Actionable analysis: Individual Header Rating (IHR) determines choices in blockers vs runners
In the new season 2024–2025 I want to try something new. I’ve been dabbling a lot in innovation (think of creating new metrics) and using analysis with existing data, but that’s not enough for me. It’s quite standard and I want to give more insight to what happens when you tailor your metrics whilst working for a club and make it actionable.
That’s the major issue with online content, it’s mostly present because it speaks to an audience — and I’m definitely part of that issue. I write not for myself only, but also in part because I know my audience would like the result of what I’m researching. I don’t think I can change that, but I can show you a little more about what the design of a metric means in the light of actual analysis within a club or organisation.
In this first installment I will focus on header ability. I want to create a metric that gives a rating to individual heading ability. With that rating that will change after each matchday, I can show probability of a player winning a duel. The actionable part here is that we can use that particular info when we are training and preparing for set pieces in the next game. IF we employ 3v3 runners vs blockers, we want the best winning probability in the air to maximise our delivery.
Why do we need a metric that measures ability
Football is a lot about tactics and avoiding trouble, but there are a lot of areas of the game where we have to focus on duels. It is a contact sport so we need to be aware that this will always be part of the game. To win critical battles on the pitch you need to make sure you can win little parts of it and that’s where aerial duels come into place. If you want to control the central areas of the pitch, you will have to deal with long goal kicks for example. This means there will be a battleground in those zones, so you need a strong force in the air.
I was trying to come up with ideas to look at aerial duels and came across this article about a metric designed by Statsbomb. Because, of course I was too late to actual invent a new idea.
It’s a very interesting concept and I wanted to recreate it, but also make it different. I want to look at aerial duels per 90, aerial duels won in % and height, but I also want to look further. I want to see if more metrics can be incorporated or see if different weights can be used for specific metrics. Recreating metrics is a great exercise for data analysis and data engineers, but more often than not you realise you are not completely satisfied with things that have been done before. That’s why I wanted to do it a bit different: I want to evaluate heading ability for individuals, with a link to probability and come with Power Ranking for Individual Set Piece Strength based on heading ability.
What do I need to make this happen?
There are several things I need to make this happen. First of all I need the full data of full season in a specific league or number of leagues. For this particular research, I’ve chosen the Eredivisie 2023–2024. This data can come from StatsBomb, Wyscout, Opta or any other data platform you are using. I’m using Wyscout data and specifically look at four specific metrics:
- Height: the influence of height on the aerial duels is not to be underestimated. There is a correlation and we will see later what that looks like.
- Aerial duels per 90: the number of aerial duels means something because it indicates how often you come in these duels and therefore your rating can constantly change.
- Aerial duels won in %: yes the number of duels is very important, but the win percentage tells us everything about quality. And, that’s where the real advantage will come from.
All the players will have played at least 900 minutes and will be excluded when they are goalkeepers. Their aerial duels are from a completely different calibre and need to be addressed in another research.
Methodology
There isn’t one specific methodology, because I want to make different things. I will be making a rating (and a score based on that rating), a probability and a power ranking based on that.
For the rating I will make sure there is a weighting for the metrics used. I will give a weight of 0,5 to the height, a weight of 1 to the numbers of duels and a weight of 1,5 to the win percentage of those duels.
After that I will use the glicko method to calculate the rating. I can also use ELO, but there is a reason I don’t. Glicko2 is more accurate in terms of predicting match outcomes or win probability and that’s essential for my process in doing this all.
Lastly, I will be making a power ranking and for all of these things I need to make a lot of calculations. These calculations will be made using Python.
Individual Header Rating (IHR)
By using glicko calculations I can calculate a rating based on the weights I’ve described above. By doing so I get a list of the players who score highest. For this example, I will take the top 15 performers according to this rating.
When we look at the table above we see the top 25 players and their rating. The rating is based on glicko2 method, which can be compared to an ELO rating, but again, the calculations are quite different.
When we want to look at an easier comparison for players instead of a ranking, we will convert them into a score from 0–100.
As you can see in the table above we have managed to show the top 10 players and their corresponding score. In this case we can then state that these players are most likely to win their aerial duels against other opposition.
Win Probability
We we will go into changing ratings and scores in a bit, but before we can get there, we need to talk about win probability. If we think a player is going to match up with a player during the length of the game, we can look at how likely it is that the a player will win.
I want to look at two players that might come across each other in a match. In the attacking side I’m going for Luuk de Jong (81,24) who is a threat in the air for PSV. He will play against Lutsharel Geertruida (83,85) of Feyenoord, who might come across to defend him and is very strong in the air too.
Using just the score, we can conclude that the win probability for Geertruida is 54,70% against 45,30% for De Jong. That matching up can be favorable for Feyenoord and PSV might want to think about a different match up. More on that later.
New ratings
So from that probability we move on to the new ratings and effectively making a power ranking. We will continue with these two players and look at the effect of the probability being the actual result.
Geertruida had the win probability and also won the aerial duels (absolute numbers in the majority) and that means his new rating is 2211,73 from 2210,45. De Jong lost and went from 2177,66 to 2176,38. In the grand scheme of things their ratings did change, but their place on the ranking did not. Geertruida stayed at place 33 while De Jong also stayed at place 40.
The place might stay the same, but over a whole season, things can fluctuate and become different. That’s how we can have an individual power ranking based on the Individual Header Rating.
Actionable analysis
Let’s pick Ajax for this analysis. They need to play against NEC in their next game and they need to know how well their own players are doing in the air and how NEC is doing as a whole.
In team performances for this metric, Ajax scores 7th and NEC scores 16th. From this you can conclude that Ajax is better in the air as a team that NEC is. That would lead to soft conclusion that this would also work to Ajax’s benefit in defending set pieces and attacking set pieces.
Now, let’s take a closer look at the individual players.
In the top 5, NEC has 3 defenders and Ajax 2. This means that in general we can see that NEC is stronger in the air with defending than Ajax. Ajax has more players overall who do well in the air in defence, but also in midfield and attack, while NEC isn’t that great in attack in comparison.
Suggestions could be that Ajax needs to focus more on the attacking and see solutions there, because the NEC defence is tough. Their attacking however isn’t as threatening, so the worry shouldn’t be there for Ajax.
The next step is how you match up in attacking set piece to make sure you create the upper hand. Let’s for a moment assume there are runners vs blockers of 2v2. If that would mean that Ross-Pereira would defend, Ajax would do well to get their highest rating players to go against them. That would lead to a higher probability.
Ross still has a higher winning probability, but it is much closer than when Ajax employ another player like Berghuis there;
If this were the case, according to our data, it would be very easy Ross to defend Berghuis. This would also lead to a difference in ratings after the game, but that’s purely for academic purposes.
Correlation between Height and Rating
Of course, there is a relation between the height of a player and the way it’s easier to win those duels in the air. You can see that in the scatterplot above. I don’t think it’s really weird to expect these results, but what can really help in this analysis is to look at outliers — they stand out and can lead to conclusions about someone’s ability to jump or go into an aerial duel. These outliers are marked in red.
Final thoughts
I had a lot of fun creating/writing this because this is something I use to determine my suggestions for set piece positioning to teams. Of course, data is just a part of the story and should always be backed with video to explain routines.
What I really wanted to show is that when you make metrics, there should be a practical use for it when you work in analysis with teams. It should add value to the process of the staff and the players. There are so many bugs, tweaks and turns I need to look at, but it’s an example of how making a metric can help in set piece analysis. Especially when most data metrics focus on delivery and shooting.
