Defending in the EPL
Football the English Way
For as long as I can remember the British football pundits, media and commentators have talked about how the English game was faster / more physical / had less space (delete as appropriate) than the football that was played far away in a land named Europe.
I have often wondered how they arrived at that conclusion.
The hard pressing, high line tactic that has started to make its way into England in recent years has its roots firmly established in Europe. All other things being equal, this gradual acceptance in England of the hard press should therefore mean that when in possession, players in England should have more time on the ball than their foreign counterparts.
In this piece I’m going to use my Expected Goal data to demonstrate that it really is much tougher to score from any given shot in the Premier League than in any of the other Big 4 European leagues. This assertion is based on the 2012/13 season – it is limited to last season due to a lack of prior data.
Can we demonstrate that it is more difficult to score in England?
In order for the numbers to “prove” that it is more difficult to score in the Premier League we are going to have to make just one assumption. I am fine with this as I think it is a reasonable assumption to make, but I’ll let you decide.
The one assumption required for these numbers to hang together is that “Strikers in the EPL are on average no worse than those in the other Big 4 leagues”.
I certainly think this is the case. For every Bale and Ronaldo that have left the EPL at the height of their powers we have an Aguero, Negredo, Navas, Soldado to replace them. The money that permeates throughout the Premier League means that this assumption is perfectly valid in my opinion. Remember, the quality of the strikers doesn’t need to be better than the other leagues, just no worse on average. On this basis I think we are OK to proceed.
Regular readers of my pieces will know that Constantinos Chappas and I have developed a model that assigns a goal probability (ExpG) for each shot. The probability is based on the location, the type of shot and a few other factors, and was developed using shot information from across the 5 leagues.
The summary ExpG, shots and goals table for 2012/13 for the Big 5 leagues is shown below:
|League||ExpG||nShots||Avg ExpG / Shot||nGoals||Goal / Shot||Goals / ExpG
The ExpG values are a work in progress - this explains why I now have the EPL at 84% efficiency (Goals / ExpG) whereas I had mentioned 90% in a previous article - as we tinker and fine tune them. Note that Own Goals are excluded from the goals totals shown above.
For the 2012/13 season, the quality of the average chance attempted was marginally higher in England than in the other leagues at 0.114 Expected goals per shot. Germany and Spain followed closely behind at a clip of 0.11 goals per shot.
However, look at the Goals / Shot ratio.
The Premier League only actually recorded 0.096 of a goal per shot, which is at just 84% of our expected rate. On the other hand, the Bundesliga teams recorded 0.112 goals per shot. So from what appeared to be less productive shots the German teams managed to score much more goals per shot than the Premier League and indeed, in absolute terms, more goals than our model had expected them to score.
Ligue 1 is also interesting. This league is notoriously low scoring, but this is due as much to the lack of shooting (less than 24 shots on average per game) as it is the average shot quality.
Please keep in mind that for this exercise we are not interested in the total number of goals or shots achieved in each league. We’re interested in the average shot quality in each league and then the subsequent conversion of those average shots as I look at how our ExpG efficiency, which is an objective measure, varies by league.
Possible explanations for the EPL under achievement
I can come up with a few reasons why our model has seen goals in England converted at just 84% of the rate that we had expected:
- Defensive Pressure
- Better standard of Goal Keeping than in other leagues
- Our model, although good, is missing other certain ingredients to capture exactly the rate that any given chance is converted
- Chance or random variance
Most of those possible explanations are fairly wide ranging, and ones that I’m not sure how to tackle at this stage. But using an idea that Sam Green posted in an Opta blog in 2012 I have come up with a method of evaluating the amount of the shortfall in shot conversion that is due to defensive pressure, and it’s this that I want to investigate.
Up until this point, our Expected Goal model looked exclusively at the factors present at the point that the shot was struck.
We have used a similar framework to create an Expected Goal model which is based on where the shot was aimed for, and I’m using this article to introduce this metric.
In order to differentiate these two different Expected Goal values we will now refer to the expected goal probability at the point the shot was taken as ExpG1 and the value after the shot has been taken as ExpG2.
A shot that is off target or blocked has an ExpG2 value of zero as those shots have no chance of resulting directly in a goal, whereas a shot that is arrowed right for the very top corner will have a high ExpG2 value as the likelihood of a goal is strong. An extreme example would be a penalty that is aimed at the top corner; such a shot would have an ExpG2 value of approx 0.96 with an ExpG1 value of approx 0.75.
Let’s have a look at what the ExpG2 values looked like for each of the 5 leagues last season, and most importantly how they compare with their respective ExpG1 figures.
|League||nShots||ExpG2||ExpG1||Diff in ExpGs
I’m ignoring the actual goals scored for the purposes of this exercise, and instead am concentrating on the differences between the ExpG1 and ExpG2 values for each league.
For four of the leagues the differences between the two sets of values are small, at 2% or less. But the difference between the ExpG1 and ExpG2 values for the Premier League last season are extremely significant in comparison at 10 percentage points.
Let me step you through exactly what this means, and the significance of this.
Looking at the situation at the point when the shots were struck (ExpG1) our model had expected to see 1201 goals in the Premier League last season.
Here’s the important bit. When we then take account of where the shot was aimed for the expected total goals dramatically reduce to 1083. Remember that the other four leagues showed comparatively little movement between the two measures. This suggests that, on average, the other four leagues all exhibited similar tendencies on average in terms of the change in probability of a shot being scored after it has been struck compared to before the shot took place. But for some reason, the English Premier League reacted totally differently.
Let’s now remind ourselves of the assumption that I made earlier in this article.
If we assume that the quality of the average striker in the Premier League is no worse than the average of those appearing in the other four leagues then the low Expected Goals total that has been calculated after the shots have been struck can only be due to defensive pressure. As we’re happy that EPL players are no worse at shooting, and we have normalised the quality of the shots, there is no other reason to explain why such a large proportion of shots are blocked, miss the target or are aimed towards the centre of the goal. This same amount of defensive pressure simply seems not to be present in other leagues.
As we have normalised for shot location and are ignoring whether the keeper actually saves the shot or not, we are left with the defensive pressure as the only remaining explanation for the large drop in goal expectation in the EPL between ExpG1 and ExpG2.
In my opinion it’s not important that we break down the reasons for this low ExpG2 totals between missed shots and blocked shots. It’s enough to know that this heavy defensive pressure insures that the probability of a goal being scored in England from any given shot is much lower than our original model predicted, and indeed lower than in any of the other main European leagues.
Due to the lack of proper defensive measures we were always aware of the fact that our model didn’t take into account defensive pressure, but for the first time thanks to the introduction of the ExpG2 metric we can demonstrate the impact of defensive pressure that is epplied in the Premier League.
I am aware that this conclusion is based on just one season of data, but due to the size of the difference in the EPL, the defensive styles that tend to be played in England and the fact that there were almost 50,000 shots in this sample I would be confident that this phenomenon will hold true in other seasons.
Why Go To All That Bother?
Having done the work to demonstrate this, I now find myself asking if I needed to go to so much work in proving something that summary stats could have told us.
For example, Germany, Spain and France all had more than 34% of their shots on target last season; it was just over 31% in England. This alone tells us that defensive pressure was greater, right?
Likewise, only 9.6% of all shots were scored in the EPL last season compared to 11.1% in the Bundesliga, 10.7% in Spain and 10.2% in the traditionally low scoring Ligue 1. Again, I could have just used those stats and said “Due to lower percentage of shots on target and goals I conclude that there is greater pressure on attackers in England than anywhere else”, right?
However, although that would have been a short cut it might not have been correct. The difference in shot on target percentage could have been due to shot locations, ie on average EPL shots were attempted from less advantageous shooting positions, alternatively the lower scoring percentage could have been due to the same reason combined with better performances from goalkeepers in the English league.
This detailed analysis has ruled both of those possible explanations out – hence why I feel it was required in order for me to be able to definitively conclude that the pressure exerted on the player shooting is at a much higher level in the Premier League than in other leagues.
Quantifying the Impact
Over the 380 game 2012/13 EPL season the Expected Goals fell from 1201 to 1083 on the ExpG1 and ExpG2 measures respectively. This amounts to 0.31 goals per game, or just over 1% for each shot as there was an average of 28 shots per game last season.
At this point we reach the take away message from this piece:
Any specific shot in the EPL is scored at a rate of 1% less than the other four European leagues due to the pressure applied to the shooter, or due to other general defensive pressure
A 1% reduction in scoring per shot doesn’t sound like a huge amount, but when shots are only scored at an average rate of approximately 10%, that extra defensive pressure soon makes a difference to the number of goals that are actually scored.
Reasons for the Phenomenon
The central finding in this piece may come as a surprise to many, and indeed will be contested by quite a few, as conventional wisdom would hold that players in the continental leagues have less time on the ball due to the high pressing game that would be prominent in those leagues.
In fact only yesterday, a German based reader made this exact point in a comment in relation to an article where I made the point that our ExpG1 model over-estimated the number of goals that should have been scored in the Premier League:
You´re saying that there is a lot more defensive pressure in the PL than in other leagues e.g. the Bundesliga or Primera Division or even League 1. Well I don´t think that’s true. When Ozil moved to Arsenal, Fabregas said Ozil will enjoy PL because of the space OMs get in the PL. Then Ligue 1 is considered the league with less goals than other leagues. Furthermore the defensive compactness is definitely higher in the PD and the Bundesliga than in the PL. The athleticism may be higher in the PL but as our teams had to learn last year that doesn’t cover the fact that they are almost always split in two parts (an offensive and a defensive one) whereas other teams such as Dortmund, Valencia, Donezk and of course Bayern are defending with 11 men (even barca is defending with 10) and attack with 11.
I would suggest that although the hard press, high defensive line used in Europe certainly reduces space in the middle third of the pitch and leaves it more difficult for teams to be in control of the ball, once those defensive lines are breached the attackers are then faced with easier scoring chances. The defences are not as well set on the edge of the box and there certainly won’t be as many bodies between the striker and the goal.
A perfect example of this was AVB during his time at Chelsea. He attempted to play the high defensive line, perhaps his players weren’t ready for it, but once that line was breached they were powerless to prevent the opposition converting the chances they created.