xGenius, page 3
Figure 1.3: Expected Points per Wage Expenditure, Premier League 2022/23
Unsurprisingly, Brentford and Brighton come out as the best teams in terms of Expected Points per £1 million spent on wages during 2022/23. While it’s true that this view does penalise the big clubs for spending several times more than the smaller clubs, the gulf in financial efficiency between Brentford and Brighton and the rest of the division is undeniable. Based on wage spend, the strongest predictor of success in football, both teams should have been relegated to the Championship. In reality, their xG-driven philosophies allowed them to thrive in the best league in the world. This book will explain how Brighton and Brentford have used data to turn conventional football logic on its head.
These teams are the poster boys of football’s analytical revolution, and other clubs are sitting up and taking notice of their methods. In May 2023, Tottenham chairman Daniel Levy began sounding out betting experts in an attempt to replicate the Brighton-Brentford model. Levy realised that Spurs were struggling to keep pace with other big Premier League clubs, while the two data-driven teams were able to achieve sixth- and ninth-place finishes respectively despite having considerably shallower pockets. Brighton and Brentford’s success was born from the betting markets; their owners’ skill at analysing football and predicting future outcomes. Both clubs live and breathe the science outlined in this book and, as such, will feature regularly as case studies. Their journeys weren’t always smooth. From terrible runs of form to dramatic clashes between the analytics department and management staff, all against the background of immense budgetary pressure, these two Davids have managed to take on and defeat entire leagues worth of Goliaths.
The third club in the triad of analytical frontrunners could certainly be described as one of these giants. Liverpool are European heavyweights who have used data analysis to achieve both domestic and cross-continental glory in recent years. They have faced a different set of challenges to Brighton and Brentford. The latter clubs were in dire straits before being rescued by lifelong fans in Bloom and Benham. Bold and innovative strategies are far easier to deploy at the bottom of the ladder. Beggars can’t be choosers, but football royalty certainly can be. A club the size of Liverpool, despite operating on a smaller budget than certain other teams in England and across Europe, will naturally find it more difficult to achieve buy-in for new or risky ways of thinking. The Reds offer an interesting case study when examining how to implement the science of winning matches at an elite club, and one we’ll visit periodically throughout the book.
xG Discrepancies
Critics of xG might argue that Brighton’s near-relegation in 2020/21 contradicts the argument that data analysis should be a cornerstone of each football club’s operations. The Seagulls represented an xG believer’s perfect blueprint for how a team should be run. They possessed a forward-thinking manager who played an attacking brand of football, an innovatively-structured front office, armed with the best data and analytical toolset anywhere in the world, a highly-skilled management staff packed with a diverse range of experiences, perspectives and ways of thinking, and a recruitment team with a proven track record of mining the transfer market and identifying high-quality players at bargain prices. It all came together on the pitch, as the team created great chances and prevented their opponents from doing so. Yet they still almost got relegated simply because a handful of opposition goalkeepers played out of their skin, some penalty decisions went against them, and their strikers miskicked the ball more often than usual. This team became the laughingstock of the footballing world. Why bother to innovate and forward-think when freakish bad luck means even the best-run clubs can struggle in any given season?
Another data sceptic might use Brighton’s infamous season as ammunition to launch a different attack on Expected Goals. ‘When the xG is so drastically different from the actual number of goals a team scores, it shows how useless and inaccurate Expected Goals is as a metric,’ such a critic might shout. Condemnation of this sort came to the fore during the 2022 World Cup. Only 15 of the first 32 matches (47 per cent) at the tournament were won by the team who created the higher xG figure. In other words, the team who created the most xG failed to win more often than not. A handful of noticeable xG scorelines captured the attention of xG deniers, notably:
Argentina (2.27) 1-2 (0.14) Saudi Arabia
Germany (3.53) 1-2 (1.33) Japan
Belgium (0.86) 1-0 (2.83) Canada
‘Surely,’ the cynics argued, ‘this disproves xG as a valid statistic.’ Other sceptics claimed that these scorelines proved the model was flawed. Ironically, the results experienced in the above matches and throughout Brighton’s 2020/21 season are the situations when we need our xG x-ray glasses the most. The critics are missing the point. The whole purpose of xG is to highlight when performances aren’t matching expectation. In the above matches, the xG isn’t differing from the goals. The goals are differing from the xG. Expected Goals shows the actual ability of a team; the final score is the product once you add a whole bunch of randomness and noise into the mix.
The purpose of xG is to highlight when performances aren’t matching expectation.
Suppose you decide to throw a normal six-sided dice 100 times and count up the total number of ‘points’ that are accumulated. Before the experiment, you might expect an overall points total of 350, given that the ‘average’ score of a dice throw is 3.54 . Say your 100 rolls actually produces a score of 500. Does that mean your dice is broken? Not necessarily. It just means your 100 rolls threw up a series of unusually high numbers. This is useful information. In a football match, both teams are attempting to increase the number of rolls (shots) that they’re allowed, while also increasing the numbers on each side of the dice (the quality of their shots). But those teams who manage to succeed in both tasks aren’t necessarily going to end the game with the highest points total (number of goals).
There have been other, more valid criticisms of xG, and @xGPhilosophy’s Twitter/X presence in particular. The account is set up to post xG scorelines in the following format: Brighton (2.06) 1-2 (0.24) Crystal Palace, with the bracketed numbers representing the xG. The benefit of this format, compared to different accounts who post xG shot maps, timelines, and other graphics from matches, is simplicity. The account’s success has stemmed from the adage, ‘less is more’. The xG stats are presented in an instantly digestible manner and can be easily consumed without having to open an image or read through numerous lines of text. This has facilitated widespread engagement, which in turn has helped xG to reach a wider audience and permeate mainstream football consciousness.
The downside of @xGPhilosophy’s format is the lack of context, which the audience is expected to provide for themselves. It doesn’t show whether there was a penalty in the game, nor whether a team played some of the match with 10 men. Game state can also play a factor in the final xG totals – a team who scores early might sit deep and defend in an attempt to protect their lead. These are important bits of information that cannot be captured in a simple xG scoreline, just as they cannot be captured in the usual post-match presentation of a scoreline (e.g. Brighton 1-2 Crystal Palace). Data alone is inevitably limited in its capacity to describe the world. The application of context is crucial.
Each Expected Goals figure isn’t reality. It’s an attempt at representing reality to the best of our ability. The US federal reserve defines a mathematical model as ‘a representation of some aspect of the world which is based on simplifying assumptions’. Essentially, a phenomenon will be represented mathematically in order to produce a simplified ‘pretend’ version of reality. Each shot’s xG number signifies an attempt to define the chance of that shot resulting in a goal given the outcome of hundreds of thousands of previous shots with similar characteristics. Different models churn out different xG figures for the same shot, but the differences are usually minimal and over the long run all models tend to more or less align. Think of xG models like judges at a boxing bout. One may score a shot one way and another a different way due to their own perspectives. They all have strong points and blind spots. Back in football, one model may occasionally underrate the value of big scoring opportunities, another might not take into account the position of the goalkeeper. Usually, the xG models agree on the winner of a game, but sometimes it comes down to the equivalent of a majority or a split decision.
The reason xG models don’t always agree with one another in the short-term, and the reason xG can never completely mirror reality, is because they offer non-verifiable judgements. Consider a prediction being made about the peak temperature in Barcelona tomorrow, or the result of a presidential election. If you disagree with a friend about these questions, you will, at some point, find out who is right. Now consider a company deciding which of two candidates to hire as CEO. Whichever choice they make, they’ll never truly be certain they made the right one. They can never be certain that the other option would or wouldn’t have performed better than the candidate they chose. Similarly, if an event which a forecaster assigned a 90 per cent chance of occurring fails to happen, the judgement of probability wasn’t necessarily a bad one. After all, outcomes that have a 10 per cent likelihood of occurring do occur 10 per cent of the time. Any probability judgement (i.e., anything between 0 per cent and 100 per cent) can never be truly confirmed or denied. Expected Goals data is an example of such non-verifiable judgements. We can say a shot has a 30 per cent of hitting the back of the net, but we will never be able to verify this in the same manner as we can verify the peak temperature in Barcelona tomorrow.
Chapter Summary
Brighton’s 2020/21 season was one of the unluckiest in recent football history. The team consistently played well but struggled to turn chances into goals and good performances into points on the board.
The Expected Points per Game vs Weekly Wages graph illustrates how well Brighton and Brentford have been performing on the pitch relative to their wealth, validating their use of xG in everything from recruitment to opposition analysis.
A wider context is required, above and beyond xG, to understand how a game has developed. Game state often plays a critical part in adding context to the final xG totals – and that is information that cannot typically be used to capture an xG scoreline.
xG has moved into the mainstream thanks to a growing knowledge of Brighton and Brentford’s use of it for recruitment, aligned with increased online interest in the idea thanks to vehicles such as the @xGPhilosophy Twitter/X page.
Notes
1 That isn’t just rhetoric. If the match was simulated thousands of times according to the xG figures at the end of the game, Brighton would have won in 93 per cent of instances and Crystal Palace just 1 per cent.
2 We will study Expected Points later on. In simple terms, it tells you how many points a team deserved to take from a game based on their xG performances. Note that Expected Points doesn’t try to tell us how many points a team were expected to achieve before the game took place, but rather tells us how many they deserved to win based on how well they played (in the same manner as how xG tells us how many they should have scored based on the chances they created). A team who dominate a game in terms of chance creation but end up losing will amass close to 3 Expected Points, while a team who get an incredibly lucky win will get close to 0 Expected Points. Adding up a team’s figures over the course of a season will give an alternative, more truthful representation of their standing.
3 Although different sources give slightly different figures of Premier League wage expenditure, the same trend prevails: Brighton and Brentford were the two poorest teams in the division.
4 3.5 is the average of the following list of numbers: 1, 2, 3 ,4, 5, 6.
2
The Field of Play
How Randomness Interacts with Football
‘The only sure thing about luck is that it will change’
Bret Harte, 19th century American poet and author
Sergio Agüero. Troy Deeney. Marcello Trotta. These three players have few, if any, noticeable similarities. They each hail from a different corner of the planet; Argentina, England, and Italy respectively. They’ve played football at many different levels and in different leagues, from the Argentine Primera Division, to Serie C, to the Midland Football Combination Division Two. You’ve almost certainly heard of one of these players, probably heard of two of them, but almost certainly not heard of all three. The only substantial connection to be made between the triad is their participation in a last-minute, season-defining moment.
On the last day of the 2011/12 Premier League season, with the score tied at 2-2 and Manchester City needing a win against Queen’s Park Rangers to secure the Premier League title, Agüero collected the ball just outside the box and drove towards goal. He cut past a defender before unleashing a powerful right-footed shot that flew into the back of the net, sparking wild celebrations at the Etihad Stadium and securing City’s first top-flight title in 44 years. The goal was a pivotal moment in the club’s history and the defining moment of Agüero’s career.
The next season, Vicarage Road staged one of the most extraordinary play-off moments in history. With Watford and Leicester City tied 2-2 on aggregate and the game entering the final seconds, Anthony Knockaert won a soft penalty for the visitors. A goal would have sent Leicester to the final, but Manuel Almunia intervened by saving both Knockaert’s penalty and the resulting rebound. Vicarage Road erupted as the ball found its way to the feet of Fernando Forestieri on the right-wing. The Watford player surged upfield and crossed into the box, whereupon a knock-down was smashed home by Troy Deeney. Pandemonium ensued. The footballing world had never seen anything like it before. Although that wasn’t quite true. Two weeks prior, a remarkably similar and similarly remarkable incident had taken place just across the capital.
In the final game of the League One season, Brentford needed to defeat Doncaster Rovers to achieve promotion to the second division of English football for the first time since 1993 and the second time in 50 years. Anything less than a Brentford win would see Doncaster promoted instead. With the score tied at 0-0 heading into the 95th minute, the Bees won a penalty. Up stepped Marcello Trotta, a 20-year-old Italian on-loan from Fulham. Twelve thousand fans held their breath inside the packed Griffin Park. Many turned away, not daring to watch. The referee blew his whistle. Trotta ran forward and sent the ball high to Neil Sullivan’s right. The goalkeeper watched as it smashed hard into the underside of the crossbar and bounced down back into the penalty area. There was a momentary melee in the Rovers box, before the ball was eventually cleared. Brentford had committed players forward in an attempt to scramble home the rebound, which allowed Doncaster to run straight down the other end and score.
It was one of the most dramatic endings to a league season in football history. Doncaster had been one kick away from heading into the play-offs in third position, but 20 seconds later had secured the title. As for Brentford, they ended up losing the League One play-off final to Yeovil Town a few weeks later. For the Bees fans, it felt like a pivotal moment in the club’s history. The team had blown their best chance of promotion to the Championship for two decades. Little did they know, a far more pivotal moment had happened a couple of weeks prior at the club’s training ground, Jersey Road. This moment was a first meeting between Matthew Benham and Rasmus Ankersen, whose union would not only end up changing the trajectory of Brentford FC, but fundamentally alter the way the sport of football is played.
A Game of Chance
Rasmus Ankersen grew up as one of Denmark’s most promising young footballers before a knee injury on his first ever senior appearance effectively ended his playing career. Despite, or perhaps because, his own gift had been snatched from his hands, Ankersen developed a deep interest in talent, particularly in where it comes from and how to identify it. He gained his UEFA A License and became the assistant coach at FC Midtjylland U17s, the same team he played for at youth level. The club lacked the financial might or locational appeal to draw in high-quality players and so relied heavily on producing young talent. Ankersen gained exposure to one of the best football academies in Europe and a club with a strong culture for innovation. He drew upon this experience to write a book entitled DNA of a Winner. The book sold well enough for him to quit his job at Midtjylland, move to Copenhagen and start working as a performance coach. There, he wrote two more books, Leadership DNA and The Goldmine Effect. The latter of these caught the attention of Matthew Benham, the owner of Brentford. Benham’s club lacked the stadium infrastructure or fanbase to make money from ticket sales. The club had little commercial appeal, meaning lucrative sponsorship deals weren’t an option. And the television income for League One teams was next to nothing. The only way Brentford could generate enough revenue to cover their costs was through the smart trading of players. Benham realised Ankersen’s expertise in the identification and nurturing of talent could prove a useful asset to the West London club.
At the time of their meeting, a few weeks prior to Trotta’s penalty miss, Brentford were fourth in League One with five matches remaining, though with games in hand on the teams above. Ankersen asked Benham, ‘What do you think? Are you guys going to be promoted?’ Benham gave a response that Ankersen didn’t quite expect. It wasn’t an answer full of excitement, nervousness, or any other glimmer of emotion. He simply looked at him and said, ‘At the moment, there’s a 42 per cent chance we’ll be promoted.’ In that moment, Ankersen realised he’d met someone who thought completely differently about the game of football than anyone he’d ever met. Benham and Ankersen met regularly over the next couple of months to exchange ideas about how a football club could be run differently. They were particularly interested in how it would be possible to break the strong correlation that existed between spending and results. A team’s wage bill is the strongest indicator of their success and Brentford’s pockets were shallower than those of their opponents. The two men embarked on a mission to outthink, rather than outspend, the competition. Their weapons of choice in this battle were data, analytics, and a profound understanding of the laws of probability.
