Salah’s Season in Statistics: A review of the Egyptian’s first year at Chelsea
With Mohamed Salah’s debut half-season at Chelsea over, we here at KingFut have decided to take an in-depth look to see how the Egyptian starlet fared. With Chelsea already busy in the transfer-market (and stacked in attacking midfielders), Salah’s future may hang on his performance over the last 6 months; so how has he done?
Revisiting Chelsea vs. Liverpool*:
*My little ‘I told you so’
Back in February, I examined whether Salah made the right move picking Chelsea over Liverpool. Chelsea ended up edging the analysis with a score of 3 categories to 2 for Liverpool, and most of the readers disagreed with my analysis. The big sticking point for most reader’s was my claim that Salah could end up getting more minutes at Chelsea in the near future. It seemed counter-intuitive, due to the extremely strong competition at right-wing in Chelsea (Willian, Schurrle) as opposed to Liverpool (Sterling, Moses). The form of Suarez & Sturridge was the deciding factor, however, as it forced Brendan Rodgers to play them together leaving only one option open. Although I couldn’t predict Sterling’s great form, the fact that he is younger than Salah, English, and has been at the club longer, points to Salah being 2nd string at Liverpool.
Mourinho’s penchant for inverted wingers, as well as rotation forced on him by a deep Champions League run, allowed my prediction on Salah to come true, as he played over 500 minutes for the Blues (around six full games). This, combined with six straight starts to end the season, allowed Salah the time to settle in and show the coaching staff his qualities. Doubtful if he would’ve hit that number with Sterling in the form he was in late in the season.
According to Squawka & WhoScored:
– Picked in the squad: 14 times
– Appearances: 10 times (545 minutes)
– Starts: 6 times
– Subbed on: 4 times
– Subbed off: 4 times
– Average time per appearance: 54.5 minutes
– Goals: 2
– Assists: 1
– Shots: 21 (6 on target)
– Key Passes: 9 (passes that lead to shots)
– Passing %: 181/220= 82%
– Successful dribbles: 16
– Dispossessed: 18 times
– Fouled: 14 times
– Interceptions + Tackles : 14
– Aerial duels: 2/4
A quick look at the raw numbers seems to imply that Salah has had a pretty good opening season with Chelsea. Shooting percentage (only 6 shots on target out of 21) seems to be an issue, but a return of 2 goals isn’t bad and 9 key passes seems pretty good. His dribbling seems extremely high, although he does get dispossessed too often it seems.
Here’s the problem with all these numbers, we have no context with them! Is 16 dribbles high? Is it low? Can we compare across players, positions, leagues? The first thing we have to do is turn them into per minute, or better yet, per 90. Dribbles per 90 shows us how many dribbles Salah completes per game basically. We also introduced Non-Penalty Goals per 90 in the piece discussing the best goal-scorers in Egypt, a better measure for goal-scoring.
Salah’s Per 90 Stats:
– NPG: 0.33
– Assists: 0.16
– Shots: 3.47
– Key Passes: 1.65
– TB: 0
– Dribbles: 2.64
In order to compare Salah’s season to others across Europe and the EPL, we will be using the excellent Ted Knutson (of Statsbomb) radar charts. Ted has gathered data over the last 5 years in all of Europe’s top leagues, and therefore can give us context for Salah’s numbers compared to other players. His radar charts are normalized so that the outer boundaries mean the player has hit top 5% in Europe for this stat, while his lower boundaries means bottom 5% in Europe.
Statbomb has also created a different radar charts for different positions. Here are a couple of examples:
A CM Radar for the excellent Ivan Rakitic:
The raw numbers are bottom left. Scoring contribution equals Non-Penalty Goals per 90+ Assists per 90. The inner-most circle means bottom 5%, while the outer-most circle means top 5%. Rakitic is top 5% in Europe both in Key Passes and in Scoring Contribution.
As you can see above, the amount of fouls, the times dribbled past, tackles & long balls are all stats that are more important for a CM than a striker or AM like Salah.
Here is the adjusted AM chart, with a certain CR7 visualized:
Ridiculous NPG rate of 0.89 this year for Ronaldo (doesn’t show up since its above top 5% in Europe), as well as 7.65 shots (wow). 2.34 dribbles while being dispossessed only 1.49 times is also pretty impressive. So as you can see, TB, Shooting % and Goal Conversion % is added to the AM chart.
So how does Salah compare? Using data from WhoScored and Squawka, I tried to replicate the StatBomb radar:
Not bad at all from Mo Salah. Top 5% dribbler with 2.6 dribbles a game. Great shooting with almost three-and-a-half a game. Good passing percentage (almost top 5%). His scoring rate is OK for a wide forward, not spectacular.
Dispossessed far too often at almost three a game however, and his Key Passes numbers are quite poor. Also, Goal Conversion is not good enough at 12%. Non-existent through-balls, although that can be a tactical effect (and an Eto’o effect).
So how does Salah look compared to his fellow AM at Chelsea?
The dribble monster Eden Hazard:
And direct competition Willian:
Willian has had a good first year at Chelsea, and his defensive numbers stick out (especially in a Mourinho side!) with over 3 interceptions+tackles a game. An above-average in almost all areas; the one area where he struggles a little seems to be shooting % and turning shots into goals.
In comparison, Salah has had a pretty good year. The one thing that sticks out in a positive way is his number of shots, much higher than both Hazard and Willian, which is a positive sign. The negative seems to be his passing, as both Hazard & Willian blow Salah out the water in Key Passes (and they play with the same strikers).
Problems with the Radar:
Three things stick out when comparing players using radars:
1) The area does not necessarily represent anything in the graph because two individual stats next to each other do not correlate.
2) Some of these stats are team-level stats (goal conversion depends heavily on quality of chance, for example)
3) Some of these stats aren’t repeatable, meaning they are random, and show more luck than skill.
Even goals (somewhat counterintuitively) aren’t a repeatable skill due to their fluctuation. So Salah’s goal rate this season of 0.33 NPG per 90 doesn’t tell us a lot about his true goal-scoring ability or how much he will score next year. In fact, according to the great work of Omar Chaudhuri over at Prozone here is a graph showing the relations of goal rate in one year with the next:
The correlation between goal rate this season and goal rate next season is a very weak 0.13. So despite the radars being useful for a sense of descriptive statistics (what happened) it doesn’t show us actual level of performance and add future prediction.
Goal conversion is also even weaker:
Let’s dive a little deeper into Salah’s underlying/advanced statistics to get a better indicator of how his first season went and find some repeatable statistics.
Introducing Expected Goals & Assists:
So what are the two things that differentiate great attackers from bad ones?
1) Getting into good positions
To measure both things, we use Expected Goals. Expected Goals is a formula that uses past data to assign a probability to every shot that was taken based on its location, type (head vs. foot), assist (through-balls are finished at a higher rate), etc. So for example, a shot from 1 yard out has a very high probability of being scored (say 80%) whereas a header from 50 yards out has a very low probability of being scored (less than 1%). The average shot has a ExG value of 0.10, or 10%. A striker that accumulates a lot of ExG is a striker that gets into good positions.
Getting into good positions is also a repeatable skill, unlike scoring goals (from this article by 11tgen):
Players who get into good positions the first year, get into positions the second year proving that it is a skill and not luck.
As for finishing, we can measure them by using Residual Goals = Actual goals- Expected Goals. If an attacker has ExG of 10, it means the average player would score 10 goals from the shots he has taken. If that player has scored 15, then he has overachieved as a finisher.
So are residual goals repeatable?
Interestingly, they aren’t. There are 3 possible explanations:
1) Finishing is pure luck
2) ExG doesn’t take into account defensive pressure
3) Shots are too rare to show a pattern
Regardless, getting into the proper positions seems much more important than the actual finishing (where Salah has struggled with a goal conversion of around 10%).
So how has Salah done in terms of getting & shooting from good positions?
Pretty impressive shot chart from a right winger! Only 3 shots from outside the penalty box, a good indicator that Salah gets into threatening positions often. 2.3 Expected Goals according to Micheal Caley‘s ExG formula. Salah has an ExG per game of 0.4, meaning every game he plays he gets into shooting positions that would result in almost half a goal. Assuming even average finishing, that is a great result as it would lead to lots of goals for the young Egyptian attacker. Here are the top 20 Barclays Premier League players in ExG/90 (i.e. getting into good shooting positions) with a minimum of 300 minutes:
Pretty impressive list, and our boy Salah comes out at 18th! Really really impressive by the Egyptian, especially considering that he comes out 3rd if you take away pure strikers. 3rd! Creating 0.41 expected goals a game with his shots, only Walcott (almost a second striker) and Harry Kane are higher without being strikers. The list also fits with conventional knowledge, with Aguero, Dzeko, Suarez & Eto’o leading the way. Since this is the most repeatable (and useful) skill, it is encouraging to see Salah get into such good positions on a consistent basis from the right wing. As he develops, (and since finishing has not been proven to be a very consistent skill year-to-year), this ability should serve him well in boasting his goal-tally and keeping him at Chelsea.
Compared to other Chelsea attackers:
Pretty solid for Salah, coming out as the best out of all his direct competitors. There may be a slight substitute effect here, as well a relatively small sample size, but regardless extremely impressive numbers here for a right winger, and an obvious edge he has on Willian (his direct competitor).
*Gedo is first on the list with 1.5 ExG per game but with only 30 minutes worth of data. Elmo is in the bottom 100 with 0.02 ExG per game.
Just like creating goals by shooting is one job for attackers, creating goals for those around them is also an important objective. Although we use Key Passes as a proxy for chances created, the advanced stats community has taken it one step further and incorporated the ExG formula onto assists to create Expected Assists. Expected Assists is basically weighing the key pass (which is a pass before a shot) with the probability of scoring and assigning it to the passer.
In other words, not all Key Passes are equal:
Attacker A splits defense and gives Attacker B a shot from 1 yard out with an ExG value of 0.99. Attacker B gets 0.99 on his ExG value and Attacker A gets 0.99 on his ExA total.
Attacker A gives a key pass for an Attacker B header way outside the box worth 0.01 ExG. A gets 0.01 ExA.
So a Key Pass > Assist because it doesn’t penalize the passer for an attacker that missed an easy chance.
And ExA > Key Pass because it doesn’t treat all key passes/shots equally and weighs the probability of them getting scored.
Here are the top 20 in Expected Assists in the EPL:
So David Silva’s passes create 0.4 assists a game assuming average strikers finish them.
Any list that starts with David Silva and includes Ozil & Nasri in the top 10 seems fine by me! Here is the one area Willian seems to have a clear edge on Salah.
Salah comes in at 116 with a value of 0.13.
Here is a chart of his Key Passes:
Not nearly as impressive as his Shot Chart, a lot of Key Passes that lead to shots from outside. Salah really needs to work on his creating and passing if he wants to take the next step. 9 Key Passes and 0.7 Expected Assists is a poor number (shown in the radar for ‘KP’ as well).
Here are the Chelsea attackers, again:
Salah’s competitors seem to be better passers than he is, an area of concern.
Combining ExG+ ExA:
Since a goal and assist are equally valuable (both lead to a 1 on the scoreboard), how does Salah compare when adding both?
According to this Micheal Caley graph, Salah comes out in total top 20 for creating (by passing & shooting) goals in the EPL this year per game:
Seems like the numbers are a little different to the ones I cited earlier, but not by much. In the end of year spreadsheet Micheal sent, here are the Chelsea attackers:
Salah: 0.54 (!!!)
The other attackers passing numbers allow them to almost catch Salah, but he is still the biggest contributor of ExG for Chelsea other than their three strikers. This is great news, especially since Expected Goals are the best way to predict actual goals & final league table!
Great first season from Salah in England. Settled in, got minutes, got into great positions. Finished at an average rate which is fine considering how good his movement is. Elite dribbling. Needs to get dispossessed less, as well as work on his passing & creating. Defensively sound, but his competition in Willian is a defensive monster so will need to keep up the pressure. Schurrle seems a clear 3rd choice behind Salah and Willian on the right, but his versatility helps.
Here are a couple of radars that are interesting to look at and compare Salah too:
Sterling’s improvement over the last year from Sami Hernia (good thing Salah wasn’t competing with him!):
A certain Mr. Reus:
Salah on the Statsbomb radar using this create a radar tool:
* For more radars, Ted Knutson is your guy
*Data from Opta and Whoscored
*Expected Goal data from Micheal Caley
/ 4 hours ago
Egypt international Amr Warda has been named Man of the Match as Atromitos beat...
/ 9 hours ago
Al Ahly head coach Hossam El-Badry mentioned that he is aware of being compared...