If there is an Expected Goal (xG) stats for an average of 20 shots in a match, why not give this data for 600-700 passes?
We had a 2-step problem in front of us to measure the Expected Pass and Pass Value.
First, We had to determine the probability that a pass with various features entered would be accurate.
Secondly, the contribution of a pass to the game.
According to the numbers of Whoscored, Zanka made 14 matches with Fenerbahçe jersey and made an average of 62.3 passes per game. This value stands out as 60.1 in Serdar Aziz and 51.4 in Luyindama.
So how much of this is a valuable pass?
Although there is a distinction between various data providers such as forward pass, lateral pass and back pass, we think these are insufficient when considered in detail. We will try to explain this situation more clearly by giving examples with videos.
Let’s create our “Expected Pass” model first.
Expected Pass
A pass either goes to its target or not. So we have 2 possible outcomes.
So we have to develop a model that can say “95% chance it will be successful.” when we enter the features of the pass.
What do we need to find out what percentage of a ball sent from point A to point B will succeed?
First of all, the information about the point where the pass was thrown and the point where it was sent. It would be great if we add the length of the pass.
The type and speed of the pass also seem to be important because a pass in open play and set-pieces will be different.
The height of the pass and how it is given is also important. A ball thrown with the right foot from the ground and a ball sent from the air with a head may differ.
What if there is pressure on the player to pass? Then things get a little hard, right?
We think we have agreed so far. However, two more important variables remained. Let me explain them visually.
Consider a pass from point A to point B. The point where the ball comes out of the foot sees the goal with 9 °, whereas it reaches 23 °. As a result, the fact that there are more defense obstacles in the central area of the field reduces the chance of the pass to be successful.
Another issue is the distance of the starting and ending points to the goal. A pass thrown right in front of the opposing goal should not be too high.
We accessed LaLiga data shared by StatsBomb and tried to make models using the variables we highlighted above.
We create our final model with Logistic Regression with all the variables in the highest result from the Roc curve.
We encountered a result that we predicted exactly. No matter how well the place where the pass goes to see the goal or how close it is, the rate of accuracy of the pass decreases.
Since StatsBomb shares the 2018 World Cup event data, we have a chance to try our model on 64 games played in Russia. Let’s see who were the ones who successfully hit the hardest passes.
When we examine the above graph, we see that the passes with low xP (We will call xP from now on.) are mostly inaccurate.
This situation does not surprise us; however, let us answer the questions like what is the high xP pass or what does the pass representing the opposite look like in terms of the full revival in our mind with 2 short videos.
There is no pressure from the opponent, you are in your safe zone, you are very far from the opposing goal, etc.
For the xP to be high, there are all kinds of joints that are exactly what you watched on the video.
Let’s take a look at the ones that are unlikely to be accurate.
The distance of the pass has been extended. The ball traveled to the dangerous area. The angle to see the goal where it went has widened. As a result, xP decreased significantly.
When we filter the players who made more than 100 accurate passes, we see the graphic above.
If you look carefully, the more passes a player makes, the higher their average xP.
Before doing this work, we thought that the best passer is the player who sends even the most difficult balls to his target and we guessed that the results will be in that direction.
If we look at Luka Modric, this estimate may be partially correct; however, when we analyzed the results from position to position, there was a point that we missed.
Sending low xP passes with accuracy is undoubtedly an important success.
But we ignore something.
The algorithm lowers the xP of long-distance, airborne, oppressed balls. This pass from Tosic has all the features we counted. Its xP is 0.05.
So is it really a valuable ball?
Although the passes of Tosic and Witsel are almost identical, one of them is going to a very dangerous point.
So we need another parameter to evaluate the passes.
New Doors Opening xG
Models that measure the value of each action in the field, such as Goals Added or VAEP, use expected goal.
How is it used?
(Expected Goal of the post-action location – Expected Goal of the pre-action location)
The formula is pretty clear.
If the action resulted in a more dangerous area, the player made a valuable move.
So if we integrate this idea into our xP model, we can measure the value of a pass.
Thus, a pass that has both low xP and high xG ground will be more valuable than others. (If hit)
We have created a simple Expected Goal model with location, distance and angle variables.
Now we can take the xG difference of the start and end points of each pass and combine it with the xP.
When we take into account only the accurate passes thrown in open play, we found that on average, the ones sending the most valuable balls were the back and wing players.
The main reason for this is that the zone they are playing in sees the goal from a low angle.
What does this mean?
Points close to the line are points that negatively affect xG as an angle. Therefore, the next stop of the ball will likely be the center of the field.
The ball will go to the 1st area of the opponent or return to the central midfielder who cames for support.
Therefore, it would be useful to evaluate everyone according to their position.
Now, with a few passes thrown in the 2018 World Cup, we are leaving the data world and sailing to the game itself.
The value of this hard ball with an xP of 0.05 is 0.38.
The touch of Iniesta, which is half the probability of being accurate, has a value of 0.04.
We have to remind you here that it is a high score when hit by an average of 0.04. Usually, the matches form passes of 0.001.
We would like to share 3 passes with the highest value in open play.
The most important reason for Busquets’ seeming simple pass’ high value, the point where the ball touches the ball is very narrow angle of seeing the goal. At the point where the ball arrived, there is a very high xG. Diego Costa has only one touch left.
Don’t you think James’ fine touch to the ball shows the quality right away?
Although the action before the pass is not taken into consideration in our model, the Brazilian winger’s run with the ball is also remarkable. The intensity and target point of this pass is just as wonderful.
Smart Passes
Although not as high-value as the examples above, there are some smart passes. What we mean by the word “smart” is that it is not very likely to be accurate, but it is a valuable pass.
The xP of Isco’s pass to Iniesta is 0.70. So it will be highly accurate. But its value is 0.01. As we mentioned above, this figure is far above average. Isco actually made a strategic pass.
The xP of the ball sent by Griezmann seems to be 0.39. It is not an impossible pass, but an important pass. Its value is about 0.02.
Mbappe’s one-touch accuracy is much higher, but that one-touch puts his teammate against the goalkeeper. So it has a value of 0.03.
Of course, when we examined the data that we came across, we encountered various oddities.
We want to share two of these with you.
The value of Marco Reus’ pass is exactly 0.39.
When we first watched, we said this was it. A fleeting ball may have reached its goal even by chance, but its location is a dangerous point. We are curious about your opinions.
We are going back to Iniesta again. In the example we will end, we wanted to use his position.
Although this pass fails to achieve its purpose, it is an accurate pass according to the data provider.
Its value is 0.26.
Should we really give this value to Iniesta? Or should we punish?
Overtime
In this article, we used successful passes thrown in open play as an example.
Because we did not bend too much on how the failed passes and set-pieces should turn into a score.
Is it important to throw a ball with an expected pass of 0.01? Is it a good passer’s job to try a ball that will be hit by a hit even a thousand times? Or does the player who makes the ball think of throwing simple but effective passes?
As these questions are a new research area, we did not want to mention them here.
If we come to the question of how to develop or analyze the model, we must definitely determine each player according to the standards of his position. So we can only break the hegemon of wing and back players.
Much larger match data is required for this. we focused only on the 2018 World Cup within the framework of the possibilities at hand.
We must have many more examples to train the model. The more examples, the more successful models…
How happy if we could explain the “essence” of the work.
Although it is not yet spoken in our country, there are people and institutions working in this field. You can find their link from here.
Stay with data.
Kurucu Ortak & Şef Scout, the FA ve the PFSA lisanslı scout, maç & veri analisti, Gürcistan sorumlusu.
Henüz yorum yok