Expected Shots
Can’t there be a benchmark for the number of shots?
Let me ask you a question. Have you ever felt that your team was taking a few shots? Your team has the ball, the part of the heat map that shows the opponent’s half is red, there are a lot of passes left and right in the final third, they make relatively a lot of crosses, there are a lot of touches inside the penalty area, there are a lot of corners and set-pieces, but not what? No shots. There are, but only a few.
Shouldn’t there be a benchmark for this?
We had 20 shots! Wow, that’s a lot (is it?).
That’s exactly what I thought about and decided to do in this study.
When does a team shoot? As a football watcher, this has a lot to do with the distance to the opponent’s goal. Not only in terms of xG, but if we take a shot map from anywhere, it is possible to see that most of the shots are taken when inside the penalty area. This is not a surprise to anyone. So there must be a positive correlation between shooting and touch in some zones.
The more you touch the ball in dangerous zones, the more you shoot. At least we expect it to be so.
Or let’s think about the opponent. We pressed high up the pitch and didn’t allow them to build up the game from their own half, we indirectly increased our chances of taking a shot. We forced them to make a mistake and created a chance when we regained possession.
The opponent’s touches in their own third and our number of shots didn’t have much correlation, but the other one also has a message.
Of course, it is not enough just to go through the touch. It might help to focus on the type of actions. Passing is the most common action on the pitch. If a team makes a lot of passes into the final third or even into the penalty area, if they try a lot of through-balls, we expect our shot numbers to increase.
Life for me is like a classification model. The probability of an event happening is somewhere between 0 and 1. An infinite number of factors can influence the outcome. If we humans want something, we identify the variables that will affect the outcome more and work on them. Isn’t football like that? Possessing the ball and increasing the number of effective passing attempts increases the probability of reaching the goal, and you need to shoot to score.
Enough philosophy.
I don’t want to bore you by explaining each variable individually. In general, if we carry the ball into the danger zones, we directly increase the probability of shooting. Some factors indirectly affect this number. For example, dominating the middle third allows us to move the ball into the final third. Winning duels or aerial duels doesn’t necessarily cause us to shoot directly. It keeps the ball in our possession or allows us to take possession and eventually our target will be the opponent’s goal.
Of course, the opponent’s style of play also has the power to influence the number of shots. For example, if the opponent puts you in a difficult situation with counter-attacks, you may have to play a more cautious game.
I think I’ve made my point. Now I want to take one more step with an linear regression and xgboost models.
I used backward elimination to choose the features. First, I took all the variables, then I removed the variables with a high p-value for the Linear model and the variables with low importance for xGBoost. I evaluated the models I built on MSE and RMSE metrics. I found that the Linear model gave better results. You can find these codes on my GitHub page. I also put the training and test set there, in case anyone wants to try it for themselves.
I think we’ve come to my favorite part, the results!
I had La Liga as my test set and the most interesting result I noticed was the Barcelona vs Gradana match played on September 9, 2021. The match ended in a 1–1 draw. Barcelona took 17 shots. So why is this number strange for our model? Let’s look at the match statistics.
53 touches in the opponent’s penalty area, 53 progressive carries, 12 corner kicks, 8 completed crosses into penalty area, 76% possession, etc.
According to all these figures, Barcelona should have taken ~28 shots, but it is clear that the desired result has not been achieved.
I don’t know if you have this feeling, especially when Guardiola was coaching Barcelona, I used to get this feeling a lot in Barça’s games. They would pass and pass and never shoot. I think they were always waiting for the best chance. Maybe it was because I was just a child at the time.
How about a look at this season’s Premier League?
One of the biggest disappointments of the season, Chelsea took almost 50 shots less than expected according to the model. West Ham took almost 20 more shots than expected. Whether you have the ball or not has nothing to do with this difference. Bournemouth is bottom of the league in this respect, averaging 39.2% possession per game. West Ham or Nottingham Forest, despite having the same style, have had more chances.
This model only makes one point. Is taking too many shots good or bad? We can’t know that this way. Man City has the most shots in the league. They are also first in terms of expected shots, although the difference between the prediction and the actual value is negative. However, the quality of the shots we can’t know that way. For that, of course, we should have a look at xG.
What would the expected “Expected Goal” be like?
As I said at the beginning, as a football fan, there are teams that I sometimes find to have too few shots. For example, in last week’s match between Fenerbahce and Konyaspor, the home team played with a clear advantage because a player from the away team was sent off. The possession rate in the first half was 74.5% to 25.5% but Fenerbahçe only had 5 shots. How can it be possible? There was an obvious lack of creativity and this is one of Fenerbahce’s biggest problems this season.
So, just out of curiosity, I decided to take a look.
Finally, let’s look at the two craziest games in the EPL this season. Finally, let’s look at the two craziest games in the EPL this season. In both matches, there is a positive difference in the number of expected and actual shots.
Yes, that’s it.
I have been rolling up my sleeves for a bigger project. I’m obsessed with squad building. I believe that this time I’ll make a successful model test. I want to believe it!!! Wish me luck!
