Introducing my newly developed Player Impact Model
For a long time now I’ve been toying with the idea of trying to create a metric that can measure a player’s contribution in all phases of the game – and after many hours of creation and fine-tuning, I’m finally in a position to share Version 1 of my Player Impact Model.
What is the Player Impact Model?
Taking every single action captured in a match, using a similar framework seen already in Expected Threat (created by Karun Singh) & On-Ball Value (created by StatsBomb), a value can be assigned dependent on many different variables.
Without getting into too much depth, actions are valued based on the probability of a goal being scored either increasing or decreasing. For example, a successful cross from a wide area into the 6-yard box would be quite a high scoring action, but an unsuccessful cross would be a low negative value. It makes sense to reward a player highly for a successful cross to a dangerous area, and in turn it makes sense to not punish them too harshly for attempting to create a good opportunity for a teammate.
On the other hand, a successful pass from once centre back to another across their own box would be worth a negligible amount, but an incomplete pass in that area would come with a pretty big negative value as it will likely lead to a good opportunity for an opponent.
All actions captured for the match can be assigned a value, they’ll then be totalled up for each player, where then we can start to compare them in two different ways.
How can the Player Impact Model be used?
There are two slightly different uses for the model. Firstly, it can be used on a single game basis. In these instances, the total valued actions of each player will be added up, and then compared to all the players who have been on the pitch.
From this, we can create a relatively simple ‘match average’, and see which players are above and below that average level. Unless a player has had a real standout performance, in general you would see around half of the players on the pitch having a ‘Positive’ Player Impact, and around half having a ‘Negative’ Player Impact.
In this example using Nottingham Forest’s recent 3-1 defeat to Newcastle United, despite being on the losing side Murillo had a Player Impact Score of +0.68, ranking 4th of all players. The model does take into account goals conceded and penalises players for this, but despite conceding 3 goals Murillo still has a high Player Impact Score.
This was due to many things – Murillo scored the opening goal of the game, in my model that is obviously a very highly valued action. He was involved in lots of progressive passing actions, and more importantly in this example he had lots of successful defensive actions in and around his own box.
From the same match, we see a Player Impact Score of -0.36 for Morgan Gibbs-White. We can tell from his action map that there is not really much going on in terms of progressive or successful actions – and for that reason he has a low Player Impact Score.
The other way we can use the Player Impact Model, and perhaps the best and most useful way to do so, is by looking at it over a longer term such as a full season or a large set of matches. For this to work well, we have to change the model very slightly so that we compare players to their peers in each position, rather than the single match view where they are compared to all others on the pitch on that day.
In this example, we compare Ola Aina on the Pizza Plot to other right backs in the Premier League this season. He ranks in the 64th percentile of all eligible players for Player Impact, suggesting his performances would fit in well for a team looking to finish in Upper Mid Table.
The other very interesting thing we are able to do with the Player Impact Model, is look at previous seasons at that & similar levels. This can help us identify trends in a player’s performance & development – here we see that Ola Aina when compared to other Right Backs in the Top 5 European Leagues, more than holds his own this season. We also see that he joined Forest after a very decent season in Italy with Torino, but that he seemed to struggle slightly in his first season back in England in 2023/24.
It's important to remember that the model takes into account lots of advanced metrics (some publicly available, some developed by myself), and this allows us to really dig into a player’s performance.
It would be fair to say that Bruno Fernandes hasn’t had the output in terms of goals and assists this season that he, Manchester United or the recently departed Erik ten Hag would have hoped for. But according to my model his underlying numbers have been extremely good, and it would seem it would only be a matter of time until things turned around for him.
By turning the Player Impact Score into a percentile ranking for the purpose of this chart, we see the model ranks him as the best in the Premier League this season (for context he is just the slightest amount above Cole Palmer and would’ve been below him before last weekend’s matches).
But what this also tells us is two things – firstly he has still experienced a slight dip in Player Impact when compared to other Attacking Midfielders across Europe (his yellow line is down slightly on last year), and secondly there are quite a few Attacking Midfielders in Europe who have a Player Impact higher than he does – so there’s plenty of room for improvement for him as the season goes on under new manager Ruben Amorim.
Are there any flaws/limitations to the Player Impact Model?
Of course, like with pretty much every single metric there are always flaws and issues – no metric should ever be taken in isolation and should always be used to complement other metrics as well as what you see with your own eyes when assessing a player.
The main limitation is that the model can only take into account actions on the ball – for example a player making a really good run off the ball where they are not found/spotted by their teammate is not taken into consideration for this metric. Similarly, a player making a very good pressing decision to cover a potentially dangerous passing lane is also not included. This is purely because publicly available player tracking data simply doesn’t exist.
For that reason, there’s always going to be some players and examples where the Player Impact Model doesn’t seem to pass the eye test. As a Forest fan the most obvious example I can point out is that of Nikola Milenkovic.
The vast majority of Forest fans will tell you that Milenkovic’s addition has been instrumental to Forest’s significant defensive improvement this season, however my model only has him down as a playing at a level equivalent to the bottom half of the table. The reason for this is two-fold.
Milenkovic doesn’t actually engage in too many defensive actions; this can sort of penalise a player because he has done very little defending in his own box compared to other central defenders in the Premier League. That’s not a bad thing though, in fact it could be argued that the presence of Milenkovic at the back forces the opposition to find alternative ways to goal – this is a good thing but unfortunately it just can’t be accounted for in my model with the data that is available.
What are the plans for the Player Impact Model going forward?
As I gather more data I will continue to make tweaks to the model throughout the season to try and make it as accurate & insightful as possible. I also have plans to include & create some more metrics that can be used within the model that should hopefully improve it further.
Unfortunately, at this point in time the model doesn’t include goalkeepers. I’d like them to be included in the future but they are quite difficult for me to analyse in the same way with the way the data is currently structured.
I would also like to eventually incorporate it into using it not only as a tool to measure what has happened in the past, but also to try and measure and predict a player’s development in the future. There’s still a lot to be done to get to that stage but it would definitely add another string to the bow of the model.
Where can people get access to the Player Impact Model?
The Player Impact Model is currently exclusively available to players & agencies who work with the RDA Coaching & Performance Analysis. The team at RDA combine individual coaching for players with bespoke data analysis to help players improve and reach their potential. Get in contact with the team if you or someone you know would like more information!
How did you come up with the idea to create the model?
Since getting started with football analysis in the summer of 2023, over time I’ve become more comfortable in creating more advanced and interesting visualisations to help tell a better story about players through their data. An all-encompassing model like this was the next step in my journey and so far I’m pretty happy with the output from it.
I’d like to just shout out a few people who have been a great help over the past 18 months. Firstly, to Ben Griffis (@BeGriffis) who has been a huge help whenever I’ve bothered him with questions. Some of the stuff he creates and puts out is fantastic and really helps me in thinking about other things I can do & create. Liam Henshaw @HenshawAnalytics was also a huge inspiration with some of the stuff he put out around my club before he deservedly got his break in the game – he too has always been gracious with his time.
Rich & the guys at RDA Coaching have also been kind enough to bring me on board to help their players in their careers, and it’s also been fantastic to be invited to join the guys here at the Analysts Bar too.
If you’d like more information on the model, or on any of the things I produce you can contact me on Twitter (@WT_Analysis), BlueSky (https://bsky.app/profile/wtanalysis.bsky.social) or on e-mail (wt_analysis@outlook.com).
Comments