Xgenius, p.5
xGenius, page 5
Large accounts which are part of ‘Football Twitter/X’ perfectly demonstrate the ‘all-in’ nature of such predictions. They’ll post forecasts like ‘Marcus Rashford is 100 per cent going to score today, mark my words’ or ‘Barcelona are going to beat Real Madrid 3-0, save this tweet’. If their unlikely prediction comes off, then great. They’ll milk it for all it’s worth. If it goes disastrously wrong? No matter, they’ll still have hundreds of replies from opposition fans mocking them. These Twitter/X accounts are often anonymous, represented by a display picture of a footballer from the team they support. They won’t be bothered by the mass derision headed their way when they’re wrong. In fact, they’ll enjoy it. For them, any engagement is good engagement. Some corners of the punditry world follow a similar philosophy. Being wrong or right is an afterthought; the main objective is to stay relevant. This approach contradicts sound analytical practice which aims to define the truth and has no care for acclaim or attention.
This problem stretches beyond football. Open any newspaper or turn on any TV news show and you find experts who forecast what’s coming. A few are cautious. Most are bold and confident. A handful claim to have supernatural abilities to foresee years into the future. With few exceptions, they aren’t placed in front of the cameras because they possess any proven skill at forecasting. Accuracy is seldom even mentioned. Old predictions are like old news; easily forgotten. Pundits are rarely asked to reconcile what they forecasted with what actually happened. The prevailing, undeniable talent that these commentators have is their ability to craft a compelling story and relay it with conviction.
Old predictions are like old news; easily forgotten. Pundits are rarely asked to reconcile what they forecasted with what actually happened.
Pundits and the media don’t have any checks on them. There is no inherent need for them to be right. It doesn’t matter if a pundit makes a handful of failed predictions – they will still be offered more punditry jobs. Hansen didn’t get sacked because of his disastrous forecast. If anything, his increased publicity would have secured him even more job offers.
Football punditry is a fairly trivial exercise compared to, say, that of the political realm. Politics is full of so-called ‘experts’ who offer strong opinions on the world around them – which candidates will be elected, whether one country will invade another, what policies are likely to be introduced in the near future. These predictions have real-world consequences, but studies tracking the accuracy of these forecasters have shown many bear more resemblance to chimpanzees throwing darts at a dart board than knowledgeable specialists who have a strong authority of their subject matter. Again, these pundits will skew more towards ambitious and outlandish forecasts than measured and considered ones. The analyst who correctly calls a recession that no one else saw coming will gain fame, whereas the analyst who never strays far from consensus will remain obscure. That’s not to say that all political pundits are useless. The task of peering into the future and perceiving future outcomes is a difficult one. There are obstacles to foresight that may not be surmountable, and our desire to reach into the future will almost always exceed our grasp. But football, in particular, has fostered a culture where open-mindedness and inquisitiveness is being suffocated under the weight of uninformed jargon. Without these qualities, we’ll always struggle to cultivate the requisite forecasting skills.
Roy Keane is a classic hedgehog. The Irishman is not one for sitting on the fence and his famed scowl is enough to cast terror into the hearts of his fellow panellists. Don’t expect him to meet you halfway or to assess both sides of an argument. He’s more inclined to make bold arguments, annihilating any opponent who dares challenge his point of view. Keane seldom takes into account the role of probability; uncertainty is a weakness which makes it look like you don’t know what you’re talking about. Cross him at your peril.
The interesting thing about Roy Keane is that he seems to live a dual life. ‘Roy Keane the Pundit’ has all the qualities outlined above, but ‘Roy Keane the Person’ seems to carry a completely different presence. Watching him off-screen, in podcasts or features where he’s not acting as a pundit, he displays behaviour unrecognisable to the man described above. He’s calm and considerate with his choice of words and even exhibits, dare I say it, fox-like tendencies. The curious case of Roy Keane hints at an important truth: the nature of punditry turns foxes into hedgehogs. The viewing audience love ‘Roy Keane the Pundit,’ not because he’s a particularly astute analyst, but because he offers incredible entertainment value. The most viewed videos on the Sky Sports YouTube channel feature vociferous debates between pundits, and Roy Keane is usually involved.
Conflict gets traction. But in pursuing this ‘entertainment factor’, pundits risk losing the analytical, shrewd elements of their character. Punditry is a game of ‘survival of the abrasive.’ Just as the theory of evolution dictates that only the fittest survive, punditry tends to weed out moderate opinions and make successful those who are outlandish and strong-willed. This is not necessarily the fault of pundits or the media. We, the consumers, must share the blame. We’re the ones relentlessly consuming this form of content – conflict over consideration, argument over analytics.
A Brave New World
Historically, the hedgehog-like nature of punditry has had implications for the broader football community. The way fans talk about and analyse football stems from the personalities they watch on TV. Pundits dictate the language of football and, in the past, this language has disregarded a smarter dialect driven by data and analytics. However, punditry has come on leaps and bounds in the last few years. Sky Sports’ Monday Night Football has put xG and other advanced metrics at the centre of the conversation. The BBC were the first movers in this space in the UK when Match of the Day started showing xG scorelines after the highlights of games back in 2017. The Athletic are perhaps the leaders when it comes to data journalism in football, and have even hired a team of specialised analytics writers.
The media are certainly getting better at incorporating data into our native vocabulary by blending advanced stats with engaging narratives. In yesteryear, stats like shots, shots on target and possession were the most advanced you’d read about in a national paper or have beamed into your living room by a television. Now, the likes of xG, Field Tilt and Expected Threat – all of which we’ll study in more detail in latter sections of this book – are being used in mainstream reportage. High-quality, statistically oriented journalism requires investment: the data costs money, you have to build tools to harbour and present the data, you need specialist staff members and you need to invest time into building a culture and environment where analytics can come to the fore. Outlets like The Athletic have taken the game to the next level, telling engaging stories that are founded on cutting-edge research and data analysis. They’ve proved that fans are interested in data journalism, and that investment in this remit can reap rewards.
There is greater scrutiny on pundits nowadays, which in turn has forged better analysis. Their role used to be to provide colour to games, and while that is still the case to some degree, there is now a greater emphasis on providing insight. Pundits nowadays get deeper into the tactical side of the game; they delve into the actual performances of the players. Their aim should be to replicate the sort of insight that a manager or an analytics department will be going into inside clubs. That’s the bar. Studying the little things, taking the scientific examination of the game to the next level, putting players under the microscope, carrying out a forensic examination of each match. They should be using xG to see how well clubs are playing, then cross-reference that with tactical insights to see why teams are performing the way they are.
Not only have the media become better at integrating stats into their broadcasts, they’ve started using visually aesthetic graphics to engage their audience. The presentation of xG has been key to its wider adoption. Analysts have taken inspiration from an unlikely source in their visual representations of xG data. In drama, three unities represent the Aristotelian theory of dramatic tragedy. There is the unity of place, which dictates that the action of a play should exist in a single physical location. The unity of time determines that the action should occur over the course of a limited timescale. Finally, the unity of action decrees that a play should be defined by a series of principal actions.
The unity of place has been imitated by football analysts in their use of xG shot maps. Figure 3.2 shows one such map, that of Harry Kane, which Sky Sports put out across broadcasts and online media. The gradient of each dot represents each effort’s post-shot xG – a metric that we’ll study later on. The stars show the shots that ended up in the back of the net, while the circles are efforts which were either saved or missed the target. These maps are sometimes comprised entirely of dots, the sizes of which represent the xG value. A shot with a high xG value will mean a larger circle, while low xG efforts will be represented by smaller dots. Shot maps have become popular because of their intuitiveness – an xG map clearly shows the quantity, quality, and locations of each team’s efforts on goal.
Figure 3.2: Harry Kane Shot Map, Premier League 2020/21
Shot maps offer a great insight into the chances that a player is accumulating. Harry Kane clearly generated a large number of chances in and around the six-yard box during the 2020/21 campaign. At the point in the season when the graphic was put out, early April, he’d also scored four long-range efforts from outside the box. His chances predominantly came from the left-hand side of the penalty area, perhaps an indication of his right-footedness: he prefers to shoot when he can open his body up and curl it towards the far post rather than hitting it across the goalkeeper from the right-hand side of the box. Clubs can use these graphics to identify the areas that opponents like to operate in, or to assess a potential signing’s ability to generate scoring opportunities.
Sky Sports’ broadcasts have emulated Aristotle’s second unity, the unity of time, by featuring analysis of xG timelines. These graphics show the development of each team’s cumulative Expected Goals throughout a game. A football match does not consist of a random cluster of attacks from either side. Anyone who has watched the sport will recognise that teams will often exert spells of dominance over their opposition. The tide of a game ebbs and flows. Commentators remark that ‘it is important to score while you are on top’, because soon enough it will be your opponent’s turn to enjoy a period of sustained pressure.
Figure 3.3 shows the xG timeline of a match between Norwich City and Manchester United in late 2021, as analysed on Monday Night Football. On the horizontal axis, we have the minute of the match (from 0 to 95). On the vertical axis, we have the cumulative Expected Goals total for either team. Each time either side takes a shot, the line representing them increases by the xG value of the attempt. For instance, the first shot of the match was taken by United in the 11th minute and was worth 0.03(xG). Thus, their line moves above Norwich’s. A shot which results in a goal is signified to have a dot at the top of it. The only goal in this match was scored by Cristiano Ronaldo in the 75th minute, from a penalty worth 0.77(xG). This incident is clearly noticeable in the graphic.
Figure 3.3: xG Timeline Norwich 0-1 Manchester United, 11 December, 2021
The xG timeline gives analysts a unique insight into the spells of pressure which each side endured during the match. Figure 3.3 gives the impression of a cagey game between Norwich and United. Both sides struggled to create clear-cut chances in the first 35 minutes or so. Each team had a half chance to open the scoring just before half time. Norwich missed a good goalscoring opportunity just before the hour mark. With the xG scoreline reading roughly 0.7-0.5 with 15 minutes left, United scored a penalty. Norwich carved out a few low-value opportunities in the remaining time, but struggled to find a winner. Clearly, there were periods when either team was on top, and also spells when the game was more open and spells when the game was more closed up. Expected Goals timelines can tell the tale of any match without the analyst having to watch any of the action.
The final of Aristotle’s dramatic unities is that of action – a performance should be made up of a series of consecutive events. The Bundesliga’s pioneering use of xG in broadcasts has made this unity a reality. As well as showing Expected Goals data for each team after every match, they’ve mimicked Aristotle’s unity of action by showing the individual xG of goals as and when they’re scored. Suppose Joshua Kimmich scores for Bayern Munich, a graphic will pop up in the top-left of the screen showing the xG of his effort. The great advantage of this form of graphic is the precision with which it presents the data. We are able to glean the exact amount of xG that each team or player accumulated in pretty much the exact moment the chance is created. The Bundesliga have also launched explainer videos to educate their audience and have even attempted to combat the confusing term ‘Expected Goals’ by rebranding it as ‘Goal Probability.’6
Gaffers and xG
A foxier, more analytical tone has certainly permeated football broadcast journalism in recent years. This has been reinforced by managers openly discussing xG in post-match interviews and public appearances. Chelsea manager Thomas Tuchel sat down to talk with the media after his team’s defeat to Arsenal in April 2022. The Blues had conceded 3 goals from just 1.12(xG), before a late penalty provided Arsenal with a fourth goal. The result came off the back of a damaging Champions League quarter-final defeat to Real Madrid. Those two legs were preceded by a 4-1 home defeat to Brentford. Chelsea’s form was all over the place and the fans were unhappy.
Speaking after the Arsenal match, Tuchel said, ‘It was a kind of freak result. But it feels like a pattern because we had the Real Madrid and Brentford games.’ That trio of back-to-back home defeats meant Chelsea, who had kept six clean sheets in nine games prior, had since let in 13 goals in six matches. ‘We’ve conceded 11 goals in three home matches,’ said Tuchel. Unprompted, he moved on to the xG stats. Chelsea had conceded 4 goals from 2.48(xG) against Brentford, before Karim Benzema scored 4 goals from 1.85(xG) against them over the two legs of their Champions League clash. At that point in the season, Chelsea should have let in roughly eight goals since the March internationals according to xG. In reality, they had let in 13. The Arsenal result had rubbed yet more salt into Chelsea’s gaping xG wounds.
When Tuchel was pressed on his own views of xG, he gave an unusually in-depth and academic response for a Premier League manager. ‘We’ve known about it for a long time and now it is out there in public, which I think is good because it gives you a more realistic view on your performance,’ he said. ‘You can lose games in football by being unlucky and you can win games with luck and the result very often does not reflect what happened on the pitch. So it gives you a clear view: how many chances you allow, how many shots of which quality you allow. It’s good to have that figure.’ The analysts in the press room were delighted.
‘Like with every number,’ he continued, ‘the more you look into it, you find your benchmarks. So we know over a season or half a season what the level of Expected Goals is if you play in a certain structure and if this is suddenly higher we ask ourselves, “what’s happening within the structure?” If we concede double the amount we are tempted to say we’re in a very unlucky streak at the moment because obviously the quality that we give away is not enough to concede so much and still we concede. That’s hard to take.’
The story of Thomas Tuchel’s xG education began almost a decade earlier. At the end of the 2013/14 season, having just guided Mainz to a seventh-place finish in the German Bundesliga, Tuchel decided he needed a break. Despite securing Europa League football for Mainz, he struggled to see how the club could reinvent itself. Tuchel was determined to expand his football knowledge at this point in his career, so he took some time off to learn about more innovative coaching techniques. Tuchel’s quest for inspiration led him directly into the path of Matthew Benham.
Brentford were flying high at that point. Benham’s team had recovered from Marcello Trotta’s 2013 penalty miss to gain promotion to the Championship in 2014 for the first time in two decades. By the turn of the year, the Bees were pushing for a remarkable back-to-back promotion. Tuchel became aware of Benham’s work and a meeting was arranged through a mutual contact. They were first introduced at a hotel in Hamburg, where their football philosophies instantly clicked. Tuchel was fascinated by how Benham’s data consultancy, Smartodds, used complex algorithms to analyse football matches.
In February 2015, Tuchel came to London for a couple of days. Brentford had controversially decided to part company with manager Mark Warburton (which they eventually did that summer) in order to fully implement their data-driven philosophy. Although it was not a formal interview, Benham launched a charm offensive with the secret ambition of recruiting Tuchel as Brentford head coach. Tuchel was given a tour of the Smartodds headquarters alongside Arno Michels and Rainer Schrey, who had been his assistant and fitness coach respectively at Mainz. The group were granted an exclusive breakdown of Smartodds’ intricate football analysis operation. Tuchel was particularly intrigued by how they evaluated the likelihood of a team winning a match in real-time, using a system founded on xG data. Tuchel also shared his own nuggets of wisdom. He told Smartodds staff members working on player analytics that he was a big admirer of a teenage striker at Stuttgart – Timo Werner. ‘He absolutely loved him and went on forever about his potential and skillset,’ a former Smartodds employee recalls.
