze the clubs that have conceded a below-average number of goals, Liverpool faced the highest quality shots but conceded only 17 goals.
To be more technical, Liverpool were expected to concede at least 24 goals based on the quality of shots on their goal but managed to concede only 17 goals, about 3 goals less than the league average. This is an overperformance of their PSxG and might suggest they have a very good goalkeeper.
Southampton on the other hand were the highest underperforming team in terms of PSxG. They faced just below the average number of high quality shots yet still conceded way more goals than the competition average. In simpler terms, they were expected to concede at least 18 goals yet managed to concede 27 goals. A massive underperformance of their PSxG by at least 9 goals. At least a few of these conceded goals could have been stopped. This could indicate that their goalkeeper was poor at stopping shots during the period being analyzed.
In the bottom left quadrant, we see that Arsenal faced the lowest quality shots and conceded the fewest number of goals, but in terms of shot stopping performance, Newcastle edges it, as they were expected to concede 13 goals but conceded only 11, the same as Arsenal.
There are of course more ways to interpret this, and one of them could be that Arsenal have generally defended better than Newcastle as they allow lower-quality shots on goal.
The bottom left quadrant generally groups teams who conceded below average number of goals and below average quality shots but there is a catch. Manchester United, for example, while conceding fewer goals than the league average, actually underperformed their PSxG. They were expected to concede at least 16 goals but conceded 20. Brighton were also expected to concede at least 15 goals but ended up conceding 19.
In the upper right quadrant, Bournemouth were expected to concede 26 goals but instead conceded 32 goals, the most in the league, and also recorded the second highest underperformance of PSxG among all clubs, which may indicate that their defense has not only done a poor job of protecting their goalkeeper to limit the quality of shots they faced, but they also have a goalkeeper who has underperformed in terms of keeping the ball out of the net.
Context is everything. While PSxG analyzes performance based on key metrics, the reason why a club might be conceding more or less goals than expected could be due to a host of factors.
For example, in a good defensive team (let's say based on goals conceded less than the average number of goals conceded), a player makes a bad pass that allows an opponent attacker to get in on goal, or a keeper makes a save but there is a rebound, and so on, which makes their keeper face a higher quality chance. So conceding higher quality shots is not always the result of a bad defense.
As my statistics tutor used to say, “Jeff, C, not C!” which was code for "correlation is not causation."
There are other factors that could be used to support a goalkeeper’s shot stopping performance, but this serves as a baseline. From here, more indepth insights can be extracted by factoring in context and other variables to subject shot-stopping performance to
For goalkeepers, we had to take some players out. Clubs like Everton, Villa, Bournemouth, Chelsea, Nottingham Forest, Fulham, and West Ham fielded 2 goalkeepers during the duration that has been analyzed. The goalkeeper with more games was used, or their preferred goalkeeper based on appearances.
Alisson overperformed his PSxG massively and, according to this metric, was the highest-performing goalkeeper based on the number of actual goals he conceded.
Other honorable mentions are Pickford and Nick Pope, who faced high quality shots but conceded fewer than they were expected to.
Gavin Bazunu and Meslier were underwhelming, conceding way more than they were expected to. Bernd Leno could also be considered as a good shot stopper if we adjust our perspective a little. For goalkeepers who faced above-average quality of shots and conceded above average number of goals, he was the best performing goalkeeper, conceding only the 24 goals he was expected to concede which suggests that he is quite adept at stopping lower quality shots.
SWEEPING
We would also like to further which goalkeepers stay further away from their line on average and measure their efficiency based on how many defensive actions they performed. A defensive action could be a header, an interception, a clearance, etc., performed by a goalkeeper away from their goal.
By analyzing sweeping performance based on this metric, would like to know which goalkeepers, stay higher up the field more and which of them make the most defensive actions outside their penalty area.
Here we notice that on average, the goalkeepers who stay higher up the field the most are Nick Pope and AlissonBecker. Generally speaking, more goalkeepers stay higher above the field, with the goalkeepers who stay close to their goal most of the time being Martinez, Fabianski and Sa, in that order.
West Ham’s Fabianski however made the least defensive actions in the league and Newcastle’s Nick Pope making the highest. To contextualize it, Fabianski made only 4 defensive actions outside his penalty area while Nick Pope made a staggering total of 36.
I would also like to highlight the performance of Ederson here too. He is regarded as the best sweeper goalkeeper in the league but is outperformed by Alisson and Nick Pope here. This does not purely mean a bad thing, but could actually mean more if you contextualize the possession, passing, and defensive statistics of his club, as they are a more attacking side, hence the fewer defensive actions he has had to perform in comparison to the aforementioned goalkeepers.
Despite that, his statistics are above the league average for both distance from penalty area and defensive actions made.
Distance from goal can provide information about how a team plays, such as whether they defend more or attack more, but we must be cautious because there are teams that defend deeper or stay higher on the pitch, but their goalkeeper performs more or less defensive actions outside their penalty area and/or stays higher or lower up outside his penalty box. Some fair examples of this are Ramsdale, Mendy, Henderson, De Gea, etc.
CROSSING
Aston Villa’s Emiliano Martinez was generally the best for claiming or stopping crosses. Despite facing fewer crosses than the league average, Martinez stopped 26 of them, four more than Raya, who faced 43 more crosses.
When it comes down to the percentages, Martinez is still the best cross stopper. He saved 15% of the crosses he faced. Raya, Ederson, and Wolves' Sa finished third, each with a 10% cross-stopping success rate. Bazunu recorded the 3rd highest cross stopping success rate with 9%
This metric may also suggest that teams like City, Arsenal, Liverpool, Fulham, Brighton, and Leeds stop more crosses into their penalty area, while other teams like Forest, Everton, Brentford, Southampton, Manchester United, Leicester, and Crystal Palace fail to stop more crosses into their penalty area. The crosses-stopped metric here is only looking at it from the goalkeeper’s perspective to see which keeper is the most authoritative in his box, Martinez, and which keepers are less authoritative, Pickford, Alisson, and De Gea, who all posted a meager 3% success rate for stopping crosses.
This metric does not factor in cross accuracy, as a cross might be successfully cleared, headed away, or intercepted by a defender, for example. It only factors in the number of crosses claimed by the goalkeeper as a fraction of the total number of crosses he faced.
PASSING AND GOAL KICKS
Now let us examine an arguably more exciting statistic: passing and goal kicks.
Based on the data we used, we can only check for average pass distance, accuracy of long passes, excluding goal kicks, average goal kick distance, so we can see which keepers find their teammates the most with long balls, which keepers kick the ball highest up the field and the passing range of each goalkeeper.
From our data, Raya attempted the most passes among all goalkeepers in the league. Bazunu and Ederson came in second and third, respectively. An interesting observation here is, although Raya attempted 49 more passes than Bazunu here, Bazunu’s passes were much higher up the pitch among the two goalkeepers. In terms of pass length, however, Bazunu came in second behind Everton’s Jordan Pickford.
Neto attempted the fewest passes out of all goalkeepers but came in at joint 4th for the longest average pass distance alongside Raya despite attempting a whopping 363 passes less than Raya, indicating that he is a goalkeeper who prefers to hit his passes hard and long. We could actually determine if he plays a higher proportion of his passes longer if we had his full pass statistics, but we do not have the data for this.
Also, Alisson, Ederson, Ward, and Leno, for example, attempted more passes than the league average, and the distances their passes traveled were less than the competition average, suggesting they are keepers who like to keep it short and (don’t do it, Jeff!)… sweet.
A goalkeeper’s average passing distance could suggest a team plays out from the back more or likes to go long, but then again, remember. C, not C!
Ederson has the lowest average pass distance, suggesting he attempts more short passes than longer ones, and if you add Manchester City’s style of play, you find out this holds some weight.
LONG BALL ACCURACY
Now on to long passes. Excluding goalkicks, we will like to analyze the average pass distance of the goalkeepers. This is similar to their general passes, but it only focuses on long balls, that is, passes of at least 40 yards. We’ve seen how high goalkeepers launch the ball up the field, but this metric would help us zoom in further to see the accuracy of only the long balls a goalkeeper hits during open play.
From the previous plot, one might have asked? Why do keepers like Bazunu, Pickford, Henderson, and Raya attempt so many long passes?
Well, if you put it into perspective, it worked for them, and much more so for Raya and Pickford, for example, as they attempted and completed the most long balls. If it ain’t broken, don’t fix it, right?
One might ask, "What about Mendy, Ederson, and Alisson, for example?" Why do they attempt and complete even fewer long balls? Well, the fewer long balls they play, the more they do not complete these passes, and they are either intercepted or go out of play, so why keep playing them long and giving the ball away when they can play it short and keep the ball?
Generally, most keepers in the league attempt and complete below the average number of long passes attempted and completed. One would ask if it would have been more prudent to use the median as the measure of center rather than the average. This is because the values for both measures are not too far apart, and by using the average, we can better visualize and preserve the stark difference in styles, for example, rather than using the median, which punishes long ball experts like Bazunu and Raya.
The bulk of keepers are crowded around the midpoint but still below the average attempts and also, below the average completion rate, indicating that, longer passes fail to find their teammates more for them so they are attempt them less.
GOAL KICKS
For goal kicks, Tottenham’s Hugo Lloris took the shortest goal kicks, and Newcastle’s Nick Pope took the longest goal kicks. Surprisingly, Arsenal’s Aaron Ramsdale took the second-longest goal kicks based on average distance among all goalkeepers in the league, while Ederson and Alisson took the shortest goal kicks for keepers in that range. Henderson, Raya, and Bazunu, who are already known for their long ball kicking ability, also took the most and longest goal kicks for keepers who attempted more goal kicks than the average number.
To generalize, we can assume keepers who restart the game with shorter goal kicks (goal kick distances below the league average) could be classified as keepers who play from the back often, and vice versa for keepers who restart the game with longer goal kicks.
However, there is a catch here: keepers like Ramsdale and De Gea are widely regarded as playing for teams that play more often from the back, yet they go long frequently; this could indicate that they are uncomfortable doing so or that they have a preference for longer goal kicks.
The same can be said for the teams above too. Keepers who take longer goal kicks do not necessarily play for direct teams, and vice versa.
To add a little context here, both Brentford and Manchester City have very physical forwards in Erling Haaland and Ivan Toney, but notice how City’s goalkeeper Ederson typically takes shorter goal kicks and attempts shorter passes? And how Brentford’s Raya customarily take longer goal kicks and attempt longer passes?
This offers some classic context to the style of play of both clubs. While Raya is more than happy to launch goal kicks and passes to, say, Ivan Toney, so Brentford can use his very physical build to win aerial duels or hold off defenders and then utilize his technical quality to link the midfield and bring them into play to move Brentford up the pitch.
City is very much averse to playing this way. They use Ederson’s passing ability and the technical ability of their defenders and midfielders to progress the ball up the field, as that suits them more.
To add more context, Brentford has fewer technical players to do what Manchester City does, so they play to their strength by going straight to Toney.
While Haaland might be a very clinical scorer and also dangerous in the air, his ability to link up play is widely regarded as inferior to Toney’s, and given the fact that City love keeping the ball, they would not want to waste this advantage by playing directly as opposed to the other way.
One might think being direct is only for low-block teams or generally defensive teams, but based on the sweeping metric we measured above, Raya averaged more actions at a distance higher than the league average, once again telling us why “C, not C!” is so important. He actually ranks 4th, just behind Ederson in terms of keepers who stay higher up the field.
This shows us that these metrics, while they are baseline metrics that can be developed upon by adding more context, generally inform how a goalkeeper is performing.
CONCLUSION
To sum it all up, based on the expected number of goals to be conceded, which is calculated based on shot quality, Alisson Becker was the best performing keeper, overperforming his PSxG of about 24 expected goals by 7. He conceded only 17 goals when he was expected to concede at least 24. Despite conceding fewer goals than Alisson, 11 each, Ramsdale and Pope faced lower-quality shots than the competition average.
Bazunu and De Gea were the worst-performing goalkeepers, judging by the quality of shots they faced. De Gea faced less than the competition’s average shot quality yet conceded more than the average number of goals, while Bazunu faced shots above the average quality yet conceded more than seven goalkeepers who faced even higher quality shots.
In terms of sweeping, Alisson Becker and Nick Pope averaged the highest distance away from the penalty box among all goalkeepers, but Pope made 11 more defensive actions than Alisson, making Nick Pope the busiest sweeper goalkeeper in the league. Martinez, on the other hand, was the goalkeeper who stayed closest to his goal and made only seven defensive actions in all of the games he played in. Martinez, however, was not the least busy goalkeeper; that honor went to West Ham's Fabianski. He made only 4 defensive actions in all the games he played in, which was way less than the competition’s average of about 17.
Martinez, the recent World Cup winner, was the most authoritative keeper in terms of the number of crosses he stopped. He was able to stop 26 crosses, 4 more than the next-best keeper, Raya. It is worth mentioning that Martinez faced not only 43 fewer crosses but also fewer crosses than the league average. At the other end, Alisson stopped the fewest crosses in the league.
Passing, passing, passing! Ask most Premier League fans who attempts the most passes in the league, and a huge number of people will say Ederson, but surprise… He comes in third! No, it’s actually Raya who completed the most passes in the competition, followed by Southampton’s Bazunu. Mendy and Neto attempted the fewest passes, but both did not play so many games, so if you adjust to measure the fewest pass attempts, Nick Pope was the goalkeeper who hit the fewest passes. Everton’s Jordan Pickford averaged the highest passing distance out of all keepers, trailed closely by Bazunu and Henderson. Ederson averaged the lowest distance in the league, which, given his extremely high passes attempted and if we were able to factor in a couple of variables like touches, might suggest he is the goalkeeper who is most involved in build up play, as he does not play it too far while still producing an extremely high volume of passes. Alisson Becker was also averaging only a few meters less than Ederson’s average length as well as attempted passes. Among the goalkeepers who hit fewer than the average pass distance, Neto and then Fabianksi averaged the longest pass distance which seems to indicate that they like to hit it to find someone higher up the field. In the case of Fabianski, an argument could be made that he does that because West Ham have Antonio and Bowen, two quite physical players who are higher up the field.
Again, when it comes to long passes attempted and completed in open play, Bazunu and Raya again, made and completed the most long passes in the league. Bazunu attempted a whooping 324 long passes, finding his teammates 121 times while Raya found his teammates with 129 out of his 321 long balls. This approach, while less efficient because it took an awfully high number of attempts to achieve that, was effective for their respective teams as they found their teammates more than any other goalkeeper in the league. Pickford came in third with a 105 successful long balls from open play out of 288 attempts. Mendy, Meslier, Ederson and Alisson completed the least long balls, in that order, with Alisson coming in at fourth if we adjust it because Mendy played fewer games than the others. All the goalkeepers combined produced an average of 35% accuracy with their long pass attempts, which might suggest that it is not a very effective tactic. Neto recorded the highest long pass accuracy with about 47% accuracy. However he played in only 9 games. Alisson Becker played in all games and had the next highest accuracy with about 43%, with Fabianski finishing close behind in third with about 42% accuracy.
Finally, Nick Pope averaged the longest goal kicks while hitting way below the average number of goal kicks. Aaron Ramsdale was next, hitting longer goal kicks with fewer attempts. Hugo Lloris took the shortest goal kicks not only among keepers who took more goal kicks than the league average, but also among all keepers in the league as well, hinting that he is a keeper who likes to restart play with passes to his centerbacks or fullbacks more.
Alisson Becker outperforms most goalkeepers in the league in key goalkeeping metrics. Although he ranked very low for crosses claimed, he was the best shot stopper, and in terms of sweeping, he came in second to only Nick Pope in terms of most defensive actions outside their penalty area. He also ranked first in open play long ball accuracy.
Honorable mentions include Nick Pope, Ederson, Raya, and Bazunu, who were specialists in specific areas but not as well rounded as Alisson.
Thanks for making it to the end of this lengthy article. I will conduct more detailed analyses of key areas or key metrics for Europe's top leagues, as well as analyze player performance for specific positions on occasion and suggest players who might fit a team’s style of play more based on how they play.
You can reach me here on LinkedIn.
A link to my Tableau page in due course after I finish organizing and putting together the worksheets in a dashboard or as a story.
The statistical programming tool I used for this analysis was R, and all ploty were made by me Jeffrey O. with data from FBREF.