I have been writing on soccer analytics for over a decade now, and one of the most difficult problems in the field has not really changed. Everyone wants your opinions of transfers, but there are few sound bases from which to begin analysis. There are no “wins above replacement” statistics in the sport with solid grounding. It is easy, rhetorically, to take a player and point to one statistical output or another and argue that it means the player is worth a certain fee. But it does not offer much room for debate or discussion: what if someone else points to a different statistical measure and says it means the fee was too high? What can you use to adjudicate differences and back up your opinion?
“Wins above replacement” does not answer every question under the sun, but such statistics are very helpful for establishing a common ground for discussion. I have struggled with writing articles about transfers because there is always this crucial gap between the thing I want to argue—”this player is worth the fee”—and the thing I actually can demonstrate—”this player has 98th percentile ball progression by pass and carry”.
I am not introducing WAR for football here. I wish I were. That would be cool. But I have developed some methods that I think can be useful for creating a different kind of grounding for discussions in player analytics.
The Comps System: Remembering Some Guys as Analytics
The player comparison system is not a new thing at Expecting Goals. I introduced the comps system in the Elliot Anderson newsletter, and I used it in the Morgan Rogers newsletter as well. But I wanted to give it a more thorough introduction here because I think it offers a way to ground these discussions in something more than just figures.
I cannot argue, for example, that Elliot Anderson’s progressive passing and possession involvement numbers created a certain number of goals for Nottingham Forest and so his transfer fee is fair given the price of goals. But I can say that the players with the most similar statistical production to Elliot Anderson’s last season at Nottingham Forest were Jorginho, Granit Xhaka, Hakan Çalhanoğlu, Toni Kroos and Marco Verratti.
With a method for identifying similar players and seasons, I can translate Anderson’s statistical profile from a set of numbers to a set of names. This comps list raises questions like, what would you pay for a young Jorginho, a young Kroos or a young Verratti? It creates a historical context for discussion. And if the comps list seems wrong, it is possible to isolate the statistical indicators that make the difference and ask whether they are misleading in some way.
The comps method may not be a breakthrough in player valuation, but it is enough to make me feel comfortable writing a long series of transfer articles.
Leon Goretzka and How the Comps System Works
Aston Villa signed Leon Goretzka at the end of August, and the move came without a transfer fee because Goretzka’s Bayern Munich contract expired in the summer. The contract terms and length have not been precisely reported.1 Goretzka has not generally been treated as one of Villa’s most important signings, but the comps system suggests he may be their most important signing for the 26-27 season.
I will walk through how this was built.
First, the system takes a large set of player statistics. For each statistic, it begins by taking a combination of the player’s raw per-90 numbers and the player’s percentage of team production in that statistic while on the pitch. Goretzka, who has been playing in one of the most extreme possession-oriented systems in world football, would show up as an absurdly elite possession player without this adjustment. As a raw number, Goretzka appears far above the league median in touches.
But as a share of his team’s on-ball touches, Goretzka appears only slightly above average.
Notably his teammate Joshua Kimmich took a far larger percentage of Bayern’s touches when he played midfield, and appears as a league leader on both axes unlike Goretzka. A chart of the full list of qualifying seasons shows how these numbers work across a larger population.
Both of Goretzka’s qualifying seasons show up as notably stronger in raw rate than team share, and I have highlighted a few of the more extreme examples. Kimmich appears here again, with otherworldly involvement numbers that are only partly accounted for by playing in an extreme possession system. The 2012-13 version of Sergio Busquets is a fun one, because that was the season Busquets split time in midfield with Xavi, Thiago Alcântara, and Andrés Iniesta. While over time Busquets would become one of the best passing midfielders in the world, on that team it was other players who had more of the ball while Busquets handled more defensive responsibilities. On the other side, I have highlighted Pascal Groß’ season at Ingolstadt, before his move to Brighton, where he was the entire ball progression plan for that team, as well as one of Mark Noble’s seasons at West Ham.
The 50/50 split between team share and raw rate was chosen primarily because, for most of the statistics in question, it performed a little better than pure raw rate or pure share at projecting future output.2 I think these examples are useful for demonstrating why. You would not want to project Mark Noble either as an average possession midfielder because of his raw numbers, or as the equal of Joshua Kimmich from 24-25 because of his team share numbers. Somewhere in the middle makes more sense.
These numbers are then combined into, for midfielders, five axes. The first four are reasonably straightforward: involvement in possession, ball progression by pass or carry, ball-winning, and shot contribution. The specific statistics which make up each axis are listed in the chart below. The fifth, the lowest-weighted axis, is “usage”, distinguishing players who had a substantial percentage of their minutes and appearances as substitutes rather than as starters.
Players are compared based on the difference in their scores on every statistic, as well as in every aggregate axis. These two scores, between statistics and between axes, and added together to create a comps score for a player’s statistical record in one or more seasons. For Goretzka, I took the aggregate of his 2023-24, 24-25 and 25-26 seasons. The player with three consecutive seasons most similar to Goretzka was Ivan Rakitić in his final seasons at Barcelona.
Fabián Ruiz in his Napoli seasons appears as similar to Goretzka in most of his output statistics, but he was a starter not a rotation player. Adrien Rabiot earlier in his career at PSG would have been similar if he had shot output, but at that point it was not part of Rabiot’s game. (He also had a bit more defensive contribution than Goretzka).
These calculations all come together to create a full comps list for Leon Goretzka, based on his statistical profile over the last three seasons.
Rakitić at Barcelona scores as three of the top four comparables. Obviously there is some overlap between the seasons here, but it still demonstrates this is the player type. A more recent set of Fabián Ruiz seasons, as he became more of a rotation player at PSG, also joins the group. These players are mostly around 30 years old, good all-around midfielders on elite teams, with an unexpected amount of shot production, who cannot or are not asked to carry a full-season minutes load anymore.
If this is what Aston Villa gets out of Goretzka, this could be one of the most important short-term signings of the summer. While Goretzka cannot replace all the minutes that Youri Tielemans played, he has been contributing at a level on the ball that is reminiscent of Tielemans. The risk of injury or collapse is real, but Goretzka’s recent production level has remained at a very high level across all the on-ball categories, comparable to many star midfielders in their decline seasons.
There is one other aspect of the comps system to introduce, by means of another Aston Villa signing, and then I can finally get around to talking about two of Arsenal’s signings.
Johan Manzambi and Comp Distance
Villa signed Manzambi from Freiburg this summer for a reported fee of just over £50 million guaranteed and just under £10 million in add-ons. Manzambi has shown obvious skills at Freiburg, as well as at the World Cup for Switzerland, but an initial perusal of his comparables list is somewhat baffling. Among his top comparable seasons are Rodrigo de Paul at Udinese, Naby Keïta at RB Leipzig, and Jude Bellingham at Borussia Dortmund. None of these are terribly similar players. What they have in common is they all did an unusually large amount of progressive ball-carrying for a central midfielder. And this helps get at the problem. Manzambi’s ball-carrying statistics are so unusual for a CM that he has almost no real comparables at all. Keïta in 17-18, the closest comparable season, has an overall score of 0.68, which would not place among the 30 most similar seasons on Goretzka’s list. The system identifies a few players who were ball-carrying outliers at central midfield, and they are the best it can do, but none of them is truly similar to Manzambi.
Comparability is both a ranking and an overall score. For Manzambi as a central midfielder, the overall score is the most useful. He has zero players with a 0.5 comp score or below, and only a few with a score of 1.0 or below.
The main fact the system identifies about Manzambi, then, is not that his statistical production resembles this player or that player, but that exceptionally few players have played anything like him.
There is a further oddity about Manzambi’s numbers that can maybe provide a few reasonable analogs. While Manzambi primarily played CM for Freiburg, his statistical output involved a large amount of ball-carrying and a high take-on rate, relatively few touches in possession and only moderate defensive work, and significant amount of shot production. In other words, he has the statistics of a winger.
If Manzambi’s numbers are run through the attacker template instead of the CM template, the list of similar players seems much more reasonable.
