Finishing
Is it possible to measure finishing? Is the xGOT metric sufficient for this purpose? What are the pros and cons? And a new approach…
When my brother and I play PES on Playstation, I prefer simple passes, as Cruyff said, and bring the ball to the front of the goal. There is only a goalkeeper left or no one. All that remains is to touch. My brother always says: “Come on, you scored a cheap goal again.” And I reply to him: “Didn’t you see what I did to get the ball there?”
To tell the truth, my brother is right about this because he’s looking from the player’s point of view. An easy shot, that’s all. If I speak in football analytics language, it’s not that hard to turn a chance with very high xG into scoring. The player doesn’t need to put an extra quality. That’s why, as it is always said, xG doesn’t measure finishing. The player scored that goal because he was in the right place at the right time.
OPTA’s xGOT metric offers us a different view. This metric paints a nice picture of finishing, as it calculates the value of the shot after the player hits the ball.
OPTA subtracts xG from xGOT to find out which player adds more value to their on-target shots. They call it “Shooting goals added”. The table below shows the figures for this season. The xG of Salah’s on-target shots is 7 (total), but the xGOT figure is much more than that. This means that Salah is a good finisher.
The answer may not be like that.
In the second minute of the Fenerbahçe-Antwerp game played on October 21, the away team’s forward player Samatta scored the goal that put his team ahead. According to FotMob’s numbers, the xG of this shot is 0.82. The xGOT number is 0.97. Samatta added an extra value to his shot. Can we say that this man has shown a good example of finishing?
No, we can’t! Here is the problem.
Look, a chance with almost the same “shooting goal added” value. Saint-Maximin shot the ball into the left corner of the goal. The xG is 0.06, the xGOT is 0.21.
Don’t you think this is a better finishing than the first one?
The goalkeeper was out of position on Samatta’s shot. Let’s look at another clip where the goalkeeper is almost out of the position.
It’s just one of those other chances that add the same value as the other two chances above but clearly differ in finishing ability.
I figure I explained my problem. Now let’s think about something for the solution.
I don’t think it’s effective enough to measure finishing by looking at shots that have a high probability of scoring. These shots will most likely be scored, but that doesn’t mean we shouldn’t evaluate chances like Samatta’s shot. Perhaps the player would have hit the woodwork. My point is, if we’re going to measure finishing ability, it has to be taken into account whether the xG is high or not.
Let’s continue with a simple formula.
Finishing = xGOT / xGxG number is between 0 and 1, the same is true for xGOT. Therefore, the higher the denominator (xG), the less impact the numerator (xGOT) has.
When we apply this formula to the goals above, we get the following results.
0.21 / 0.06 = 3.5 => Saint Maximin0.19 / 0.04 = 4.75 => James0.97 / 0.82 = 1.18 => Samatta
According to these results, while giving the others their due, we took Samatta’s finishing to the level it should be. (I think it’s not even enough.)
In this formula, 1 point represents a standard finishing. If the shot is missed, the result of the formula will be 0, as there will be no xGOT digits, but logically zero.
So is this method perfect? Not enough. The biggest problem is that the shots taken from outside the penalty area are exaggerated according to the formula.
Long shot and finishing are different skills, some will say. They are not wrong. Some players are hard shooters from long distances, but the conversion rate of their shots into goals is quite low. I will keep saying “finishing” without making any distinctions, in order not to deviate from the main theme and not to distract attention.
Where was I? Exaggerated results…
Longstaff’s incredible shot has xGOT of exactly 29 times his xG. Almost all of those with such a high finishing score are shots with an xG of less than 0.02. One of my goals is to measure the finishing ability of the shooter by somewhat standardizing the different types of shots taken regardless of distance.
What are the obstacles that keep me from this aim?
First of all, as I just mentioned, the xGOT limit needs to be a little higher for those with low xG to have a standard finish.
I figure this shot has a standard finishing it should have. Are we waiting for his chance to score? I think no. Is it a bad shot? The answer is again no.
However, this is the case for shots with a %2 probability of scoring. Remember Samatta’s shot. There is no goalkeeper in the goal. If we call a hundred people from the street, chances are, ninety-nine of them will do it.
I’ve talked a lot about two opposite extremes. How about ”normal” shots?
Azpilicueta’s shot near the penalty area border has a 5% goal probability, while the xGOT value is 17%.
Mahrez’s chance is better than Azpilicueta’s one. The xG value is 0.08, but his shot didn’t generate excitement. According to OPTA’s model, it has 0.21 xGOT.
Semedo was one-on-one with the goalkeeper. He added value to his shot, but it wasn’t a great finishing.
In case you have noticed, the shots we just saw don’t have a touch that goes beyond standard finishing. However, their xGOT exceeds their xG. For %2 xG, %8 xGOT, that is, four times xGOT, is standard, while for %40 xG, %47 xGOT is standard.
Here you can direct me the following criticism: “The players you showed shot pretty well.”
The limit I set is completely subjective. It was determined by my perspective on football, completely with an eye test. We can also lower this limit.
Therefore, a faster increase is required in cases with less xG, and a slower increase in cases with higher xG in terms of xGOT for a standard finishing. In other words, what we need is logarithmic growth.
I used a logarithmic growth function to find the xGOT value representing the standard finishing for each xG. It produced exactly the result I wanted at first sight. Take Samatta’s goal, for example. There is an xGOT expectation of almost 95% for a shot with an 82% probability of scoring. In general, we are on the right track, but this formula also contains very solid problems. The growth between 1% and 10% is very fast. Remember Mahrez’s shot. The xG was 0.08, and we thought ~0.21 is enough for xGOT, but the function gives us approx 0.50 xGOT. We need to reduce these exaggerated results to their normal level.
The logarithmic growth is growing faster than we expected. We need to add a rule to the function such that values close to 0 should almost be divided by 2, and values close to 1 should remain almost the same. Thus, we both slow down the growth and prevent high probability shots from going beyond standard finishing.
I came across this page while surfing Google to overcome the problem. It seemed logical to try the s-shaped curve, and I wrote the function below.
s_shape_func = function(x) {2 - (1 / (1 + (x / (1-x))^-3))}
Here is the view of our curve showing our standard xGOT value for each xG.
It’s time to test our model. Our formula is clear, if the result is 1, it’s a shot with standard finishing.
Finishing = xGOT / ((0.5 * log10(xG * 100)) / s_shape_func(xG))Let’s go back to Samatta’s shot. The output is “1.024293”. Nice one! Shots with almost 70% or higher probability of goal are telling us, you don’t need to look for finishing here, if the player didn’t shoot the ball to the spectators, it’s standard finishing.
It’s good that the vast majority of goals scored are above standard finishing. I should point out that penalties are not included. Also, since the majority of the distribution is between 0 and 3, I decided to fix values greater than 3 at 3.
I want to place particular focus on blocked and missed shots. As I said at the beginning, if it is missed, the xGOT value becomes 0 reasonably. However, I believe this unwritten rule isn’t true of blocked ones.
Here is an example of a blocked shot. It doesn’t have the xGOT value because the ball was blocked before it went to the goal, but it’s kind of cruel to count this shot as 0 finishing. You know I set a standard xGOT for each xG. If the xGOT and xG numbers are the same, the finishing will be less than 1. I think I will avoid punishing the player too much when I make the xGOT of the blocked shots the same as the xG because if I do that, the result will be a value between 0 and 1. The same goes for shots that hit the woodwork. Even if those shots aren’t on target, they’re not examples of worthless finishing.
We overcame all obstacles. Let’s take a look at the EPL’s finishing table for this season.
According to “Shooting Goals Added”, Salah was the best in the league. Our model also says he has pretty good finishing. As the number of shots increased, the finishing score also increased faster than the average. Ronaldo, on the other hand, is a complete disappointment. It’s surprising that Bruyne also is below the line, but the finishing value of his goal against Chelsea reached the sky. Therefore, do your heart good.
Let’s not skip Salah right away.
We can easily say that the Egyptian player finished very well, especially shots with a high probability of scoring. It is no coincidence that most of his goals come from such chances.
When we look only at the shots within the penalty area, the result is not much different.
While Salah has more shot chances in terms of quantity, he also adds value to these chances. His partner on the other wing, Mane, remains below the line, but still closes to the average. Players below the line generally have lots of missed shots. For example, 42% of Benteke’s chances within the penalty area have gone wide. Of course, we don’t know under what conditions he hit the ball. A different scenario can be produced for such cases with a variable such as freeze frame.
By the way, speaking of this, I’ve also encountered some absurd results.
In this shot, the finishing value is very high, but it is not a situation caused by the model. According to Opta, the xG of the shot is 0.5 while its xGOT is 0.78. We can minimize such problems with the size of the sample.
Or a similar situation exists in this shot.
You watched the goal with the lowest finishing points among the shots that resulted in scoring. What’s wrong with it, you can say, but since the xGOT is a lower shot than the xG, this is how the finishing model works. There’s not much to do for the low finishing.
I don’t want to blame only the xGOT model for this absurdity. The finishing model isn’t perfect either. I’ve also encountered examples where the standard xGOT point we put for each xG is sometimes too much.
Mahrez displayed a very simple but very high-quality finishing example.
It would not be good to forget the shots from outside the penalty area. Of course, as I mentioned at the beginning, we can call it shooting, not finishing.
Raphinha scored 3 out of 5 goals outside the penalty area. He is the best in the league in terms of number of shots and manages to stay above the average line. Despite Harry Kane’s good numbers, his failure to score any goals is something that needs to be examined. If I go into more details, I will never be able to finish this blog. I haven’t analyzed the finishing in terms of the goalkeeper yet.
Final Word
The starting point of this whole issue was Samatta’s goal. That’s why I often emphazised his goal. It was obvious that there was a problem, and I tried to find a solution from my own perspective. Thus, I had an opportunity to return to my personal blog page again.
Thank you for reading.
New content coming soon!
*All data here belongs to OPTA via Fotmob
