Two sides of a sphere: The importance of attack and defence in different positions

Introduction

At its beginning, association football was only barely an organised sport at all. In the FA’s first set of laws, there were no stipulations as to the number of players, or how long they were expected to play for. There were prohibitions against holding, hacking, tripping, pushing and certain kinds of footwear; but there were no fixed penalties for breaking these laws and no match officials to enforce them. The only mention of a free kick was in Law 8, which gave the right to claim one to any player who made a mark in the ground with his heel immediately after catching the ball. Such marks were the only ones a pitch could be expected to have besides a set of flags around its edges. The prevailing tactics were as rudimentary as the rules. The man with the ball would simply charge towards his opponents’ goal with his colleagues backing him up, ready to take over if someone knocked the ball loose. The notion of players having set positions was at first as unheard of as it had been in the whole-village scrambles from which the game had descended; but as the game developed, it became clear that a teams stood a better chance of winning if it left a few players behind the line of chargers to concentrate on guarding their own goal. The specialist position of goalkeeper was officially recognised in 1871; and by the time the first official international match was played a year later, the English and Scottish teams had both realised that he could do with some protection.

The question of how much protection one could afford to give him, and what form that protection should take, was investigated by trial and error. Organised formations were on their way; and after a few years, the classical formation had become established everywhere as the best way to balance defence and attack. The Cambridge Pyramid of two full-backs, three half-backs and five forwards remained ubiquitous for forty years and formed the basis of thinking about football’s tactics and strategy for several more. It survives as a common attacking formation even today, although hidden behind a defensive base such as 4–4–2 or 4–3–3. Today, players are usually described in terms of the positions they occupy while cloaked in their defensive formation, and those positions themselves are often verbally identified with specific roles. Instead of full-backs, half-backs and forwards; many of today’s sportswriters, radio and television commentators and analysts talk of defenders, midfield players and attackers. (If the classical terms are used, they are often used without much thought as to what they were intended to mean, but that’s a different story.)

But how much truth is there in these descriptions? It makes intuitive sense that a player whose usual position is close to his own goal should be concerned primarily with defending that goal, and that a man who spends his playing time hanging around in the opposition’s half should be focused on attack. On the other hand, everyone who plays or watches football knows that this isn’t the whole truth. When a team has the ball, even the goalkeeper has a part to play in building up an attack; and that when it loses the ball, even the centre-forward can contribute to the effort to win it back. Fifty years ago, when the modern terms for playing positions were beginning to spread, Bill Nicholson was complaining that this sort of language was a gross oversimplification which restricted tactical thinking even among players, and I can think of Premier League players today who seem to illustrate his point. But what do the numbers say? How important is each position in a team to attack, and how important is each position to defence? We all weigh the two against one another whenever we make assessments of player’s contributions; but most of the time, we have nothing to base our weightings on but gut feeling. Here, I’m going to have a go at being a bit more precise.

Method and Data

How is this to be done, you may be asking? What statistics can one use? If you’re a trained number cruncher, you may have some ideas already. If you’re more of a football person; or if, as in the case of some of my readers, you’re neither; you might be tempted to throw your hands in the air and give up already. For those who haven’t studied statistics before, the question of how important one variable is in determining another seems daunting, perhaps even unanswerable. But I promise you that our quest is not as hopeless as it may seem.

The key concept that needs to be understood here is that of the variance, which is exactly what it sounds like — a measure of the degree to which a given variable varies across a population. The variance is calculated using the equation:

(1) V(Y) = Σ [Yi— E(Y)]2 / N

The variable of interest is Y and individual values of Y are shown as Yi. The arithmetic mean, or average, of these values is E(Y), where E is the expectations operator.

(2) E(Y) = Σ Yi / N

The Greek letter Σ is used as a summation symbol, meaning that all the values of [Yi — E(Y)]2 are added together. N is the number of observations in the population. Where the entire population cannot be observed, a sample of size n must be used instead. Applying the summation operator only to those observed values of Y within the sample and correcting for sample bias gives the sample variance, v(Y):

(3) v(Y) = [(n / (n-1)] * Σ [Yi — E(yi)]2 / n = Σ [Yi — E(yi)]2 / (n-1)

Taking observations of Y under different conditions and calculating the variance tells us something about the importance of those conditions. The more important the conditions, the greater the variance. But what is Y? We are looking for a measure of a player’s contribution to his team’s offensive and defensive performance. If there were a statistic that truly captured and quantified every player’s offensive and defensive contributions to his team’s results, one could simply break that statistic down into its offensive and defensive components for every individual player in every individual game and calculate the variance of offensive and defensive contributions at each position within a certain formation. Unfortunately, no such statistic exists. Any attempt to create such a statistic is almost certain to end up over-valuing attacking play relative to defensive play, for the simple reason that things which do happen are easier to notice and quantify than things which don’t happen. However, we can do something.

We cannot directly see how much any player contributes to his team’s performance, but what we can directly see is how many goals it scores and concedes when he is on the pitch. Giving him credit for this creates an Offensive and a Defensive Personal Goal Difference for each player.

4) Oi = Gf, t, i * 90 / Mi — L

(5) Di = L — Ga, t, i * 90 / Mi

In these equations, Oi and Di are the Offensive and Defensive Personal Goal Differences for a player i. Gf, t, i and Ga, t, i are the number of goals his team, t, scores and concedes respectively when he is playing. Mi is the number of minutes a player spends on the pitch and L is the average number of goals per team per game in the competition being studied. Oi and Di are therefore the number of goals per game better than average team t is, offensively and defensively, when player i is active. Combining the two gives a player’s total personal goal difference per game.

As rating systems for individual players, personal goal difference and its components leave much to be desired. A relatively good player may be stuck in a relatively bad team and end up with lower personal ratings than his individual contributions warrant. A relatively bad player may be pulled up by his relatively good colleagues and end up with higher ratings than his own poor performances merit. But if one were to control for the competence of a player’s colleagues by comparing his results to his team’s average rather than to the league average, one would not necessarily improve the accuracy of one’s measurements. A second-string Liverpool player, whose team does better with his first-string colleague playing in his place, may still be more productive than a first-string Watford player whose team can’t do without him. Ideally, one would want to at least control for the quality of the opposition. That second-string Liverpool player may look better than he is because his first-team appearances tend to come in games against weaker opponents, when his manager is giving the first-string player at his position a rest. I could account for this by using the opposing team’s averages as a comparator; but, as hard as it may be for some of you to believe, I do in fact have a life. Generating opposition-neutral versions of Oi and Di would require me to meticulously collate game-by-game data for every player I wanted to include in my study; and until and unless someone pays me to do so, that’s not going to happen.

Anyway, rating individual players is not our concern here. We are trying to investigate from scratch the importance of attack and defence at different positions; and the metrics we are using are good enough for our purposes. Good attacking teams, by definition, score a lot of goals; and good defensive teams concede few goals. It therefore makes sense that good attacking players will tend to play in teams that score a lot of goals and good defensive players to play in teams that don’t concede many. As long as that’s true, the variance in personal goal averages at a position should be a pretty good indicator of the variation in personal contribution to attack and defence at the same position. If the identity of a team’s centre-forward makes more of a difference to the number of goals it scores than to the number of goals it concedes, the variance of Oi among centre-forwards can be expected to be greater than the variance of Di within the same group, even if the individual values of Oi and Di are far from comprehensive measures of individual contributions.

So far, so good. But before we plug the values of Oi and Di into Equation 3, we have one last adjustment to make. Some players play more than others, and their statistics carry more weight than others in the calculation of a competition average. They also carry correspondingly more weight when one is calculating the variance, the average squared deviation from that average. The weighted sample variances of Oi and Di for a given position are therefore:

(6) Vw (Oi, p) = [np / (np — 1)] * Σ (Mi, p [Oi, p — Ew(Oi, p)]2) / Σ Mi, p

(7) Vw (Di, p) = [np / (np — 1)] * Σ (Mi, p [Di, p — Ew(Di, p)]2) / Σ Mi, p

The subscript p indicates that we are focusing on a particular playing position, while w indicates that the averages and variances of Oi and Di are weighted in accordance with how much a particular player played at that position.

It may have occurred to you that, holding the playing positions constant, the weighted averages of Oi, p and Di, p are by definition zero across any sample of matches; and that Σ Mi, p will be twice the number of matches multiplied by 90. You may be wondering why I have written the latter the way I have, and why I bothered to include the former in my equations at all. The answer is that there are two complications, one concerning the data available to me and one concerning the nature of the game itself.

Firstly, as I stated earlier, my data are imperfect and my time limited. For this study, the most accurate and easily accessible data I could find were on fbref.com. This website tells me which positions a player tends to play, and how his team did while he was playing in a particular competition in a particular season. However, it does not tell me how well it did while he was playing a particular position. It can tell me how many goals Manchester United scored and conceded when Paul Pogba was playing; but the records of Paul Pogba the inside-forward, Paul Pogba the outside-forward and Paul Pogba the half-back are not separated. Given this limitation, I have decided to include only those players who made a majority of their appearances at a single position, with all of a player’s playing time counted as having been spent at that position. The averages and variances of Oi, p and Di, p derived from the sample are therefore rough estimates rather than precise calculations. For clarity’s sake, a list of players will be provided at each position in the appendix. Because players are initially classified by team in the dataset, a player who represented two clubs during the same season is counted as two separate players.

The second problem is that today’s formations are too varied for positions to be usefully held constant across a large set of matches involving a large number of teams. One can, if one likes, map the classical formation onto every modern formation and classify players accordingly. However, to do so without acknowledging the differences in formations would be to obscure the fact that different formations may require different types of players at equivalent positions. A player who would be an aggressive, overlapping wing-half in a W-W formation might be a returning wing-forward in an M-M, and an inside-forward in the former might play as a deep centre-forward in the latter. Once one allows for differences in formations in one’s classification of positions, the uniformity of average goal difference and minutes played across positions disappears even with perfect data. Where opposing teams can field a different number of players at the same position, there is no guarantee that every formation will be used with equal frequency or equal in the results its adherents achieve. The mathematical guarantee as to the values of Ew(Oi, p), Ew(Di, p) and Σ Mi, p thus disappears at every position except goal.

How are the playing positions to be classified? One could, if one wanted to, say that every team’s formation is different from every other team’s, its practical application depending as it does on the playing styles of the individuals within it; and that even the same group of players will vary slightly in its structure from one game to the next according to the opposition. This would be true enough; but for the purposes of this study, it would also be useless. Perhaps no two players play precisely the same position. Perhaps no one player plays precisely the same position in any two games. But in order to discover general truths about playing positions, one must find a way to group players who occupy similar positions. How these positions are to be defined is a matter of judgement. To be too narrow in one’s definition is to shrink one’s sample for each position to the point of statistical irrelevance; to be too broad is to increase statistical significance at the expense of sporting significance.

I’d like to study every formation in use, coding every player’s position for the formation in which he plays; but once again, time constrains me. I’m probably already spending more time on this project than I should be; and the more formations I study, the more data I will need to collect and sort. To keep my task manageable, I have decided to base my positional groupings on those used by whoscored.com. This website divides outfield positions into a grid of 5 rows and three columns, similar to those used in the great Irish sports of Gaelic football and hurling. The rows, running from back to front, are listed below.

· Defenders

· Defensive Midfield

· Midfield

· Attacking Midfield

· Forwards

The columns represent the right, left and centre of the pitch. In a row of four or five players, the left and right columns are reserved for those who play the extreme wide positions. Players occupying a row on their own are assumed to be central. Players occupying a row of two, I have observed, are usually but not always assumed to be central. The wide positions in a row of three are usually counted as central in the three deepest rows and as right and left in the two more advanced rows.

If you think this is inconsistent, you are not alone. This is not the way I would have chosen to classify playing positions, either. My preference would be either to use the classical positions as outlined above; or to use a 5*5 grid, including the inside channels, instead of a 5*3, and classify several players in different rows. Nonetheless, I have to work with what I have. For the purposes of this study, a player’s position is the position in which, according to the squad list available on whoscored.com, he played a majority of his games during the observation period.

The observation period is the 2021–2022 English Premier League season, the most recently completed European season at the time this project started. The data are taken from the English Premier League. For every player who appeared for any club in this competition during the season of interest, team performance data covering the competition and season are imported. For every player who can be classified by position according to the aforementioned criterion, values of Oi and Di are calculated using Equations 4 and 5. Equations 6 and 7 are applied to each positional group of players to compute Vw (Oi, p) and Vw (Di, p). Once Vw (Oi, p) and Vw (Di, p) have been calculated for a given position, one can easily compute attack-defence ratios expressing the relative importance of offensive and defensive contributions at that position.

(8) Fo, p = Vw (Oi, p) / [Vw (Oi, p) + Vw (Di, p)]

(9) Fd, p = Vw (Di, p) / [Vw (Oi, p) + Vw (Di, p)]

(10) Rp = (100*Fo, p):(100*Fd, p)

Equations 8 and 9 express the relative importance of attack and defence at a given position as fractions. Fo, p is the weighted variance of Oi at a given position divided by the sum of the weighted variances of Oi and Di. Fd, p is the weighted variance of Di. divided by the same denominator. It follows that Fo, p and Fd, p by definition add up to 1. Rp is the ratio between the two, scaled so that the sum is 100.

Results and Conclusion

The table below summarises the results of my study for every position, with all decimal numbers rounded to two decimal places.

Table 1: Summary of results

Immediately, one can see several results that would almost certainly surprise a football novice, and some that may surprise several experts.

A) The identity of a team’s goalkeeper (GK) appears to be slightly more important to its offensive output than its defensive security, the effects of constructive play from the back varying more than the effects of shot-stopping and defensive organisation. This, to me, is probably the most surprising result of all. I expected constructive play from goalkeepers to be more important than is commonly assumed, but not that important.

B) The same holds true for those players classified as outfield defenders (DR, DC, DL), whose offensive contributions are more likely to include advancing to join the forward line during play and heading goals in dead-ball situations.

C) Wide midfield players in 3–4–3 or 3–5–2 formations, assumed by whoscored to be defensive players, appear by Rp to be some of the most offensively oriented players of all. In this dataset, left-sided defensive midfield players (DML) have the highest Rp of all and their right-sided equivalents (DMR) the third-highest.

D) Further forward, the opposite seems to happen. All three so-called “attacking midfield” positions (AMR, AMC, AML) show a greater variance in team defensive performance than in offensive performance; and for centre-forwards (FC), the variances are almost exactly equal.

E) Central defensive midfield players (DMC) appear not to greatly affect their team’s performances in attack or defence. This is the only one of the so-called defensive positions in which defensive performance varied more than offensive performance, but neither varied a great deal with personnel at this position. Whereas those central midfield players not classified as defensive or offensive appear to contribute a great deal to both, this category of player having the third-highest offensive variance and the fourth-highest defensive variance; those classified as defensive have the fourth-lowest offensive variance and the joint-fourth-lowest defensive variance. Given the centrality of this position within a team, and the way in which several coaches and analysts make a point of highlighting their contributions, this surprised me almost as much as the results for goalkeepers.

However, these findings are not to be accepted uncritically. Before I can claim to have revolutionised football’s understanding of player roles, numerous caveats must be considered. As I have previously acknowledged, the small sample size and the incompleteness of the data make it unwise to draw any firm conclusions. Data taken from a single season of a single competition, and aggregated in the clumsy manner I have described, may be suggestive but they are far from conclusive. The small-sample problem is especially obvious when one considers the wider and more advanced positions within a team. A sample of 89 centre-backs (DC) is one thing, a sample of only 5 left-sided forwards (FL) another.

The role of the averages at each position should not go unnoticed. For those positions which have a large sample, the weighted averages of Oi and Di are close to zero, the deviations in the case of goalkeepers being visible only at the 5th decimal place. However, for some positions, Ew(Oi, p) and Ew(Di, p) differ markedly from zero; and this may be an effect not only of imperfect data and variation in formations, as described above, but also of the changes in a team’s style of play that goes with a change in formation. Even if central defenders vary more in their offensive contribution than in their defensive contribution, the fact that the average defensive goal difference for central defenders is positive and exceeds the average offensive goal difference indicates that, while such players may vary more in their offensive contributions than in their defensive contributions, playing with more of them nonetheless tends to make a team more defensive. The fact that Ew(Di, p) is even greater for wide defensive midfield players and negative for wide defenders supports this hypothesis further, the former typically being present and the latter absent in teams which opt for a third central defender. Similarly, playing with a designated attacking central midfield player (AMC) seems to make a team more offensive, with an average offensive goal difference of 0.28 and an average defensive goal difference of -0.18, even if the variation in the contributions made by such players is primarily defensive.

The effect of playing style may be related to that of the way the data are classified. In the whoscored data, a trio of central midfield players is usually split into defensive and attacking players when there are two of the former and one of the latter but not when one of the three holds his position behind the other two. This may explain to an extent the relatively high variation in offensive and defensive goal difference among “MC” players and the relatively low variation in both among “DMC.” Perhaps players classified as pure midfield players vary so much in offensive and defensive goal difference precisely because the category includes so many holding players and so many advanced players. Perhaps players classified as DMC vary so little in their contributions because playing with twin pivots reduces the effect that either can have on a team’s performances individually.

Something similar may be observed in the wide players. Four of the ten players classified as right-sided forwards and three of the five left-sided forwards spent the season on interest playing for Liverpool or Manchester City, the top two teams in the division. Meanwhile, all of the wide midfield players not classified as attacking or defensive players came from teams in the bottom six of the league table. Part of this is down to these teams’ formation and styles of play, and part may be attributable to notation.

Liverpool spent the entire season playing 4–3–3, and Manchester City usually used the same shape. Burnley, Everton and Southampton (the teams that supplied most of the players classified as ML and MR) usually used a 4–4–2 formation. It is therefore not surprising that the average goal differences, offensive and defensive, should be positive for advanced wingers and negative for withdrawn ones. Much of this may be attributable to the relative strength of the teams in which these men played, the samples being insufficiently large and insufficiently varied for team effects to cancel each other out. The slightly negative averages for centre forwards is probably attributable to the same cause: within the sample, teams that played with two central strikers instead of just one tended to be worse than average. Whether they were worse than average because they used a 4–4–2 formation, or whether they played a 4–4–2 formation because they had worse players, or whether the two are just coincidental, is not within the scope of this study.

The strength of a team may also play a part in how its players are classified. If a team is playing well, its wingers will have plenty of chances to advance. If it is being dominated in terms of possession and territory, those chances will be less frequent. Therefore, what is called a 4–3–3 formation when practiced by a good team may look more like a 4–5–1 when used by a bad team. I do not think this effect is especially large in the case of this dataset, but it is something to consider.

With the above caveats, it would be premature to call the results of this study conclusive. They are, however, suggestive, although wider and deeper research would be required to confirm or refute the suggestions. They suggest that while the common descriptions of players as defenders and attackers are not entirely false, they are not entirely true either. Deploying more centre-backs may make a team more defensive in style, and using one or two designated attacking midfielders may make it more aggressive; but it appears that once a formation has been selected, the competence of most of the players within that formation will vary the most in those aspects of football not reflected in his job title. What separates a good centre-back from a bad one is less his ability to head the ball out of the penalty area than his ability to build up attacks from the rear. A manager who wants a more attacking style of play may be well advised to use an advanced central midfield player; but in looking for that player, he should pay close attention to what that player does defensively. To those who know nothing about football, such ideas may seem crazy; but to most of those who know the game well, I expect that they will make a counterintuitive kind of sense.

References

https://www.whoscored.com/

https://fbref.com/en/

Appendix: List of Players by position

GK

Ãlvaro Fernandez, Aaron Ramsdale, Alex McCarthy, Alisson, Alphonse Areola, Angus Gunn, Asmir Begovic, Ben Foster, Bernd Leno, Caoimhín Kelleher, Daniel Bachmann, Danny Ward, David de Gea, David Raya, Ederson, Edouard Mendy, Emiliano Martinez, Fraser Forster, Freddie Woodman, Hugo Lloris, Illan Meslier, Jack Butland, Jason Steele, Jed Steer, John Ruddy, Jonas Loessl, Jordan Pickford, José Sá, Karl Darlow, Kasper Schmeichel, Kepa Arrizabalaga, Kristoffer Klaesson, Łukasz Fabiański, Martin Dúbravka, Nick Pope, Robert Sanchez, Robin Olsen, Tim Krul, Vicente Guaita, Wayne Hennessey, Willy Caballero, Zack Steffen

DR

Aaron Wan-Bissaka, Ben Johnson, Cedric Soares, Connor Roberts, Diogo Dalot, James Justin, Jamie Shackleton, Jeremy Ngakia, Joe Gomez, Joel Ward, Kiko Femenía, Kyle Walker, Matthew Lowton, Matty Cash, Max Aarons, Nathaniel Clyne, Ricardo Pereira, Ryan Fredericks, Seamus Coleman, Sam Byram, Takehiro Tomiyasu, Trent Alexander-Arnold, Valentino Livramento, Vladimír Coufal

DC

Gabriel Dos Santos, Ben White, Rob Holding, Pablo Mari,­ Sead Kolasinac, Tyrone Mings, Ezri Konsa, Calum Chambers, Axel Tuanzebe, Kortney Hause, Pontus Jansson, Ethan Pinnock, Kristoffer Ajer, Mads Bech Sorensen, Mathias Jorgensen, Charlie Goode, Lewis Dunk, Adam Webster, Shane Duffy, Dan Burn, James Tarkowski, Ben Mee, Nathan Collins, Kevin Long, Antonio Ruediger, Thiago Silva, Andreas Christensen, Trevoh Chalobah, Marc Guehi, Joachim Andersen, James Tomkins, Michael Keane, Mason Holgate, Ben Godfrey, Yerry Mina, Jarrad Branthwaite, Diego Llorente, Pascal Struijk, Liam Cooper, Çağlar Söyüncü, Daniel Amartey, Jonny Evans, Wesley Fofana, Jannik Vestergaard, Virgil van Dijk, Joël Matip, Ibrahima Konaté, Aymeric Laporte, Rúben Dias, John Stones, Harry Maguire, Victor Lindelöf, Raphaël Varane, Eric Bailly, Phil Jones, Fabian Schär, Jamaal Lascelles, Dan Burn, Ciaran Clark, Federico Fernández, Grant Hanley, Ben Gibson, Ozan Kabak, Andrew Omobamidele, Christoph Zimmermann, Mohammed Salisu, Jan Bednarek, Lyanco, Jack Stephens, Yan Valery, Eric Dier, Ben Davies, Cristian Romero, Davinson Sánchez, Craig Cathcart, Samir Santos, William Troost-Ekong, Christian Kabasele, Francisco Sierralta, Nicolas Nkoulou, Craig Dawson, Kurt Zouma, Angelo Ogbonna, Issa Diop, Conor Coady, Romain Saïss, Max Kilman, Willy Boly, Toti Gomes

DL

Kieran Tierney, Nuno Tavares, Matt Targett, Lucas Digne, Marc Cucurella, Charlie Taylor, Erik Pieters, Tyrick Mitchell, Lucas Digne, Vitaliy Mykolenko, Junior Firpo, Luke Thomas, Ryan Bertrand, Andrew Robertson, Kostas Tsimikas, João Cancelo, Oleksandr Zinchenko, Benjamin Mendy, Luke Shaw, Alex Telles, Matt Targett, Matt Ritchie, Jamal Lewis, Paul Dummett, Brandon Williams, Dimitris Giannoulis, Kyle Walker-Peters, Romain Perraud, Hassane Kamara, Adam Masina, Danny Rose, Aaron Cresswell

DMR

Chiquinho, Emerson, Jonny Castro, Ki-Jana Hoever, Mads Roerslev, Matt Doherty, Nélson Semedo, Reece James, Sergi Canos

DMC

Adam Forshaw, Albert Sambi Lokonga, Declan Rice, Fred, Granit Xhaka, Mark Noble, Mohamed Elneny, Nampalys Mendy, Nemanja Matić, Oghenekaro Etebo, Robin Koch, Scott McTominay, Steven Alzate, Thomas Partey, Tim Iroegbunam, Tomáš Souček

DML

Ben Chilwell, Dominic Thompson, Fernando Marçal, Kenedy, Marcos Alonso, Rayan Aït Nouri, Rico Henry, Ryan Sessegnon, Sergio Reguilón

MR

Aaron Lennon, Johann Berg Gudmundsson, Andros Townsend, Tony Springett, Stuart Armstrong, Theo Walcott

MC

Abdoulaye Doucoure, Adam Lallana, Alex Oxlade-Chamberlain, Allan, Andre Gomes, Ashley Westwood, Bernardo Silva, Billy Gilmour, Boubakary Soumaré, Bruno Guimarães, Cheikhou Kouyaté, Christian Eriksen, Christian Norgaard, Conor Gallagher, Curtis Jones, Dale Stephens, Dan Gosling, Dele Alli, Donny van de Beek, Douglas Luiz, Eberechi Eze, Edo Kayembe, Enock Mwepu, Fabian Delph, Fabinho, Frank Onyeka, Harry Winks, Harvey Elliott, Ibrahima Diallo, İlkay Gündoğan, Imran Louza, Isaac Hayden, Jairo Riedewald, Jack Cork, Jacob Ramsey, James McArthur, James Milner, James Ward-Prowse, Jean-Philippe Gbamin, Jeffrey Schlupp, João Moutinho, Joe Willock, Joelinton, John McGinn, Jonjo Shelvey, Jordan Henderson, Jorginho, Josh Brownhill, Juraj Kucka, Kenny McLean, Kevin De Bruyne, Leander Dendoncker, Luka Milivojevic, Lukas Rupp, Luke Cundle, Marvelous Nakamba, Mateo Kovacic, Mathias Jensen, Mathias Normann, Moises Caicedo, Morgan Sanson, Moussa Sissoko, Naby Keïta, N’Golo Kante, Oliver Skipp, Oriol Romeu, Ozan Tufan, Pascal Gross, Pierre Højbjerg, Pierre Lees-Melou, Rodri, Rodrigo Bentancur, Ruben Loftus-Cheek, Rúben Neves, Saul Niguez, Sean Longstaff, Shandon Baptiste, Thiago Alcántara, Tom Cleverley, Tom Davies, Tyler Morton, Vitaly Janelt, Will Hughes, Youri Tielemans, Yves Bissouma

ML

Dwight McNeil, Demarai Gray, Dele Alli, Mohamed Elyounoussi, Nathan Tella, Moussa Djenepo, Ken Sema

AMR

Bukayo Saka, Nicolas Pepe, Jeremy Sarmiento, Mason Greenwood, Anthony Elanga, Daniel James, Kieran Dowell, Jarrod Bowen

AMC

Martin Odegaard, Philippe Coutinho, Mason Mount, Hakim Ziyech, Christian Pulisic, Callum Hudson-Odoi, Ross Barkley, Rodrigo, Tyler Roberts, James Maddison, Bruno Fernandes, Juan Mata, Son Heung-min, Dejan Kulusevski, Giovani Lo Celso, Manuel Lanzini, Andriy Yarmolenko

AML

Martinelli, Jack Harrison, Harvey Barnes, Jadon Sancho, Hannibal Mejbri

FR

Bertrand Traore, Jordan Ayew, Michael Olise, Jesuran Rak Sakyi, Mohamed Salah, Takumi Minamino, Gabriel Jesus, Riyad Mahrez, Todd Cantwell, Samuel Kalu

FC

Aaron Connolly, Adam Armstrong, Adam Idah, Alexandre Lacazette, Armando Broja, Ashley Barnes, Bryan Mbeumo, Callum Wilson, Che Adams, Chris Wood (Burnley), Chris Wood (Newcastle United), Christian Benteke, Cole Palmer, Cristiano Ronaldo, Daniel James, Danny Ings, Danny Welbeck, Diogo Jota, Dominic Calvert-Lewin, Eddie Nketiah, Edinson Cavani, Ellis Simms, Fábio Silva, Ferrán Torres, Folarin Balogun, Harry Kane, Ivan Toney, Jamie Vardy, Jay Rodriguez, Jean-Philippe Mateta, João Pedro, Joe Gelhardt, Joshua King, Kai Havertz, Kelechi Iheanacho, Marcus Forss, Matej Vydra, Maxwel Cornet, Michail Antonio, Neal Maupay, Odsonne Edouard, Ollie Watkins, Patrick Bamford, Patson Daka, Pedro Neto, Pierre-Emerick Aubameyang, Raúl Jiménez, Richarlison, Roberto Firmino, Romelu Lukaku, Salomon Rondon, Shane Long, Teemu Pukki, Timo Werner, Wout Weghorst

FL

Wilfried Zaha, Sadio Mané, Luis Díaz, Jack Grealish, Steven Bergwijn

Press enter or click to view image in full size