In writing Get Goalside, I don’t usually like being personal - but today, today I’m going to open up about my least favourite fact in the entire world.

The most generous thing I can say about the ‘fact’ “there’s no such thing as a fish” - made famous by the British TV show QI and its spin-off podcast - is that it isn’t actively harmful. It’s not going to run over your child or poison your dog. It won’t force you to watch Mikel Arteta’s Arsenal. But, as much as the categorisation of aquatic life into a single group is zoologically problematic, by heck it’s tiresome to go ‘that thing which obviously exists, doesn’t’.

Which means that it’s with great sheepishness that I say: ‘defensive lines’, they do not exist.

Sorry, I can’t bear letting the Bit last this long - defensive lines clearly exist, it’s just that they’ll be more meaningfully understood if we poke at the definition a bit. [exhales] That feels better.

Poking at definitions is something I usually do with event data. For example, a few days ago I was looking through this year’s Get Goalside posts (for reasons), and re-read this line from the end of ‘Just run some more’:

“There’s another advantage that running data has. No-one has to argue about what a ‘duel’ is.”

Running data is, in a sense, elegant. It’s pure and simple - the movement of a person through space. Sure, we divide it up at increments of speed, but ultimately it’s a case of ‘fast’, ‘not as fast’, ‘double-speed fast’, ‘Micky van de Ven fast’. Lovely.

The way that event data works is obviously different, more similar to trying to fit art into genre boxes, but if you wanted to then you could approach it in a similar way to running. The actions which event data collects are usually a case of a player enacting a change in motion of the ball. Shots are fast (usually) strikes of the ball towards the goal; passes are somewhat fast strikes of the ball towards a teammate; ball-carrying is just a player making small strikes of the ball. (Duels are difficult because you need to get granular about limbs, but they’re still just a matter of opposing players, one of whom is (usually) the ball-carrier, coming into proximity).

And defensive lines, then?

If you’ve ever heard a manager pooh-pooh the idea of formations (‘telephone numbers’), you’ll know that tactics are in the eye of the beholder. [A sidenote: I would pray that everybody keeps in mind the love they hold for the ‘fluidity of football’ next time they ask for stats about ‘goals/shots from set-pieces’. Nobody knows when a set-piece becomes open-play. If God had wanted there to be clear delineation between phases, he’d have invented the NFL.]

The best paper that I’m aware of on this area of football, identifying formations and phases from data, is this one: ‘Putting team formations in association football into context’. And even here, I think that parts of it - like the notion of ‘fish’ - are convenient fictions.

With the utmost respect to all football coaches, if it was a data analyst proposing this rather than video analysts they’d be glared out of the room. The four-phase (plus set-pieces) split may be how coaches want football to be played, but it has to be coached very strongly before it even starts to look like it falls into this pattern. Y’know who I’d call as a witness if going to trial on this? Any coach who’s introduced ‘artificial transitions’ to their in-possession play.

All of these tactical notions are also too strongly tied to coaching practices to be usefully used as generic concepts. A classification of football shouldn’t break down just because Fernando Diniz walks in the room. Running data and event data classifications will still work if you’re looking at a match from 2025 or 1955; the tactical aspects should as well.

So how can you approach this?

If running is just the movement of a player and events are the interaction of player and ball movement, tactical features are- well, first, something that they’re not. They’re not the movement of groups or teams or units, because that just creates the question of defining the unit. Instead, in all concepts I’ve been able to think of, they could be said to be the coordination of movement of players.

Happily, this allows me to say ‘yes, there is such a thing as a fish’. A defensive line can still exist in this ‘coordination of movement’ way of defining things, because the main feature of a defensive line is that it’s coordinated in one dimension. A ‘press’ is a particular set of coordinated movements*. ‘Established possession’ is (simplifying slightly) the moment at which the movements of teams have become sufficiently coordinated. ‘Rest defence’ is a coordination of positioning in deep central area when a team has advanced possession.

*different types of press, or a disjointed press, could be flagged by identifying coordinations between a small number of out-of-possession players, whose collective movements are un-coordinated with the rest of their team.

If this was a grant application instead of a blog, I’d need to argue for the practical implications of this insight. It’s fortunate that it’s not (I’d have never slipped the Arsenal joke past the editors). But the applications have already started: the Opta Forum talk by Guillaume Hacques which I mentioned last post, focused on asymmetries in defensive blocks.

(It’s probably not a coincidence that the focus on asymmetries and coordinated movement was prompted by the submission question posed by Monaco’s current Head of Sporting Insights Vignesh Jayanth (in 2023, at Stade Rennais)).

I suspect that ‘phases of play’ can come down to gradations of stability (which in itself can be related to coordination of player movement) - which will have at least one extra gradation than ‘transition’ and ‘not transition’, the tactical equivalent of dividing player movement up into ‘running’ and ‘not running’. I suspect assignation of player marking stems from the same principles (and now I think about it, think I’ve seen work doing that before(?)). And therefore so will dis-marking.

Fish exist. My life is very slightly incrementally enriched by learning that categorising animals is hard. I hope this post has been minimally tiresome.

If anyone knows of research on line-breaking passes, could you send it my way? I was reading a piece which said that there wasn’t much research on it, and searching Google Scholar that appears to be correct, but surely such a strongly-pushed metric would have some research examining it….

In true newsletter style, this will be a collection of thoughts. Mercifully, it’ll be short.

They’re both in the grand Get Goalside tradition of ‘the nature of football'. The first is a fourth-dimensional take on Rafael Benitez’s football tactical metaphor of the ‘short blanket’. The point is that, like a short blanket will always leave part of you cold, a defensive team will always leave part of the pitch exposed.

Football is a sport where it’s very difficult to score a goal, played for a length of time which is very long. What an absurdity to play a game for an hour and a half and, very possibly, not score a goal at all. This is the most popular sport in the world.

And so, there isn’t, and possibly never will be, a tactical blanket that a team can throw over a football match and cover it all. (Though of course, you can cover the match better or worse). The point is that in any tactical consideration, time is a factor. Not a revelation, but also not something that is talked about as much as all other tactical aspects.

Secondly, on a team level, is the idea of ‘stability’ in a team. This is a thought that stems from the perennial problem, ‘why do some teams game the progressive pass metrics?’. (Not ‘why do they choose to game them’, but ‘why do their stats look as if someone were gaming them’). At time of writing, three Premier League teams have averaged between eleven and twelve passes into the penalty area per game (FBRef) - Arsenal, Liverpool, Manchester City. By FBRef’s (Opta’s) numbers, Liverpool have created around ten expected goals more than City, and almost 20 more than Arsenal.

So, stability.

Teams attack and they defend. The ideal, for an attacking team, would be of being in a stable state and attacking an unstable defence. The issue which we see so often is that a team seeking stability will often allow the defensive team to find stability. Equally, a team that tries to attack while a defence is unstable will often be in an unstable state themselves. There are teams which enter the box when the opposition defence is essentially stable, which is usually still dangerous, but the ‘value gap’ in these ‘entry’ stats gets wider when you take a step back to the final third, or to ‘progressive’ metrics.

The pass stats are worthy attempts at proxying danger, but there’s a risk that teams trip themselves up by chasing them.

But what is stability? A couple of past Opta Pro Forum presentations have touched on this - Mladen Sormaz and Dan Nichols investigating shape-disrupting runs (2020) and Guillaume Hacques looking at destabilising a set defence’s symmetry (2023). [the biggest race in football analytics is between ‘solving football’ and ‘establishing a central, easily-findable location for the history of the Opta Pro Forum’].

My current sticking point is finding a ‘stability’ definition which works for counter-attacks. If a team has been caught out by a ball over the top and is running back to its goal, chasing an opposition attacker, in perfect synchronisation, it’s not exactly ‘stable’. You could just say ‘look at pitch control’, but I suspect that, when facing set defences, instability often precedes the changes that pitch control would be detecting. It could be as simple as ‘stable systems don’t move quickly’.

A side effect of dwelling on ‘stability’ is what it would do to the metrics that come out of it. You’d probably end up with something like ‘how many times did a team create instability from a stable defence’… but that would leave open the possibility that a team racks up numbers by (unintentionally) allowing the opposition defence to get stable lots of times. Aiming for a ‘high percentage of stable defences rendered instable’ is not even necessarily a good goal.

The one downside of this focus on stability? It gives all the physics PhDs in football even more to do. Football isn’t played on a Newton meter.

Over the years, I’ve met a lot of people working in football from around the globe. The nature of this industry is mobile, international, insecure, and ever-changing - my thoughts have been with those of you affected by the many life-altering circumstances that have been taking place around the world recently.

—

Every now and then, an idea just sticks with you. Most often it’s not so much the idea itself, but the way the idea is phrased. After all, people tend to echo the ideas that others have had before them, but it can sometimes be hard for a thought to take root until the right seed is sown.

‘Building capacity’ was one of these for me. It’s a phrase of long-time friend-of-the-newsletter Tiotal Football (of Absolute Unit and the Post-Script podcast). The phrase referred to the way that teams act in possession, a subtle difference to usual terms like ‘ball progression’ or ‘build-up’. We’ll get to the difference in a second.

First, the thought had to flower. It had latched, roots don’t make a fruit. (even if you can get some useful tubers). At some point, Tiotal also made a comment about cut-backs being so effective as goal-creation chances partly because they give the defending team a particular set of bad choices. It’s like the Rafael Benitez ‘short blanket’ metaphor for tactics in miniature.

‘Building capacity’ is shuffling things towards forcing these bad choices on your opponents. If cut-backs are the ‘fork’ - to play to the readerbase and use a bit of chess lingo - then building capacity is the move before the move to create the fork.

The idea of ‘build-up’, ‘ball progression’, and particularly ‘building through the thirds’ had always jarred slightly. It’s true that moving the ball closer to goal helps create chances, but how often do you see a team progress the ball through the thirds, establish possession high up-field and then… just be faced with a robust, structured defence. Ball progression is often relatively easy until you hit the point that the out-of-possession team is actually bothered about defending. Sometimes a mid block is more of a negotiation tactic than anything.

Building capacity will often mean that the ball moves forwards, but it ball progression doesn’t mean that the capacity has been built.

Neither, necessarily, does breaking a line.

I’ll grant you, breaking a line is better than breaking most things. It’s definitely way, way better to be a line-breaker than a heart-breaker, leg-breaker, strike-breaker. However, a pass that breaks a line is not necessarily ‘building capacity’. You could put together incredible highlight reels of players breaking the first line of an opposition block, which the opponent immediately recovers from, or breaking a midfield line to a teammate facing away from goal with no passing options.

A sidenote: a focus on breaking lines with passes de-emphasises the ability of ball-carrying to shake up an opponent. We need more centre-backs maraudering and gallivanting. 

The problem with ‘building capacity’ is that it’s difficult to count. But culturally we seem to be swinging back towards valuing expertise instead of just wanting easy-to-grasp simplism, so maybe that’s not a deal-breaker.

Tracking data might come sweeping in here claiming to save the day. There are already companies like [no, I’m not giving free advertising] who have models calculating both the likelihood of pass completion and the pitch value of the potential receiver. You could imagine a step further than this, some way of factoring in immediate options that the receiver would have if they received a pass.

Could thinking about ‘building capacity’ (or ‘building bad options for the opponent’) help make modelling choices easier? Let’s take backwards passes (the anti-line-breaker). Analytics people have often argued that backwards passes aren’t necessarily ‘bad’, because they could be the pass which opens up a progressive pass. This fits in the framework of ‘building capacity’. But they can also be valuable for removing risk of a turnover, which is a ‘bad option’ that the opponent is trying to force on the in-possession team. Although the quality of options might still be assessed in a goals-derived metric, you’re dealing with a much larger bunch of ‘rewards’ than if you’re looking at goal-creation directly.

Sometimes the better option might not be taking a pawn, not even waiting for an opportunity to take the queen, but starting with a fork on the bishop and rook.

Football vs Transphobia week of action, 25th-31st March: https://www.footballvhomophobia.com/fvt/

In the beginning there was darkness.

And then… Then there was the light of a headtorch, illuminating the notepad of Royal Air Force accountant, Wing Commander Charles Reep. It is 1950. Like many of us, Reep is annoyed at a football team. Unlike many of us, he’s so annoyed that he’s marking down his own data.

Within a few years of beginning his data notation at Swindon Town, of all teams, Reep will be a minor celebrity. He will be credited with helping the success of Wolverhampton Wanderers, the most successful English team of that decade alongside only Manchester United. They won three league titles, two FA Cups, and their midweek friendlies against Continental rivals helped inspire the European Cup.

This is mid-century England, and it’s the first wave of football analytics.

And now, seventy-five years later, we have team cohesion of pressing chains, among other shiny things that data company Skillcorner is promising in their latest launch of metrics. This is quarter-century England - at least where this newsletter is being written - and it’s the latest wave of football analytics.

How did we get here?

**

As much as it would be nice and neat to begin football’s data history with Charles Reep, we must go back into older and fuzzier times. Newspapers from 1920s and 1930s Hungary show visualisations that would look modern in football coverage today.

A ‘match chart’ from 1922, from ‘Three Sportviz Inventions By a Hungarian Newspaper’, Nightingale

According to Attila Bátorfy, the writer of the linked article, the Hungarian newspaper Nemzeti Sport mentioned its creations being copied by outlets in Italy and Sweden. We know disappointingly little about this spread of football datavis. Historical terms are always relative to geography and the present day, and so for the moment Charles Reep remains our most secure starting point for ‘football analytics history’. But it seems very likely that as long as there has been football, there have been dorks making data out of it. Often it hasn’t been imagination that has held things back, but finance or technology.

Those charts (which, to reiterate, were published around the time Hungarian icon Ferenc Puskás was born) focused on themes we’d find familiar. Shot maps, for the match’s chances, and a momentum chart for which team was on top. These were also themes that Reep homed in on in his data. In 1997, a paper was published in the journal The Statistician using Reep’s data. The statistician Richard Pollard - who by that point had worked with Reep for more than a decade - put together what would nowadays be recognised as expected goals and possession value models. The match’s chances, and which team was on top.

Reep, though, was motivated specifically by what helped teams to win. This was a difference from the match charts of earlier newspapers. It also seems like he was motivated by a particular tactical viewpoint, a familiar-sounding one: that the modern style of short passing was overhyped.

From our modern perspective, this seems flawed to say the least. And Reep’s analysis of the data he collected, in articles that he wrote in newspapers and magazines, often seemed cherry-picked. However, some context about 1950s football is useful here.

Imagine the dying stages of a modern football match, when both teams can still get a result but everyone is very tired. The game becomes spread, formations become less clear, as players get caught between wave after counter-wave of attack, unable to get back ‘into position’. This - based off a bit of Footballia-watching - is what 1950s football was like all game.

Teams were simply less compact, which meant that passing was much more forwards, as it often can be in the sunset period of tense, modern matches. Teams on the ball weren’t facing a defensive block so much as a defensive lattice. In this context, sideways passing instead of forward movement would allow a defensive structure to form, something which our modern experiences simply take for granted.

Technical ability was also lower. If Reep had been around for the tactical evolution of the 2010s, he’d have probably liked the high counterpressing approaches coming out of Germany. If your 1950s-calibre players were going to be prone to miscontrols, why make them in midfield and not, after a longer ball, in the final third?

Although he was probably correct that there was efficiency to be squeezed, Reep’s public analysis was contested in newspapers at the time, and his professional involvement in the ‘50s and ‘60s didn’t lead to an ‘analytical revolution’. Successful, popular English teams still played ‘nice football’. Yet Reep’s data collection being affected by his opinions about football was not to be a one-off.

**

All data collection implicitly has an opinion. Way back to Nemzeti Sport, the data that was created was the overlap of what ‘mattered’ and what could feasibly be collected. Part of Reep’s skill was developing a shorthand system that could collect much more information than you would expect to be possible. (It may have helped that he was an accountant who joined the British Royal Air Force’s administration - numbers and regimented systems was his day job).

Those who took, and possibly bastardised, Reep’s research were inspired by contemporary trends themselves. Shortly after the 1966 World Cup triumph, results turned dismal for the England men’s team. They failed to qualify for all four major tournaments between 1972 and 1978, and had disappointing results at the following two that they actually qualified for. After the disappointing 1982 World Cup, an article in the Daily Express pointed to the beliefs of FA assistant director of coaching, Charles Hughes, as the way forward:

(Although the article says that Hughes learnt from Reep’s work, it names Reep as a ‘retired naval commander’, the wrong branch of military. It seems likely that the journalist knew Hughes’s opinions far better than Reep’s).

These facts: few goals come from long passing moves, lots of goals come from set-pieces or final third turnovers. Hughes and another figure the article mentions, Graham Taylor, got their chance a decade later. After a memorable fourth-place under Bobby Robson at Italia ‘90, Taylor got his chance; unfortunately, his England side won zero games at Euro ‘92, and failed to qualify for the 1994 World Cup. Few goals came from long passing moves; fewer came from a long-ball England team.

**

Having traveled the historical timeline all this way, it’s only a short jump to the founding of two industry-defining modern data companies. As has been the case with the characters we’ve seen so far, operations were influenced by circumstance.

Opta and Prozone (like Reep, both English) came into being around 1996-1997. Fresh off the back of a memorable semi-final for Terry Venables’ England at the ‘96 Euros, Opta would be collecting data for the Premier League. But they needed to fund their enterprise. A hungry media with pages and airtime to fill helped immensely. Prozone, meanwhile, found business viability through video analysis and player running load data.

Technology helped each company create a fully systematic data collection operation, where Reep had initially been limited by needing to not only be at the games himself but to process all of his shorthand into data by himself too. Yet by the mid-2010s, the technology was not enough. The context of these companies’ origins and the context of contemporary football was beginning to frustrate people.

The wave of Pep Guardiola imitators had made it clear that incisive passes needed to be identifed, while the pressing style emerging in Germany made the conventional collection of tackles and interceptions seem… small-fry.

A German company called Impect sought to address the first, with an emphasis on how many players passes ‘bypassed’, while the second was taken on by England-based (again) company StatsBomb and their ‘pressure events’.

Prozone, meanwhile, had apparently grown used to football clubs being uninterested in the details of their data. They “had been the gatekeepers of tracking data for many years and were loath to share it with anyone,” wrote Ian Graham, Liverpool’s Director of Research from 2012-2023, in How to Win the Premier League. “The data revolution in football could have happened years earlier if it wasn’t for Prozone’s protectionism.”

Little surprise, then, that Graham helped lobby for the Premier League to institute full access to tracking data in the early 2010s. Not only that, but a couple of years after league-wide sharing of data was finally agreed, in 2016, Liverpool Football Club welcomed a new tracking data company.

It was a relatively new company, and evidently neither side of the partnership wanted to wait around for innovation to evolve slowly. For once - unlike Charles Reep, Hughes and Taylor, Opta and Prozone - it was not English by birth. It was Parisian. And it was wanting to do something completely crazy.

**

Around the time that Opta and Prozone were being set up, the host for the 2002 World Cup was chosen. It would be the first-ever shared World Cup, between South Korea and Japan; the first-ever in Asia; and the first since the 1930s when a host, at the time of announcement, had never qualified for the competition before. (Japan would shortly afterwards qualify for the 1998 edition, their first appearance).

Whether caught up in this World Cup buzz or not, this period saw some interesting work by Japanese academics. Beginning in 1996, Tsuyoshi Taki and Jun-ichi Hasegawa published a series of research papers on the ‘dominant region’ of teams - areas of the pitch that one team or the other had hold over. Or, in modern terminology, ‘pitch control’.

The idea and calculations would still appear innovative twenty years later, but a problem at the time was getting the data itself. Work was published in academic journals for computer vision (a field where computers process images to detect objects, basically), but results were either small-scale or theoretical. Taki and Hasegawa even had to propose an in-stadium camera set-up that would work for their calculations. (“The telecasting image is not useful for analysis,” they wrote, “[…] because most scenes are intermittent and are focused on a player with the ball.”)

This in-stadium set-up was the kind of thing that Prozone was able to do, but still suffered - to a lesser degree - with what Taki and Hasegawa believed made TV footage unfeasible to work with. The issues of players being off-screen occurred on a smaller scale when they passed in front of each other, blocking one player or another from the view of the in-stadium cameras.

Technology and imagination, though, find a way. Among other things, TVs got bigger.

When Liverpool partnered with Parisian-based Skillcorner in the late 2010s (yes, we’re finally getting back to Skillcorner), the camera shots of football matches were both larger and crisper than they’d been in the late 1990s. Computer vision technology had also improved. And so too had understanding about how football teams moved.

Football coaches talk about ‘shape’ and ‘units’ a lot. Teams move in coordinated ways - in defence, particularly. This is hardly ground-breaking information, but it means that with enough processing power and prior data you can make smart guesses about where off-screen players are. And with a wider aspect-ratio of television, there tend to be fewer players off-screen, who are off-screen for a shorter time.

All this combination of technological advances allows companies to create more and/or better tracking data from relatively cheap, and very available, TV footage. For the first time in history, teams could scout with tracking data.

But what to scout? Would you simply use the ‘distance run’ figures, the thing that English coaches of the ‘00s seemed to like?

The fascinating tension in innovation is finding the balance point where imagination can be pushed just enough past technology. The latter often proves to be easier to improve than the former.

As it turns out, just like Impect and Statsbomb created unique selling points born out of frustration with the industry’s existing data, the world of ‘physical data’ stats has its sticking points too. At a similar point in time, the early 2010s, researchers were trying to find better measurement points when it came to player running too.

Separation between speed thresholds, ‘running’ vs ‘high-intensity runs’ vs ‘sprinting’, was fairly established, but still dissatisfying. In a particularly punchy 2015 research paper, subtitled ‘Shedding some light on the complexity’, we get this line: “Contemporary time–motion analysis of soccer still only offers a basic snapshot, and it is imperative that future research attempts to integrate multiple approaches to unravel the complexity of the game and its performance determinants.”

Three years later, one of the co-authors was involved in a paper whose title went even further. It was called “Are Current Physical Match Performance Metrics in Elite Soccer Fit for Purpose or Is the Adoption of an Integrated Approach Needed?”.

The floors of the 2010s data scene were littered with gauntlets being thrown.

**

No longer would it be acceptable to just focus on running at different speeds. Running - in fact, all movement on a football pitch - has a tactical component. That, ultimately, is the field in which Skillcorner’s ‘Game Intelligence’ data, particularly the recent out of possession launch, is pitching itself in.

Early event data captured on-ball events like tackles; later event data added direct pressures to the ball-carrier; the latest wave adds layers of variety about what effects that pressure has. There’s even a metric for how many times a defender turned their opponent backwards, something that Get Goalside imagined for an alternative football data universe several years ago.

It remains to be seen whether this fancy new data - like all data - is as reliable and insightful as it markets itself to be. The point at which Get Goalside tends to turn from enthusiasm to scepticism is - out of step with some media outlets - when someone starts flogging it. Thankfully, no-one is (yet) shilling a service based on the historical lineages of football data innovation.

To say that history repeats, or echoes, suggests that history itself is alive. Humanity repeats itself. People echo people that have come before them. It’s no surprise that association football is an echo of other ball sports from global history, because the bounce of a ball is a magical thing. It’s no surprise that the urge of a small group of people, to understand those bounces a little better, echoes too.

We know for sure that, a century ago, there were people tracking the key events of football matches. We know that, as England was still only five years removed from the Second World War, extensive and systematic data collection of match events was underway. We know that forerunners to modern tracking data applications are almost thirty years old.

Many things about the game have changed. For one, tactics no longer have the athletic limitations of the 1950s: research suggests that high-intensity running increased 30% in the Premier League between 2006 and 2013, and by a further 10% between 2014/15 and 2018/19. This increase in athleticism may well be an inspirational spur for some of the latest data innovations, or at least making the ground fertile for Skillcorner to put its roots in.

But many things are the same.

In late 1960, Reep wrote a pointed article about Tottenham Hotspur’s victory over Stan Cullis’ Wolves in the FA Cup, suggesting that their short passing would make their success short-lived (they ended up winning the double that season). In a response to Reep in the Hull Daily Mail local newspaper, their football correspondent closed his piece with the following:

Tactical variety, tactical debate, and arguments about data. This is mid-century England.

References/Further Reading

Attila Bátorfy, ‘Three Sportviz Inventions By a Hungarian Newspaper’, https://medium.com/nightingale/three-sportviz-inventions-by-a-hungarian-newspaper-b5c0df489d6c

On Charles Reep: https://www.bbc.co.uk/sport/football/61084931, https://www.bbc.co.uk/sport/football/61648608, https://www.wsc.co.uk/the-archive/grim-reep-/

Ian Graham, How To Win The Premier League

Rory Smith, Expected Goals

Richard Pollard & Charles Reep, ‘Measuring the effectiveness of playing styles’ (1997): https://www.researchgate.net/publication/227692321_Measuring_the_effectiveness_of_playing_strategies_at_soccer

Paul, Bradley, & Nassis, ‘Factors Affecting Match Running Performance of Elite Soccer Players: Shedding Some Light on the Complexity’ (2015): https://www.researchgate.net/publication/273071207_Factors_Affecting_Match_Running_Performance_of_Elite_Soccer_Players_Shedding_Some_Light_on_the_Complexity

Ade & Bradley, ‘Are Current Physical Match Performance Metrics in Elite Soccer Fit for Purpose or Is the Adoption of an Integrated Approach Needed?’ (2018): https://www.researchgate.net/publication/322277340_Are_Current_Physical_Match_Performance_Metrics_in_Elite_Soccer_Fit_for_Purpose_or_Is_the_Adoption_of_an_Integrated_Approach_Needed

Taki, Hasegawa, Fukumura, ‘Development of motion analysis system for quantitative evaluation of teamwork in soccer’ (1996): https://ieeexplore.ieee.org/document/560865

Taki, Hasegawa, ‘Dominant region: a basic feature for group motion analysis and its application to teamwork evaluation in soccer games’ (1998): https://www.spiedigitallibrary.org/conference-proceedings-of-spie/3641/1/Dominant-region--a-basic-feature-for-group-motion-analysis/10.1117/12.333797.short

Taki, Hasegawa, ‘Visualization of dominant region in team games and its application to teamwork analysis’ (2000): https://www.semanticscholar.org/paper/Visualization-of-dominant-region-in-team-games-and-Taki-Hasegawa/beff32a0a37a8d094a471067895cf420dd2e20de

Previous Get Goalside on the subject of Charles Reep and analytics history: https://www.getgoalsideanalytics.com/36315087-analytics-is-older-than-you-think/

Get Goalside, ‘What if we didn’t care about passes?’: https://www.getgoalsideanalytics.com/what-if-passes/

A few weeks ago, I made Sportscode. Hudl, watch out.

Well, actually, the point is that Hudl doesn’t need to watch out, because - alas - there is a difference between a working DIY video clipping bit of software and an actual product.

A regular theme of Get Goalside over the last year or so has been that football clubs probably shouldn’t all be tech companies. Now, part of that is because tech companies are increasingly susceptible to megalomania, but part of it is also practical advice.

The world of software engineers is currently split between those who love ‘generative AI’ coding tools and those who think they’re kinda trash. In the ‘love’ camp are people who like to move fast and break things. In the ‘trash’ camp are people who don’t like things being broken. (No but seriously; tools introducing bugs that are hidden because of a lack of oversight are basically the main complaint).

My version of Sportscode has a video player, custom-assignable hotkeys, an option to export to XML, and was created in about four hours (which, to be totally honest, feels slow). What it doesn’t have is any infrastructure that makes me trust it won’t crash if I try and code half a football match.

And yet, and yet, and yet.

The fact that someone with a passing grasp of React (the Javascript framework that it’s written in) can ‘manage’ a coding tool through the task of creating a fully working DIY app does change the landscape for football clubs.

I think the nature of these code generation* tools still means that you shouldn’t be using them to set up critical infrastructure, but developing, say, an quick, basic expansion to an existing set of regular reports for coaches; or a webapp to let players interact with their data? Seems more plausible.

The downside, the major downside, with any of this ‘small team employee-produced’ software is that, while they’re at the club, they’re the creator and the support desk and the bug-fixer. If and when they leave, you don’t care so much about the first of those roles, but you’ll probably still care about the other two. (That’s a large part of why you don’t want to build critical infrastructure this way too).

Another way that it changes the landscape is that what can be done with these tools will change depending on the user’s knowledge level. The tools probably become more powerful the more advanced the user is, for a similar reason as to why experienced line-managers with domain expertise can be useful. Part of managing is delegating, and part of managing is experience-sharing. ‘Ah, this sounds like a problem I’ve experienced/read about before, have you tried this?’ can be surprisingly powerful. People without coding experience - or without much of it, or without much football experience - can do the delegating with code generation tools, but they can’t do the experience-sharing.

Bundle that together, and the value of a fairly-experienced data-and-tech-person may have gone up by more than a fairly-entry-level one. And that’s quite interesting.

As a serial metaphor-er, I’ve been searching for the right comparison for the way that these code generation tools lead to changes. Basic as it may be, power tools of the ‘normal DIY’ kind are probably the best thing.

Imagine everyone had access to materials and power saws and belt sanders. Everyone on your street would, in theory, be able to build you tables and chairs for your business. You probably wouldn’t want them to, unless they could already build something sturdy. But there’d be uses for it. You might be able to redo some cupboards. They might be able to replace some faulty chairs in a pinch. Again: not critical infrastructure, things that have some common templates to follow.

What kind of DIY jobs do you consider doing yourself at home and what do you think is best to get a professional in for? Those same kinds of dividing lines seem applicable to the new age of DIY software in football clubs.

Hudl don’t need to watch out. Not yet, anyway.

*I am loathed to call everything ‘AI’. The major innovations have been around language models and their surrounding infrastructure. This doesn’t quite mesh with the term ‘generative AI’ which briefly had some time in the sun, because a lot of value comes from language interpretation as well. But the coding tools specifically tend to be useful because they produce code, therefore ‘code generation tools’. Let’s all remember that we don’t need to use the buzzwords that companies trying to show value to investors try to coin.

Football vs Transphobia: https://www.footballvhomophobia.com/fvt/

In the olden days, before even the Proper Football Men were denouncing spreadsheets, football had different rules. Depending on your viewpoint, it either had too many or none at all: yes, we’re talking pre-1863 codification.

Football was very popular even before it was The Beautiful Game™️ (sponsored by Kingdom Airlines), but everyone had their own way of doing things. This limited what you could do. It made it tough to play with other teams, and with so many interpretations some were bound to be worse than others. [a cheap crack at rugby could be made here]

And it’s with this in mind that we head to Japan, to quote a recent paper about analytics: “we propose […] a unified framework designed to streamline event annotation, data standardization, and various deep learning modeling for soccer analytics.” This is OpenSTARLab, coming from a group of Japanese researchers, the latest addition to a noble line of groups who want to make things easier.

As the paper references, they’re not the first. In 2019, a group of researchers from the Belgian university KU Leuven and Dutch company SciSports presented the Soccer Player Action Description Language, in a paper on their possession value model. The fellow Central European-born Kloppy Python package has a similar concept. The idea: while data providers stubbornly produce different types of event data, there are fundamental similarities that can be mapped into the same ‘language’.

An aside: there’s a tangent we could go down here, on a favourite Get Goalside topic of the internationalisation of analytics after a period of heavy English centrality. In fact, we’ll come back to it in a bit.

At one point in time, I even wondered whether FIFA would get in on the act. In 2021 they launched their FIFA Football Language. Arsène Wenger’s opening note uses the phrase “open-source”! Unlike the other frameworks, FIFA weren’t (yet?) trying to squeeze other data providers into their football language, but it formed the basis for their own data collection for FIFA tournaments.

Just like in the 1860s, the football community would quite like things to be simpler. Over the past few years, and for the next few years as well, a wave of football clubs will be going through their first major data provider switch. In many cases, this is because data providers have different offerings - more detailed and/or more shiny - but these differences don’t make up the majority of an event data spec.

In May last year, when writing about STILL-yet-to-be-replaced-as-Chelsea-front-of-shirt-sponsor Infinite Athlete, I asked a question. It was a question mainly about data engineering, but applies to the issues that OpenSTARLab is trying to solve too:

Maybe part of the ‘something else’ will be packages like kloppy and OpenSTARLab. If they become robust enough to use as standardisation systems while still offering the unique features of chosen data providers, a football club will only have to build a data system once*.

*(Well, as much as anyone only builds a data system once. The need to re-build and re-write would certainly reduce, and/or the need for them to develop their own abstractions).

This still doesn’t mean that it’s sensible for all football clubs to create their whole data engineering and software infrastructure themselves. It’s not. Smart (and sufficiently wealthy) football associations should be helping their clubs to establish a baseline standard, particularly if they’re a country in some sort of continental competition coefficient race. Like, I dunno, Belgium.

Unfortunately (from certain perspectives), this is unlikely to be a problem that really slaps people round the face for another few years. Not only are there many leagues where ‘using data’ means using software rather than the raw data itself, but therefore many leagues where it’s a competitive advantage to quietly build things yourself even if that means building inexpertly.

Now would, I suppose, be a good moment to declare the interest of working for a company that 1) deals with multiple event data providers 2) deals with interoperability between different data providers 3) has (smart, capable, witty) employees to pay and investors to create value for. Hopefully you, dear reader, trust that Get Goalside’s only biases are towards the entertainment of yourselves and, more importantly, of the writer.

(Also to the reduction in usage of the word ‘democratise’: democracy has enough on its plate without being dragged into sales pitches).

At some point, the boring parts of using data in football will get easier. What’s less clear is how.

Football got codified when a group of English elites decided to argue until they reached an agreement (which was then tweaked and then completely rewritten for clarity in the 1930s). Their agreement was picked up as the standard by the world. Who will end up writing the standard for football data collection and processing?

…

All of that was written about a week ago, and I didn’t get around to finishing it off. And then along came the Open-Source Avengers.

🌉🌁 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐛𝐫𝐢𝐝𝐠𝐞𝐬!

We brought together contributors from different open source projects to discuss how we can align our work and improve interoperability in football analytics.

Looking forward to the next steps. #FootballAnalytics#OpenSource#PySport

— PySport (@pysport.org)
1:46 PM • Feb 26, 2025

Wonderfully, not only does it circle straight back to the origin of this post, but goes right back to the internationalism point. Six nationalities are represented here, none of them English. England by no means has a monopoly on analytics history, but it did have a commercial tracking data provider in the 90s, it did supply the mainstream breakthrough for expected goals, it does have the club(s) commonly cited as the sport’s leaders in analytics.

But it’s very exciting seeing this cross-continent collaboration, and it’s very exciting seeing interesting analytics job ads from a widening set of nations. I would love to see more of them outside the historic nations plus America; I would love to see more of them in women’s football.

The ‘globalness’ of the global game has clearly pushed players and coaches to be better. It seems likely that it’ll help push the data side of things to be better too.

There’s an idea I’ve never been able to get my head around, called the ‘Lindy effect’, which (to crib from Wikipedia) “proposes the longer a period something has survived to exist or be used in the present, the longer its remaining life expectancy.” The idea is that if something has been around a long time, the end of it is usually a long way away.

And yet, all things (usually) come to an end.

So it’s strange to see a dominant global force, a force who’s been at the very top of the tree for a long time, floundering at the moment like Manchester City men’s team. Early diagnoses pointed to a single factor, Rodri’s injury, although there are, of course, surrounding factors. Maybe most system failures come down to one central factor, but the magnitude of their impact often seems to depend on how strong the rest of the system is.

Being in the tech side of football gives an interesting perspective on this: large (and/or experienced and conscientious) software companies use ‘chaos engineering’ to test their systems. (Microsoft Azure even has a product called Chaos Studio, which is how my colleagues refer to my presence in meetings). (That is a joke).

Of course, software is an easier type of system to test for failure than others. But you can see the value.

If nothing else, the chaos scenarios can force you to think through problems that you might unconsciously file in the box of ‘not going to happen’. What happens if your data provider suddenly changes the structure of their API; what happens if your Ballon d’Or nominated central midfielder ruptures their ACL.

These ‘what ifs’ might seem annoying, but many systems may seem stable not because of design but because of lack of testing. Rodri’s absence tests parts of the City system in ways, and with a frequency, that they simply hadn’t faced before.

This goes beyond the on-pitch action too. I’ve just finished reading Miguel Delaney’s recent book ‘States of Play: How sportswashing took over football’. In multiple cases, it seems like football’s governance structures were assumed to be reasonable until a moment in time where they were stretched further than it had been imagined they’d be stretched to. Regardless of your takes on those issues, it’s clear that systems (like around multi-club ownership) have been adapted to following new circumstances rather than adapting in anticipation of their potential arrival.

Obviously, the idea of scenario planning exists outside of software and football. But situations where chaos testing feels like it would’ve been helpful keep cropping up in the Premier League. Outside of Manchester City, there’s Manchester United who ended up with a marquee midfielder in Casemiro suddenly (though somewhat predictably) unable to cover the ground a central midfielder needs to cover. Then there’s Tottenham Hotspur, where an injury crisis has combined with a continued commitment to Ange Postecoglou’s pressing with fairly spectacular results.

Maybe this is where the biggest win for tracking data simulation could be. The idea, and practice, of using tracking data as a ‘prediction’ of player movements has been around a long time. Players and coaches don’t have a lot of time on the training ground. Perhaps, looking at their artfully constructed system on paper, the senate of Guardiola’s coaching staff thought that, yes, a combination of Bernardo Silva, Mateo Kovačić, and İlkay Gündoğan would able to cover for Rodri’s responsibilities. But maybe running that through tracking data simulations would have revealed vulnerabilities.

There are far more ‘chaos scenarios’ that you could run, too. Red cards springs to mind, but you could extend this to “what if our key player in X position is having a shocker” or “what if our press is horribly disjointed for no apparent reason”. “What if the front office sign 10 players and no-one knows each others’ movements”. ”What if our expensive new signing was actually a system player and also falls out with the manager”.

The idea of running simulations might seem fanciful to some (if so, why are you reading this, it’s the most Get Goalside idea going); but equally there’ll be some who’ll think that using it for chaos testing would be wasting it. This second group will be wondering whether you could use it to test different tactical set-ups before a match. After all, Formula One teams simulate different race strategies and conditions.

Personally, I would imagine that simulating football matches will be a lot less precise than simulating F1 races, and therefore the results are gonna be fuzzier. Do you trust the system to precisely determine the dynamics at play of using one first-team wide forward over the other? Maybe one day. But giving you the broad strokes of how the team might react to unlikely, but drastic, circumstances could be valuable without needing that precision.

Ultimately, as Manchester City show us, no system is perfect and no system’s robustness is set in stone. (Evolving circumstances, like recently-injured or ageing players, can be a slow boil of fragility). Each system will have trade-offs of goal maximisation and risk avoidance too. But systems shouldn’t necessarily be assumed to be stable just because they haven’t broken yet.

When we’re young, we believe everything is possible. At a certain point we get old, tired, prone to saying “It’s complicated”.

The idea of adjusting a player’s defensive statistics to account for [whatever]… it’s complicated.

After last week’s investigation into ‘possession adjusting’, I’ve taken a look into ‘turnover adjusting’. A write-up with data tables is here, but the long and short is the same as the last one: defensive actions don’t seem linked to possession/turnover share.

Briefly, I should acknowledge an error I made in the code for the original study (now corrected and updated in the GitHub files for the project). It was a bad but ultimately insignificant one, as far as the results go. (Lesson: do basic data exploration steps throughout, particularly if you’re in an unfamiliar dev environment).

Here be numbers

A reminder of what these studies are: using Statsbomb’s open data for the 2015/16 ‘Big Five’ European men’s league seasons, running a simple correlation analysis of player defensive output (per 90 minutes, with different actions analysed separately rather than grouped) compared to a variable representing ‘possession’. Previously this was the team’s share of passes, this time it was the average number of possession sequences that happened per 90 minutes of game-time.

Some headline numbers from this latest bit of analysis, comparing to turnovers. Outside of the small group of Wing Backs, no combination of player position and defensive action had a correlation strength against the turnovers per 90 stronger than +-0.31. Only six out of a possible 45 combinations had a stronger correlation than +-0.2.

I was genuinely surprised by this. I thought there’d be a stronger link between how ‘turnover-y’ gametime was and how many defensive actions players made, given that defensive actions often are turnovers.

Even when performing the correlations for players league-by-league (which controlled a little for the fact that the Premier League had noticeably fewer turnovers than other leagues), there was little of any meaning to draw out.

So what now

Looking through the data in more detail shed some light on this.

La Liga’s correlations are really confusing, with negative correlations for Defensive Midfielders - meaning, a (very weak) relationship where a higher amount of turnovers was linked to a lower amount of defensive actions.

Why? Partly because Rayo Vallecano DMs had a really high amount of turnovers per 90 but weren’t hugely defensive active. Meanwhile players in the Spanish Big Two like Busquets, Mascherano, and Casemiro played in matches with far fewer turnovers but were much more defensively active.

And it was like that across the leagues and across positions. Shift to Premier League centre-backs (my favourite group) and at the low-turnover end you have the incredibly-active Nicolás Otamendí, alongside Otamendí aspirants like Ramiro Funes Mori and Laurent Koscielny , then at the high-turnover end you have the almost lethargic (in statistical approach) Scott Dann and Brede Hangeland.

No, so really now what

At a certain point we get old, tired, prone to low defensive output, and saying “It’s complicated”.

And there are so many complications. The play style of the match affects the statistical output, as do both the relative and absolute quality of the two teams playing it. A player’s role affects their statistical output, and can often be detected through it, but not always. A player performing their role badly may look like a different role entirely, and a player unable to perform their role may be because of them or their teammates or their manager.

Not only that, but weaker teams tend to defend with more players which affects how they can attack after turnovers which also affects how they can defend after losing the ball again. Sergio Busquets putting up ‘midfield destroyer(ish)’ tackle numbers, on a possession-heavy, low-turnover team is a clear sign that football is played on an odd, odd playing field.

Isn’t it exciting?

If you’re new to football data, you may well have missed the introduction of expected goals, the introduction of possession value models, the jump into physics PhDs and graph neural networks that tracking data has brought, and you may well be missing the first steps of body pose data (for something cool on that, see here). Yet there’s still no public consensus about how to take the tackle numbers of a player on one team and say ‘this is how best to compare them to a player on another’.

I still reckon that’s possible.

There are several motivations for this post, but one is to add to the public analysis conversation and show that (hopefully) interesting work doesn’t need to be complex. Small children often ask more interesting questions than adults.

This post is being written just after Christmas, a time of the year when the fridge is full of odds and ends that are all a little imperfect. Everyone will have their own favourite ‘leftover’ recipes, ways to squeeze the most out of the ingredients available.

Football data has an equivalent: defence. The full-bodied enormity of soccer’s contests for possession are carved and trimmed into duels, interceptions, clearances. Unlike expected goals - akin to a centre-piece cut of prime meat - they require ingenuity to work with.

Ideas about adjusting defensive stats began. Ingenious. But not all ingenuity is genius.

It’s now about ten years since I first ventured into ‘possession adjusting’ as a way of tinkering with defensive stats. (Others arrived at the idea independently before and since). I’m now pretty unsure about it.

This is partly a post about some data analysis work, but if you poke it hard enough it’s about the essence of the sport. Can you tweak an ingredient enough to make it a centre-piece on its own or are there insurmountable limits? What is and isn’t separable in football?

And how do you recognise ingenuity?

Data tables are in the eye of the beholder

Code for the project can be found here: it goes without saying that I would appreciate mistakes to be ruthlessly hunted down.

The theory behind ‘possession adjusting’ is simple. A team having possession of the ball (or not) affects what their players can do. So you adjust a player’s stats based on their team’s share of possession. In one sense, it’s the same theory as averaging a player’s stats per 90 minutes - that things need to be adjusted to make them justly comparable.

However, an update to work I did a few years ago has given me the same result as previously, one that doesn’t mesh well with possession adjusting.

We’ll see the data below, but there’s a lack of a clear link between possession share and player defensive actions output. This is not what the theory behind possession adjusting would expect.

Below are three tables that you can glance at, one for each of three positions: centre-back, full-back, and defensive midfielder. They’ll show three things, which were also features of the previous work:

1. The link between possession share and defensive actions is small if existent at all

2. The link varies by position

3. The link varies by defensive action type

The tables are ordered by the absolute strength of the correlation for each position. In none of them does the strongest correlation hit +-0.2. (A little note about the data follows them, as a reward for getting through the numbers).

Centre-backs

Full-back

Defensive Midfielders

The data being used here is the Statsbomb open dataset for the 2015/16 seasons across the men’s ‘Big Five’ European leagues. (that is, the English Premier League, French Ligue 1, German Bundesliga, Italian Serie A, and Spanish La Liga). It’s accessed via the kloppy Python package. (I’m slightly wary of discrepancies this might cause compared to working directly with the event data, but I don’t think the trends would change).

The defensive action types are ones found in Statsbomb’s dataset, although the definitions are relatively common to event data. The player positions are also from the Statsbomb data, and only feature players who played 450+ minutes in those positions - the code for this project can be found here and a full table of results here.

In fairness to the concept of possession adjusting, those three tables aren’t the entire story.

There are higher correlation strengths among other positions, although none greater than +-0.35 in Attacking Midfielders, Center Forwards, or Wide Midfielders (and most not greater than +-0.25). The Winger position gives the greatest support for the possession adjusting principle (Wing Back correlations are stronger, but a tiny group):

Wingers

There is more that can (and will, soon) be said about the gap between these figures and those for the other, more defensive, positions, but one thing is clear. This marks a blow against the principle of a uniform possession adjustment of defensive statistics.

More than meets the eye

There are a lot of interesting little avenues that can be found in this data. It seems curious, for example, that positions known most for their defending - central defenders, full-backs, and defensive midfielders - appear to be affected least by their team’s share of possession.

Defending is a team game though. It’s not unusual for ‘forwards’ on weaker teams to be more concerned by defending than attacking, while their counterparts on stronger teams are sometimes given a bit of a pass from defensive duties. Although I suspect that the Winger position correlations might look different if some lower possession 4-3-3s were classed as 4-5-1s, it seems true to my ‘viewing experience’ for possession share to have a more discernable effect in ‘forwards’ than ‘defenders’.

(An aside, although it’s covered in the code repo for this project - as Statsbomb’s event data has possession sequence IDs included with it, it’s fairly straightforward to identify when players were on-pitch and use the sequence IDs as a quick reference for events which happened while they were on-field).

This is where the fun begins

If you’re only interested in the data tables, you can safely close the email now. If you’re interested in the fine line between wonderment and futility in your numerical analysis, continue.

Let’s duck back to my theory that correlation strengths for Wingers are partly affected by the formation assigned by Statsbomb. My (unverified) assumption is that in a Statsbomb 4-3-3 formation, the wide attackers are designated as Wingers, and in some other formations - like a 4-5-1 - they’re designated as Wide Midfielders (whose correlation strengths looked much more like the Center Forwards, with a mean average correlation of the six defensive actions of -0.166 and -0.186 respectively - again, full results can be found here).

Assigning a formation to a team is notoriously difficult. It’s a subject I’ve written about before, a subject that even data-reticent teams complain to data providers about, and, as a result, an area which use of tracking data has tried to solve.

Although football punditry commonly talks about a team’s ‘in-possession’ and ‘out of possession’ shape, my conclusion in the post linked above (which I still stand by) is that the ‘formation’ a team is assigned by fans, media, and data providers is the one which they ‘see’ most often. It’s a shorthand for the roles that the players within it are playing.

When you come to the division between what the shorthand ‘4-3-3’ indicates and what the shorthand ‘4-5-1’ indicates, the difference is mainly in the defensive roles of the wide attack/midfield players.

If the switch between assigning the two formations was systematically done based on the ‘defensiveness’ of a team (for which possession share might be a plausible proxy), then this would just be a quirk of data. However, if teams are sometimes but not systematically assigned to ‘4-5-1’ or ‘4-3-3’ based on their possession share, then that might be a confounding factor in this particular type of data analysis.

(In fact, in the dataset, Wide Midfielders and Wingers in this study had the lowest (44.9%) and highest (53%) median possession shares respectively).

Tying the threads of the Get Goalside cinematic universe

As I said before, the concept of possession adjustment is sort of similar to the ubiquitous concept of ‘per 90’-ing statistics. We recognise that there are ways that data isn’t ‘comparable’, and we think we can address that.

No method is perfect though (even ‘per 90’-ing gets questioned).

Possession adjusting supposed that a team having the ball affected the defensive output of players. (In reality, many defensive actions are ends of possession sequences, and so wouldn’t be affected by the duration between these sequence endings).

You could head in the direction of a different sort of possession adjusting. Instead of adjusting based on the possession share, you could adjust based on the count of possessions/possession sequences. Given that defensive actions often end sequences, you might reasonably expect that players involved in ‘pinball’ matches will have higher figures than those without. (In fact, there was a slight indication of this in my earlier work, though I haven’t tried replicating that with this 2015/16 dataset yet).

Unfortunately, football throws another spanner in the stew. There is an obvious stylistic difference between teams who play ‘high turnover’ matches and those that don’t, and there is likely to be a quality difference between these teams too. Some previous Get Goalside work with Statsbomb open data has shown a link between ‘ball-in-play time’ and breaks in play, and a suggestion that high-possession (Guardiola-type) teams have more in-play time because of this link with breaks in play. The relative quality of teams (and possibly the absolute quality) appears to have an impact on ‘style’ whichever way you slice things.

I suspect that this is inevitable. I’ve previously tried to distil football down into teams competing over space control and ball control, and everything else springing from there. If a team has the quality to do so, it will naturally want to have more control of the ball, the only method of scoring. The hugely imbalanced value of space on a football pitch - extraordinarily heavily concentrated around either goal - steers the (weaker) defensive team’s strategy. With such a dynamically tilted field, is full comparability through adjustment of statistics possible?

So what now?

According to my own memory of my views (a trustworthy record if ever there was one), I have two longstanding opinions about data.

One is that a large part of the craft of data analysis is tuning an internal ‘fuzz-meter’, the sense of when things are ‘close enough’, ‘probably legit’, etc. I am wary of false precision.

The second - perhaps paradoxically - is that measuring discrete, specific skills is very valuable. It’s just that most of event data analysis doesn’t do that.

Take a statistic that is pretty clearly useful, expected goals. Reliably getting a large amount of xG isn’t ‘a skill’ as such; it’s an indication, or a symptom, of the various skills that a top striker has.

To go back to the (far worthier) subject of defending, there are a variety of skills that you’re looking for in a defensive player. It turns out that these skills aren’t necessarily observable purely through stats like tackles and interceptions. And it turns out that they’re probably not observable through possession-adjusted versions of these stats either.

Other types of data might let you measure those specific skills (whether that’s tracking data or other types of event data).

There are some in this world who just want to create single number models.

Maybe you can create steak out of scraps.

That would be ingenious.

Howdy. Because blogs are back, I’ve been reading some blogs.

As a quick festive gift, here are some nice links:

Football:

‘Uncovering runs: Jack Grealish in 2023 Champions League final’ - Henrik Schjøth

One of the great types of blog: pretty short, packed with ideas. One of the main ideas is looking at the variety in direction of players’ runs (there’s a neat comparison between Erling Haaland and Romelu Lukaku, for example). There’s also a couple of neat visualisations showing the attacking runs during a spell of possession, highlighting two City players making runs into the box - I can imagine this a springboard for an analysis on ‘box runs during crossing situations’.

‘Football data: finally under scrutiny?’ - Arnaud Santin

I’m a sucker for this kind of blog, a commentary on different types of data and their history. More importantly, I’m not sure how easy it is for users of football data to appreciate the differences between providers (and between the provision that different competitions get, from the same provider).

I’m not a “there is a single ‘truth’” kind of guy (as I’ve written about before: “If categories can't be 'right', then definitions can't either, only sensible.”). But, obviously, some interpretations are more reasonable than others. Santin, who is cofounder of a company SportsDynamics, writes of an example:

‘Taylor Swift, the future of fandom and a dilemma facing women’s football’ - Katie Whyatt

ok, not a ‘blog’, but if this headline doesn’t grab you I promise you it’s (at least) five times more interesting than you think. Although this piece isn’t directly linked to ‘football analytics’, it deals with the ‘what assumptions are made about fans’ questions have clear parallels in the types of things analytics nerds think about. The piece is also great for giving a subject that can get flattened to primary colour analysis the consideration from multiple angles that it merits.

Data nerds may also enjoy related a conversation in this episode of the Expected Goals podcast (about halfway through) about the data around whether someone has bought a “women’s team” shirt or a “men’s team” shirt.

Not football:

‘The beginner data science programming book that doesn’t exist (yet?)’ - Tan Ho

A set of thoughts that resonate with me, both from when I was first learning to code and when I’m learning something technical that is new. There’s probably a lot of ‘what is teaching/what is learning’ commentary that you could do relating to this as well.

‘Write code with your Alphabet Radio on’ - Vicki Boykis

A tl;dr - experienced programmers can have an advantage over LLM-code tools because we’ve worked with good programmers. Though, as Boykis references, data scientists and programmers don’t always have teams around them (which particularly applies to sports).

‘LLMs, reliability & the scientific process’ - Martina Pugliese

Interesting, readable blog on a subject I think a lot of people will be interested in. Like, I like this nugget: “You usually test their [LLMs] behaviour on a bunch of data. […] You make this dataset into an eval set which you run periodically to check for robustness. But the fact that results may look good doesn’t tell you how the LLM will behave at scale.” The last couple of years have seen best practices start to emerge for LLM-based tools (like code editors); will data science testing best practices follow?

If this is a festive time for you, have a happy festive period; if not, have a good time anyway

Yes, to take Oscar Isaac’s memorable line from Star Wars: The Rise of Skywalker: “Somehow analytics blogging returned.” (or, if country star Lainey Wilson is more your speed: “Doggone, dadgum it, didn't see that coming / Blogging’s cool again”)

Let’s get the links up top, then share some data, and finally do some ‘takeaways’.

A brief aside: One of the writing skills I most admire is drawing together the different contextual threads that contribute to a particular story. Grace Robertson is excellent at this, e.g. a piece earlier this year on how Positional Play turned defensive. In that vein, I’d recommend her recent article on this year’s Rainbow Laces campaign.

The links

• Devin Pleuler, Senior Director of R&D at Maple Leaf Sports & Entertainment (a.k.a. ‘the Toronto sports teams you know (except the Blue Jays)’) has been admirably prolific with Central Winger over the past few weeks

• the folks at KU Leuven’s Sports Analytics Lab are doing a series on design decisions for possession value models

• Houston Dynamo’s Head of Analysis Carlon Carpenter wrote about movements in the final third, using Skillcorner data for the Austrian Bundesliga

• analytics consultant Joris Bekkers just wrote up some work he presented last year on measuring pressing intensity (at a Skillcorner event, by coincidence)

• Over at American Soccer Analysis, Eliot McKinley’s put together a detailed look at the impact of capping ‘minutes played’ at 90 (a pet niche for Get Goalside, given that it touches on 1) added time decisions 2) definitions of statistics by a dominant company that have an outsized influence on the public stats understanding)

• Michael Caley’s Expecting Goals is back from US election vote analysis with a dissection of a different type of symbol of statehood, the recent bad form of Manchester City’s men’s team

• Paul Johnson with a blog on using multilevel modelling and football finances (which is slightly less recent than other things in this list but he was on the Double Pivot podcast a few days ago chatting with aforementioned Caley)

Those are recent beginnings or returns, joining things like:

• Ben Wylie’s Plot the Ball, which recently wrote about Barcelona’s use of young players.

• Scouted Football, whose output is becoming more and more delightfully statty with the recent additions of Skillcorner data (them again) and Jake Entwistle (previously of Squawka)

Send along any more!

On top of this, it’s thematically appropriate that Hudl (who now own Statsbomb) have just released some fresh data to the public. (provided you fill in one of those ‘is this really necessary’ data collection forms to do so).

This is a lot of fun. Having a ‘to read’ pile that starts tipping over and collapsing under its virtual weight is a nice change.

I asked Devin Pleuler, who’d written under the Central Winger title on MLSSoccer.com in the early 2010s*, why re-start now? “The primary [reason] is that I enjoy it! I find writing to be the best way to organize my thoughts and opinions and effective communication remains the most important part of sports analytics.

“But also, with the shift away from Twitter, it felt like the right time to reengage with the community that has atrophied over the last few years. My belief is that teams are too secretive and perhaps this can motivate others to contribute to the conversation.”

*some of it is still online! Remember Chivas USA?

Where is the data…

If this has motivated you to join in, you might be wondering where to get some data.

Outside of the recently-released J1 data, Statsbomb has a trove of publicly-available event data. There’s Champions League finals, one-off team seasons like for the 2003/04 Arsenal Invincibles, and entire league seasons for a selection of top European leagues (and more). They have a Python and R package to help get started loading the data.

If you want to try tracking data, there are a handful of matches from Metrica Sports, Skillcorner, and PFF. To be frank, these aren’t as polished as datasets as the Statsbomb event data, but at least the Kloppy Python package can help out a bit.

[online edit: after publishing, Kloppy released an update of the package, including access to some Sportec event and tracking data!]

Get in touch if you know any other data providers who have data available to the public.

If you don’t/don’t want to code, I’d just recommend scanning through FBRef. You can get a free trial of Stathead to access the database in different ways. And, although it’s a bit of a pain, you can just copy and paste tables into Excel. (My first stat explorations, many years ago, were based on flicking between one of WhoScored, Squawka, or the StatsZone app and a spreadsheet).

Beyond this, there are a range of ‘community’ data access sources. If you want to scrape data, or access scraped data, it is not too difficult to find resources. I’m retiscent to point directly to them because 1) I don’t need to use them anymore 2) unless you have a clear and specific use, you don’t need to.

If you want to plot some on-pitch data for shot maps etc, mplsoccer is good for Python, ggsoccer was my go-to when I used to use R (I still miss ggplot), and d3-soccer is good for the JavaScript hive.

Again, lemme know of any other similar packages in these or other coding languages.

What do all these new blogs teach us?

If I were starting (or re-starting) analytics blogging again after a time away from it, some neat notes of interest from the list of links earlier:

Time: In the VAR age, using full minutes rather than ‘capped at 90’ minutes matters for players (as McKinley demonstrates). When dealing with in/out-of-possession data, ‘per 30’ seems a neat standard (as Carpenter demonstrates).

Data science can be fascinating: I overheard someone at October’s StatsBomb conference say that you can bank on KU Leuven Sports Analytics Lab giving you strong work - the series linked above, taken from a master’s thesis by Lode Van Tente, is just that. In a way it’s as simple as switching one variable in a model and writing up the results - but both the reasoning and the write-up are excellently clear.

Paul Johnson’s post is also a nice example of the ‘modelling technique applied to football’ genre (a field which can otherwise produce some, to be honest, unreadable work). It benefits from focusing on explaining the technique, and not over-promising what its application means for the results.

Remixing: The KU Leuven series is, notably, riffing on methodologies that existing possession value models use; part of Joris Bekkers’s post is based on an aspect of Will Spearman’s Pitch Control model; Devin Pleuler has a whole post about data science tricks he’s learnt from others. You can get a lot of mileage out of taking an idea of someone else’s that you like and spinning it into your own area of interest.

Learning and thinking: There’s a three-minute segment of a lecture that I like to share, by Larry McEnerney, the director of the University of Chicago’s Writing Program. His aim, in that section, is to reinforce to the grad students taking the program how different it is to write for a general audience than the academic one they’re used to. But in doing that, in being specific about the way they currently write, he says:

“To help yourselves do your thinking, you have to do your writing. You have to do this, because the stuff you’re thinking about is too damn complicated to just do it in your head.”

And this is one of the things Devin Pleuler said earlier. “I find writing to be the best way to organize my thoughts and opinions,” he said.

This doesn’t need to be what all your writing is, or all your thinking; neither does it need to be on a regular schedule. There are seven bullet points in this newsletter’s first list of links, and six of those are one-offs or irregularly scheduled. And they were all still a joy to have.

It’s a matter of time before someone releases an ‘AI assistant coach’. I imagine that, when they do, it will follow the common AI hype cycle:

• product (or, more likely, product beta) released

• interface is engaging enough that normies can use it for jokes

• the ‘LinkedIn Apex’: declarations of a game-changer for the world

• within two weeks, usage drops 80%

Despite this, generative AI systems are here to stay, in one form or another. Why? Because they’re getting quite good. As someone who spends a lot of time in code editors, I can attest to that.

And, attesting to it, there are some features that guide us to what football-based systems might be like when they inevitably (but not necessarily imminently) come.

How fast the road takes us, I don’t know. But I think there are three strands to the direction of travel: tools, context, and accessible content.

Tools for tools

There’s a small part of the generative AI industry who really believe, or want investors to believe, that large language models can become superintelligent on their own. But we don’t really need that - in either a practical or dystopian sense. “Agents being able to use software is how AI becomes more general,” Amjad Masad, the CEO of a very popular company called Replit, said recently.

These models, increasingly good at interacting with human languages and ‘non-human’ languages like code and APIs, can be hooks into things. And so your football LLM systems don’t need to be trained into being expert analysts. They just need to be good enough to use the tools at their disposal.

For example, some speculative fiction: “How well does this full-back defend against overlaps?” could be turned, by an LLM-underpinned tool, into a series of requests to other tools. Maybe you have the data for ‘overlaps faced’ already available, in which case gathering it is the first step. But maybe not, in which case a data science process can be kicked off (along lines of previous work). After that, another process, gathering data about how often corners, shots, and goals are conceded shortly after the defender comes up against an overlap. Another process might grab and edit clips.

(This feels like it’ll have implications for how different services can be used alongside each other, if ‘interfaces’ are going to be more geared to code scripts and API calls than human users. But that’s for another time).

The point is, a ‘generative AI world’ in football won’t need the LLM to be a football expert; the ecosystem of football technology will do a lot of the heavy lifting.

Context is king

Context is fun. This is objectively true, because mistaken context or innuendo makes up 90% of Shakespearean comedy. But it’s not fun if your LLM is missing it.

Code assistance tools are settling on ways to help this. For in-editor helpers, users can add files as context to their question - for example, you could ask it to write some data parsing functions based on the schema files you already have.

For football, the big and slightly hypey way this might be useful is context of terminology and game model. If we go back to the ‘overlap’ example from before, a club or coach might implicitly mean ‘overlaps in the final third’, calling similar movements elsewhere on the pitch something else. Or, a slightly more realistic example, terms for types of press or phases of in-possession play or player roles.

Beyond football, but perhaps in football more than elsewhere, I think LLM systems will live and die on understanding of context. It takes far longer to learn how to use a genAI helper if everything you ask it needs to include all relevant context. It’s like talking to a particularly stubborn child who’s decided to only follow instructions very literally. Or like working with someone in Quality Assurance.

But…

Content is king too

If context is important, that means the system needs to be able to access it. And that means that more of a coach’s/club’s work needs to be on a system that plays ball with an LLM-tool.

Think about training sessions. The majority of a coach’s work is done outside of a matchday, which has been the traditional bulk of data availability. They may well have a bank of reference video clips, but a lot of knowledge might come from conversations with the rest of the staff and exist in their heads, or on paper.

Think, as well, about coaches changing jobs. Currently, the clubs are the ones who buy software - but if part of what they’re hiring in a head coach is that coach’s methodology, it’s that coach’s ‘data’ that is the most important to access. Head coaches (at the top level where it can be afforded) often want to bring assistants along with them into a new role - would an LLM-based assistant be the same?

And, of course, if we’re talking about a coach’s knowledge as ‘data’, there’s the data ownership question. If LLM-based systems are going to be relying on coaching knowledge, coaches have got to make sure they can take this with them and use it in future roles. (The same is presumably already true for scouts, whose reports have long been logged into centralised knowledge banks).

The ‘pool of knowledge’ problem

Yes, LLMs spit out text that is incorrect. But it’s amazing that they work as well as they do, pretty reliable conversation-bots created by probability.

Of course, they’re more likely to get things wrong when the data isn’t there to produce ‘good’ probability estimates of the next word in a sentence. When coding, that’ll often happen when using a small-usage package, or non-mainstream language, or just a quite specific type of problem.

Fortunately, football is the biggest sport on the planet; and if you’re reading this newsletter in its original state, then your command of English will help avoid potential disadvantages of LLMs in other languages too. Purely keeping my British Isles locality in mind, it’d be interesting to know how good LLM systems are in Welsh, and Scots and Irish Gaelic.

All that being said, while I’ve made the point that it’s the football software ecosystem that’ll do most work, maybe this will be genuinely harder to achieve outside of the major languages and/or in more niche areas of the game. Will off-the-shelf LLMs fit into ecosystems where practitioners want to organise, say, periodisation and training microcycles in Japanese? (sidenote: I’m interested in how this would interact with things like the dialectic variation in Arabic too).

The end

You could probably go a long way without specialist software.

For text notes, tools like Notion have AI helpers which can search your notes and files elsewhere, like in Google Drive. The issues with domain knowledge of LLMs would come in here, but if you’re a low-budget club then you could probably get value out of using something like it as a repository for your coaching or scouting notes.

If you are the type of coach or analyst who collects video clips, and these clips are consistently labelled, an LLM-inflected search tool might help you find relevant clips easier, without you having to act as a folder-expert organiser. But maybe this will be even easier than that: given that tools can integrate with Google Drive, perhaps you could keep a your clips there alongside a spreadsheet of names, tags, and links (essentially metadata) and use that as the entry point for an LLM-searchbot.

 Get Goalside has written about what we mean by the term ‘analytics’ before, and whether ‘football tech’ counts or is stretching the definition too far. But, regardless of your opinion on that, the end of that piece is still relevant here:

Maybe this is one of those edges.

A couple of months ago, noted Ralf Rangnick sceptic Michael Cox questioned whether data had made a dint on football’s on-pitch tactics.

I found this interesting. As I wrote shortly afterwards, maybe that’s right and, what’s more, maybe there’s a clear direction that cold, calculating data would push things.

In fact, maybe it already has.

I had a theory - and if you’ve taken enough notice of this post’s title, you’ll know it too - and I started listing things that played into it. More of the list than I’d expected were already on-pitch trends.

Across the board, I suspect that data analysis would reveal more risk-taking is better. Ian Graham, Liverpool’s former director of research, tells an anecdote in his recent book, How To Win The Premier League, about Brentford owner Matthew Benham. I used it in the previous piece - it’s about advising Brentford to Always Be Attacking - but I didn’t use the best line from it:

In other words, coaches are so affected by risk-avoidance that you need a caricature-level opinion just to drag their Overton window towards the optimal strategy. (Or so Benham believes).

After this had marinated in the back of my mind a little, my little grey brain cells made a link to goalkeeper cross-claiming. Analyst and data scientist John Harrison has previously published some work around shot prevention by claiming crosses: measuring how active and accurate claimers can save their team potential goals before shots even happen. And I remembered that David Raya (at data-savvy Arsenal) is particularly prolific in this department. (In fact, Harrison’s models have suggested that shot-prevention was a key part in Raya being an upgrade on Aaron Ramsdale’s performances in the men’s team).

Of course, we can’t talk ‘risk’ in goalkeepering without mentioning Manuel Neuer’s sweepering antics. Neuer didn’t invent keeper-sweepings by any means, but he’s a big reference point, and the position of goalkeepers in general play is a key point of analysis nowadays. Neuer was such an influence because of how successful he was.

Where else can teams play with more risk? High defensive lines, trying to keep the ball when under pressure, more take-ons, working the ball into better shooting locations, playing an attacking style even when leading, substituting as soon as necessary. (Any others?)

A lot of these are pretty familiar to modern football, even if analytics hasn’t driven take-up.

This doesn’t mean that these risk factors are a one-way street though. A push for high lines opens up a new risk-taking option for the in-possession teams: adventurous passes in behind the defensive line. But perhaps the highest line achievable is always the optimum.

At Amazon - a company even more successful than Matthew Benham’s Smartodds - they have a leadership principle called ‘bias for action’. The two beliefs which underpin this are that speed is extremely important to businesses, and that many decisions can be reversed if they turn out to be incorrect.

Risk avoidance serves an important purpose in business as in life as in football. But maybe the most effective path is to be poked a little bit more into the path of bravery.

This is the age of unstructured data. There’s your pull-quote. In the real world, ‘unstructured data’ often means ‘words’, and part of the large language model hype is that they can provide an interface for this. In football, the more important world, the big mass of unstructured data would be video.

What we call ‘event data’ - the shots, the passes - is just a ‘structured data’ way of viewing matches. But the game tape - yes, plus wordy things like scout reports or coach analysis - is unstructured.

For all the benefits of structured data, it takes time to structure, into a system that you will inevitably hate when it’s just too late to change it. (something something data engineering, too). So why not just squeeze the juice straight out of the unstructured lemon?

Unstructured lemon juice like goalkeeper save technique from video, or goalkeeper save timing from video. (Goalkeepers don’t get enough attention, but are also on camera in just the right way for body pose detection). If outfielders are more your cup of tea, here’s a 2019 paper on body orientation from video.

If we’re being picky about metaphors, video and LLMs aren’t equivalents. Video and words aren’t even the direct equivalent. In football, tracking data is a form of unstructured data that has been around for a long time; working with the video directly is like working with audio. But the point remains the same: unstructured data.

However, to leverage this unstructured data you first need the game film. As Nancy Hensley pointed out at the recent Hudl Statsbomb conference, the lower level (in quality and existence) of women’s football coverage on TV affects not just fan engagement but data collection. In Belgium, they’re making a big play of installing cameras and getting data on everything that moves.

As alluded to above, leagues centralising technical advancement is something that makes a lot of sense, not to dictate usage (which would likely stunt innovation) but to set a minimum standard. Although it does need to be a reasonable minimum standard. From Ian Graham’s How to Win the Premier League: “We also received tracking data for all UEFA games, but until 2021 UEFA did not exercise any quality control over it, so we could not trust it.”

(Related reading recommendation, The Formula by Joshua Robinson and Jonathan Clegg: “After nearly seven decades of [Formula One] teams fighting tooth and nail for every advantage […] Liberty presented them with a new reality. Instead of being rivals, these teams had to understand once and for all that they were all in business with each other.”)

The ‘pivot to unstructured data’ creates another interesting dynamic.

For a long time, access to data has been an issue for wannabe analysts or researchers. In their lifetime to date as a data provider, Statsbomb have been admirable in the amount they’ve made openly available. But if DIY collection from video takes off (not necessarily for tracking data - you could imagine someone taking this and deciding to create a shot-detection system), that would open interesting doors.

But to where?

Well, if video/tracking data is a rough equivalent of words and text analysis, maybe the recent use of generative AI can give some pointers.

There are two indisputably ‘successful’ use cases for genAI. One is coding assistants (for the more popular languages); another, though ‘successful’ is a loaded term, is art. Now, the development of lucrative tools built on scraped art, made by the profession who’ll be undercut by said tools, is the type of societal wrinkle you’d find in a dystopian novel. But these tools - think of Photoshop’s generative fill feature rather than entire artworks if it helps - produce convincing results. In the hands of artists, they produce art. In the hands of schmucks, they don’t. A lot of early genAI ‘art’ production was schmuckery.

These two use cases make sense when you think of how these kinds of generative AI work systems work. ‘Art’ does not follow the same rules of ‘factual accuracy’ that so much of the rest of the world does (photorealism, and adherence to specific styles, aside). Where else but art could Caravaggio and Kahlo, Rothko and Ruysch exist as greats. Certainly not business chatbots. Coding, meanwhile, has a much stricter sense of ‘accuracy’ but a far more limited ‘vocabulary’. LLMs work by predicting the probability of the next word; English has an estimated 170,000 words, coding languages will have far fewer (hats off to the Reddit user who asked this question a few years ago).

So, to football.

At some point in the past year, I heard someone with a lot of experience (on a podcast episode (I think) that I now can’t find) give a warning about tracking data. They said that it’s tempting to go after the gold mines of off-ball metrics, but that that could be a red herring. A marsh that one would sink into.

I’m embellishing slightly, but I think the tendency of work using tracking data to split player movement into ‘runs’ is telling. Some of it is to create physical metrics, some of it is around concepts like ‘running in behind’ or ‘overlapping runs’: clear concepts, clear vocabulary, turning unstructured data into well-understood structured datapoints.

But there’s so much that isn’t well understood, or well verbalised, in football. Leander Forcher recently released his PhD dissertation on ‘success factors in soccer defense’, much of which highlighted the lack of pre-existing work in defensive analysis. Lack of pre-existing work often means lack of clear understanding of terms. (‘what if we’d focused on different parts of the sport’ is a continuing theme of Get Goalside)

We’re in the age of unstructured data. We’re also in the toddling age of learning how to get the best use out of unstructured data.

The only thing that’s certain is that we’ll need more data engineers.

Which is harder to do well, asking questions or answering them?

At some points a few months ago, I was speaking to someone who works with physical data. A few years ago, they’d have had to be selling stakeholders on the answers th data could give them. Now, everyone eats it up. But the person was saying ‘I say to people, do you know why you want this data? What do you actually want to use it for?’.

It’s a good question to ask.

Ultimately, though, there is only one question that anyone in sport has: “how can I win more?”. And so you need to play a little game of trade-offs, narrowing the broad focus onto something narrower, a tangible area, a specific factor.

This is why this year’s Hudl-Statsbomb conference (no more capitalised ‘B’ for ‘Bomb’) steered its research competition entrants towards ‘trade-offs’ as a theme. There was one on the value of a lesser-quality left-footed left-sided defender vs a better-quality right-footer; one on the value of booting it and pressing a throw-in instead of trying to play out of pressure. (The 2024 conference research papers can be found here)

It’s not just this year’s HudStatconf papers where the strength of the question shines through. Frequently, the thing that strikes me in analytics ‘research’ work is the clarity of the question. It’s there in some of my favourite research papers (who doesn’t have favourites), like fellow 2022 Sloan conference appearances ‘Learning from the pros: extracting professional goalkeeper technique from broadcast footage’ and ‘Beyond action value: a deep reinforcement learning framework for optimising player decisions in soccer’. It’s there in my favourite broad genre of work: projects which didn’t find what they set out to find, but which were conceived clearly enough that the journey was worth it and the direction of future investigations is clear.

It’s a craft.

For example: Pep Guardiola has referred to formation notation as ‘telephone numbers’, Emma Hayes has said discussion around formations is ‘archaic’ (around 9:20) - but we talk about ‘formations’ for a reason, as a convenient shorthand. So there will be some cases where splitting data by ‘formation’, or where methods of determining ‘formation’, will make more sense than others. (If you need in-the-moment formation/shape information, then tracking data or video analysis may be your only viable options, but build-up patterns might be gleaned from event data)

But this whole thing about ‘what is your actual question’ is not just true of research.

Three years ago to the day (as this is being written), I wrote an overview of where you might spend money set aside for ‘analytics’. Quite frankly, I’d forgotten I’d written it. Let’s have a look at the end conclusion:

There are two biases I have here:

1. I work for a company you’d class as a ‘third party’, which one may want to make strategic use of (bosses and colleagues at Twenty3 Towers claim to read, so I better link the company website)

2. It feels nice when things you wrote years ago still hold up

That said, I want to highlight two specific parts of this extract. “There’s no point in the exclusivity benefits […] if you don’t make sure you still have it if someone leaves.”; apply this to your internally-created tools as well as report formats and research projects.

If you think players are the only things whose performance can drop off a cliff due to getting old, you might wanna google ‘tech debt’. Several months ago, I quoted Charlie Marshall of the European Clubs Association: “There are so, so many [clubs] and the vast majority of them are quite small businesses.” If you’re a small business — a community events business, really — do you want to also be a software company? Why?

The other particularly important part of the extract is “strategic use” of third-parties. A couple of paragraphs later in that 2021 piece, I wrote:

(I neglected to mention data engineering, and I apologise to the gods of cloud computing for this)

The problem with all that, of course, is that following this advice may mean hiring a sufficiently capable head of data to then spend money on a data provider as well as then spending further money on outside services. Maybe that won’t play well with bosses who expect some guy (often a guy) with a quantitative degree to get things going within a month or two.

But, look at Bayer Leverkusen. Thankfully not too tight-lipped in what they allow to be shared on LinkedIn, they’re on the roster of Kitman Labs, Catapult Matchtracker, MyGamePlan, and SportsDynamics, and that’s only the software platform partnership announcements I could easily find.

It may well be that Leverkusen’s eventual aim is for all of that tech to be internally created and managed (as one of the ‘Analytics in the US’ panellists at the Hudl-Statsbomb conference put forward as a general truth). It may well be that they’re not even using those platforms (which would be strange, but ‘pays for an unused subscription’ is hardly a novel situation).

But, while the best research papers ask a tight, well-framed question, the best clubs will be doing the same.

A couple of weeks ago, I wondered whether analytics should have made football be played in a radically different way to how it currently is. A recent episode of the Double Pivot podcast reminded me of an oversight: football might not have a three-point line for goals (and an associated NBA-type change in shooting behaviour) but it does have a three-point rule for points.

The change was designed to improve the incentives for teams, after a period of doldrum-esque scoring in 1970s England. In the start of that decade, the English men’s top division saw four seasons in a row with an average goals per game of around 2.5 or under. This was a huge fall from a time in the late ‘50s and early ‘60s when the average could be north of 3.5 per game.

Goalscoring alone probably isn’t the best metric to judge the rule change on, but the top division’s scoring rate picked up before the change came into effect in England in 1981. It stabilised at around 2.6 per game, until a hop upwards in the 2010s. The 2023/24 season saw the first average of over 3.0 goals per game since the ‘60s.

But this isn’t about goals.

Twitter tweet

The above is something of a riff on a theme:

Twitter tweet

Even at this high level - the level supposed to be about long-term thinking - it seems long-termism can be more theory than practice. Possibly because the incentives at play are job security at an individual level and, increasingly if not wholly, investment value at an ownership level. Regular readers of Get Goalside will identify a bias here, but ‘driving wedges between clubs and their long-standing communities removes an important incentive to generational long-term thinking’ does not sound an unreasonable hypothesis.

But this isn’t about club ownership per se either.

By coincidence, the past few days have thrown together these reminders of incentive structures with something else. On YouTube, a glimpse into Burnley’s data-gathering operation on their academy footballers, and on social media, a post about high-school American Football tracking data. Add these to the consumer products selling the dream of being scouted from afar, and soon we will be telling uncomprehending children of the days when we played sport and left behind no record other than the score. What do we do when there are no datapoints left to conquer? And what of the data already being coerced into databases?

There will, or should, of course, be data protection policies in place. But though Get Goalside is pretty pro-data, the push for player data rights is important. And it should extend to youth football. The power of data is often in its quantity, which leaves individuals - in successful data-gathering operations - as mere specks in the whole. Valuable as part of an aggregate, inconsequential on their own terms. There are football ownership models where groups will buy clubs, or large stakes in them, as a means to gaining a foothold in that continent’s ‘talent development market’ - how passé. A scouting network is such a twentieth century concept. Cycles of age group teams entering one’s data warehouses, training talent identification and development plan algorithms: that’s an investable future.

As the saying goes, football is the most important of the unimportant things. The latter half of this may be a good thing.

By coincidence, modern football coaching mirrors the now-famous line from James Clear’s book Atomic Habits: “you do not rise to the level of your goals, you fall to the level of your systems.” In football more broadly, what systems are those to fall back to? Do we trust the incentive structures in place?

Nick De Marco KC, barrister (on Twitter, responding to a hearing on Premier League rules around ‘associated party transactions’): “All that I can say is we are living in the most exciting time for sports law. I have never myself been one to celebrate the greater commercialisation and therefore legalisation of sport and its regulation, but it is a real fact of life and economic activity, such that this tendency for greater scrutiny of sports regulation is inevitable.”

Everyone knows the problems and the buzzwords. The theory is this: Teams are difficult to organise; individual magic is great but usually creates inefficiencies; a structure, aligned with meaningful goals, speeds up a team to being good.

Am I describing Positional Play or Agile project management?

For those unfamiliar with the term:

Agile is a collection of software development practices that built on existing ideas, grouped under an umbrella term, with courses and how-to books springing up for this now-a-proper-noun methodology; it is sometimes criticised as a management fad that is applied in cases it needn’t be.

Positional Play is a collection of football tactical practices that built on existing ideas, grouped under an umbrella term, with courses and how-to books springing up for this now-a-proper-noun methodology; it is sometimes criticised as a management fad that is applied in cases it needn’t be.

Good, we’re all caught up.

Now, don’t close your inbox if you’re not in software. Everywhere that there’s management, there’ll be theories and snags where implementation hits the road. Government, supermarkets, building sites, military theory. (Google ‘Sun Tzu business book’, you could start a library).

My point is that it’s easier to talk about football tactical approaches when you think about them in this way. The terms are more like genres than rubrics. Or schools of thought, within which there can be disagreements and room for leeway.

There’s an ‘analytics’ point we’re strolling towards, this isn’t completely a newsletter of amateur HBR-ing. First, though, we need to visit Munich.

Vincent Kompany, the current Bayern men’s team coach, has an unusual management career path. From Anderlecht to Burnley, with a promotion and relegation, and then to Germany’s dominant club. The style of play at Burnley translated well to their position as top team in the English Championship, very badly to their position as one of the worst teams in the Premier League, but, crucially, quite well to mega-teams like Bayern.

If we go back to the business management metaphor, what works for multinational corporations will be different, to some extent, to best practice at a neighbourhood Tesco Express. Bayern didn’t look at Kompany’s Premier League relegation as an abject failure, they’ll have seen it as a datapoint in his (sigh) management philosophy.

An advantage that football has over the rest of the business world is that a relatively wide range of manager metrics are easily obtainable. While they won’t be able to tell you for sure how good a manager is, they can at least narrow down the pool of talent to ones who ‘do Agile’, if that were important to you.

If you’re in a boardroom, that is worth paying attention to. For everyone outside, it’s the least interesting thing about football managers.

Here’s the actual interesting thing. Kompany joined Burnley and oversaw them be comprehensively promoted and comprehensively relegated. From Kompany’s perspective, you can see this as a chance to practice his Nasdaq C-suite methodology somewhere that the C-suite hirers were definitely watching, unlike Belgium. But from Burnley’s perspective, you do wonder a little more about the value in hiring a McKinsey grad for the mill town Sainsbury’s.

This is the interesting part. The amount of teams wanting to transition between playing styles is almost certainly quite large, almost certainly a difficult task, and almost certainly a very valuable one. Recent Championship-to-Premier League success stories tend to have something interesting tactically about them, either in their initial season or in their later evolution: Brighton, Leeds, Brentford, Bournemouth.

I wrote recently that I think ‘Possession Play’ football is more or less what the best teams inevitably play due to the nature of the sport. The theory is that football is about space and ball control in comparison to your opponents: teams with better players choose a style that maximises their controlling talents; teams with worse players choose a style that minimises the differences or disrupts opponent control. Better tactics or coaching can eke out edges in this battle, moving you up the table, and allowing the team to bring in better players who bring a more advantageous balance to the space/ball-control match-ups.

Who are the managers who are best at this, and does that show up in the data?

(There is a sort of ironic point to be made here too that, at a time when statistical analysis for head coach hires is taking off, there’s a trend for hiring managers in their 30s. The assumption is presumably that they’re both closer to tactical innovation and have room to grow, the latter being a similar logic to snapping up 21-year-old players.)

This brings us to a final point.

In business, it’s recognised that people who are good at scaling a company may not be the best at running it once it’s big. Not only that, but people who are good at scaling may not want to keep running things anyway. (admittedly ‘Neither Moving Fast Nor Breaking Things’ does sound like a bit of a buzzkill). Football doesn’t quite have this same type of appreciation, beyond occasional relegation-firefighter specialists. Managers who can switch up a team’s style but aren’t great at coaching ‘elite’ tactics are seen as limited, both in their aptitude and their career prospects.

If that doesn’t change, it feels like we’ll only get more teams turning their promotion-brand football straight into their relegation-brand football, because it’s the best way for the coach to get a decent job next time.

—

A note on the Positional Play-to-Agile metaphor

Given that Agile is aimed at being, y’know, agile, it strikes me that the more literal metaphor would be Agile and ‘Relationism’, where the heavily process-dependent ‘Positional Play’ would be the kind of documentation-driven software practices that Agile was aimed at replacing. The metaphor is less of a 1:1 comparison, but still works on a ‘widespread methodology’ level. And now, the frog is truly dead.

The Athletic’s Michael Cox prompted online posts this week with this article: ‘Has the impact of analytics on modern football been overstated?’. Writers don’t write headlines, of course, so as way of summary here’s the penultimate paragrah:

It’s a reasonable point. The impact of expected goals on shot distances is the most visible thing for analytics to hang its hat on, and even that is contested (numbers of long-rangers were already falling before xG became widespread, even in online circles).

Shouldn’t this be clearer by now?

Here are some possible reasons why it might not be. Feel free to skim.

Football is ruthlessly efficient

Baseball and basketball are held up as the examples of ‘analytics’ in modern sport. Really, though, we mean MLB and the NBA. The NFL too, if we’re talking fourth-down decisions. The major leagues are plural, but in their own sports they’re singular.

Maybe the closed shop (with very limited international pollination) lets stale ideas circulate, or allows a culture which suppresses new ones. Maybe the global nature of football lets experiments play out somewhere, with the successful ones later appearing on the global stage.

It may not be closed- or open-shop per se. Football is also played by far more people, a larger marketplace for ideas to appear in, and has a fairly unique interplay between club and international football. While national playing styles might once have been a barrier to new ideas, nowadays it seems national FAs will take any approach that’s working. That, in turn, filters down to their coach education programmes.

The old-school saying is that only the results matter. Maybe 21st century football has a large enough and diverse enough sample size for that to be correct.

Other sports are uniquely inefficient

The three-point line is not wildly difficult to understand. In cricket, the largest influence of data has (I think?) been in Twenty20 strategy - a new format of the sport in which there was a lower body of ‘establishment’ knowledge.

Maybe we shouldn’t be expecting analytics to make huge on-field impacts after all.

Football should actually be much different

These are theories, so they’re allowed to wildly contradict each other.

In basketball, the ‘analytics’ change was to be more brave in shooting threes instead of long two-pointers. In baseball, strategy has shifted to walks and home runs. In NFL, the push is for fewer punts on fourth downs. The insight in Twenty20 cricket was around the value of boundaries and the trade-off waste of wickets in hand at the end of an innings. There’s a theme here: aggressiveness.

This is actually mentioned in Ian Graham’s book How to Win the Premier League too, in an anecdote featuring Brentford owner and Smartodds founder Matthew Benham: “Benham told me that his instructions to the Brentford manager were to attack, regardless of the opponent and the situation in the game. In minute one, Brentford must attack. In minute 90, leading one goal to nil, and down to 10 men, Brentford must attack.”

It’s a refrain that analytics-watchers will be familiar with. That teams sit back on leads too much, that managers don’t make substitutes early enough. It’s worth remembering that the disapproval of long shots is not about a conservatism around shooting, but about the magnitude of difference between a thirty-yard and a fifteen-yard attempt.

So, maybe football should be much more aggressive - or, maybe more semantically correct, less conservative - in its outlook.

The data isn’t better than (most) coaches… yet

Apart from the (contested) xG influence on shooting behaviour, transfers are probably the biggest evidence of analytics’ influence in football. “Data/analytics have had a HUGE impact on recruitment. Literally everything about the job and process is different now,” Ted Knutson tweeted on the subject of Cox’s article.

Scouting is also the area of football with the worst ‘human-to-required observations’ ratio. A Dean Oliver line that’s quoted in How to Win the Premier League: “Your eyes see the game much better than the numbers. But the numbers see all the games. And that’s a big deal!”.

Maybe the available data had an edge on scouting operations* that it didn’t on first-team coaching. (*Or, possibly, an edge it had over managers who tried to have the last say on transfers).

‘The data’ is changing though. Maybe the increasing availability of tracking and biomechanic data (and the tools people have for them getting better) will lead to more of an on-pitch impact.

It’s also worth noting that ‘football people’ have accepted a lower level of control of the off-field stuff, retaining control of tactics and player-management (although some head coaches do gripe about the say-so of their medical departments). How might things be different if managers had demanded they keep control of transfers, and relinquished some on-field control to ‘directors of tactical methodology’ instead.

The smart people are quiet (or, vice versa)

To go back to Ted Knutson’s tweet: “Data has had a moderate impact on style of play and game models. I don't want to go into huge detail here because a lot is still amazing IP we and others have developed and I get grumpy DMs when I talk too much.”

This is something that Cox’s article touches on as well. The thing with this one is that its argument is visible to fans, but a case of ‘these changes are attributed to [some other factor], but actually analytics is responsible’.

Semi-related, a thought experiment: how different would perception of analytics’s impact at Liverpool Football Club (men’s team) be if the press coverage had been different? Their data-involved ‘transfer committee’ had a unique level of negative attention in the traditional English press, and they’re also the only club I can remember to have a New York Times magazine article written about them.

Mikel Arteta’s season-end finishes at Arsenal were 8th, 8th, and 5th before finishing as Premier League runners-up in 2022/23 and 2023/24: what would this conversation be like if a long-read on their analytics set-up had dropped in 2022?

We’re looking in the wrong place

This would fit alongside some of the other possible options. Football might be uniquely efficient at its top level, where many people focus their time. But there’s a world outside the richest leagues in western Europe, and maybe the impact is larger there.

Speaking of geography, an aside (is it even an aside?). In an open-air sport like football, ‘the right way’ to play will look different from place to place. Would the heavy-pressing football that is so tied to Germany and (through Red Bull) Austria ever have emerged in Spain, land of the siesta? Might the state of English pitches that we see in videos from the 1980s be a factor in the long-ball football that took root?

A different spin on ‘looking in the wrong place’ is, to use a singular example, the absence of new André Villas-Boas-es. The Portuguese manager, once hired by Chelsea as a sort of heir to José Mourinho, is the earliest ‘data story’ I can remember. Shortly after he was sacked by Tottenham Hotspur in December 2013, Michael Caley wrote, using a rudimentary expected goals model:

Maybe you can chalk up the (lack of) André Villas-Boas legacy, a coach who made his name with an unbeaten season as Porto manager, as an example of football’s efficiency. But the absence of AVB imitators, conscious or unconscious, is probably down to the acceptance of expected goals too.

Analytics types have long said that their biggest impact in transfers is just saying ‘no’ to bad ideas. Maybe that’s the case for on-field matters too.

Fashions come and go and, inexplicably, 90s aesthetics are back and running stats are a hot topic.

Well, maybe it is explicable.

Football has had regular running data since the 90s, but a few things might’ve held back its usage and usefulness. One of them is ball tracking - harder to do than player tracking but really helps to know what’s going on on the pitch. Another is arguably imagination (see ‘Are Current Physical Match Performance Metrics in Elite Soccer Fit for Purpose or Is the Adoption of an Integrated Approach Needed?’ (2018, Bradley and Ade))

Once you have that, it’s easier to chop the numbers up by phase, and when you can chop the numbers up, they get more interesting and useful. Split them by position and tactical phase, and you get a very quick, very nuanced picture of a player’s physical demands on matchday. As I often do, I’ll point to the FIFA Training Centre as an example of this kind of work.

That detailed matchday information also lets you work backwards. Backwards from matchday: to training, to recovery from injury. MLS team Chicago Fire are using French company Footovision to help with this. Other systems will, one presumes, be available. (Perhaps like fellow French-born company, Skillcorner).

Platforms have power. It’s not just Parisian-based companies springing up over the last few years: over the last couple I’ve noticed a few research papers using data from the Kitman Labs platform (like this one on hamstring injuries). My venturing into the sports science realm doesn’t even reach ‘amateur’ status, but I’m familiar with the ‘n=8’ sample size problem. This paper, though, is able to call on data from 36 team-seasons in elite men’s football.

Software (usually) means scale. Loop back to the FIFA article I shared earlier: that was about women’s football, data from the 2023 World Cup. As the sport tries to catch-up on its historic lack of support and active hindering of women’s football, some things scale better than others. Training infrastructure and coaches can only be shared so much. Computer vision data collection scales better than event data collection. The insight into how to split tracking data into tactical phases, to benchmark players against their peers, barely needs to scale at all. Once it’s there, it’s there.

And the global game being the global game, there are geographic implications. Check out the countries that viewers were tuning into this computer vision for tracking data Q&A stream from - there’s Turkey, Uzbekistan, India, Congo, Uruguay. I won’t repeat my previous line about ‘why not make your own tracking data instead of buying it’ (also from that September ‘22 post above), but a rudimentary system may still have value. Even if you work with large error bars, maybe you could take inspiration from the literature, gather non-quality assured data for the Uzbek second tier, and be able to benchmark attacking full-back requirements against their peers.

So, yeah, running data is cool again. Ditch the 2010s slimfit jeans, dig out the colourful maximalism.

There’s another advantage that running data has. No-one has to argue about what a ‘duel’ is.

This blog won't be about Ian Graham's book, How to Win the Premier League. I might write about it at some point when I've finished reading it, but really y'all should just read it yourselves, it’s good.

However, reading it reminded me I haven't written anything fun in a while. Let's rectify that.

Going back through the five years of Get Goalside, there was something I forgot I'd thought about. It was a piece theorising why the best teams throughout history tend to play a possession-focused game and, equally, why weaker teams tend towards deep blocks and long balls.

Here's the core idea:

"'what makes a football team 'good'?' is easily answered (though much harder implemented): being better at controlling the ball and better at controlling the space. [...] From here, from this basic framework, you can break down football archetypes into the 2x2 quadrants that high-or-low space and ball control form."

I want to poke at this. I suspect it'll seem like kindergarten tactics to some readers, but we've all got to go through kindergarten once.

Let's try filling in this 2x2(x2) matrix (each segment needs to cover in- and out-of-possession sides of things).

High space control, high ball control
In possession: Guardiola-ball (for want of a better term)
Out of possession: aggressive, zone/option-oriented press

High space control, low ball control
In possession: [unsure - defensive possession with low-technical players?]
Out of possession: effective block, but little direct pressure on the ball carrier

Low space control, high ball control
In possession: Vertical ground passes or ball-carrying (Red Bull-ball? for want of a better term)
Out of possession: aggressive ball- and/or player-oriented press (Bielsa’s Leeds)

Low space control, low ball control
In possession: hoof ball
Out of possession: low pressure on ball-carrier and porous in the middle

Trying to fill in these archetypes, it becomes more clear what each space and ball control is being distiled into. In possession, ball control is obvious, while space control feels more about the amount of controlled passing options available. If you think about times when a team feels like it's losing control despite retaining possession, it's when they're passing around a press/block, where each pass is the only one available, often going backwards.

Out of possession, space control (intuitively) would be about denying these options. People talk about compactness, but the compactness isn't exactly an aim in itself, it's a method of achieving an aim. Compactness for compactness’ sake is silly. The ultimate aim is to prevent progression through the most dangerous area of the pitch. Blocking passing lanes and being close to potential receivers, that's out of possession space control.

'Ball control' when out of possession, meanwhile, might sound weird, but strong duel ability would count, as would applying pressure to ball carriers. Maybe tenuously, you could count the pressure on a ball-carrier as an extension of the duel, or a preamble, but it's not exactly controversial to say that pressure on the ball-carrier affects control of the ball.

Simple stuff so far. But you can start to use this as building blocks for more complex things.

You should always try and maximise your advantages relative to your opponent. If your team has elite ball control in possession, that makes your in-possession plan obvious. But if your ball control in possession is only moderate and your opponent's out of possession ball control is great, maybe you change tack. You can also play in a certain style to affect how the opponent plays in response: playing a low ball-control style in possession often lowers the ball control the opponent is able to have while in possession too.

So far, sounds like basic match planning, but these balances will flow and flux moment to moment in-game. None more so than in the ‘transition’ phases. Ultimately, these moments are not so separate to settled possesson as the popular distinction might make them seem; you could just phrase them as moments where space and/or ball control is up for grabs. (This also makes terminology around ‘manufactured’ transitions easier, when teams try and coax a press to then rush through - they’re just trying to reduce the space control of the out of possesion team).

Frameworks aren’t inherently good, but they can be a useful bedrock or scaffolding or starting point for further ideas. For example, you can start to imagine how you’d use this space-ball control matrix to build out a suite of KPIs, geared specifically towards identifying space control or ball control.

Unfortunately, there are some potential flaws in this particular framework.

On a team-wide level, 'ball control' should probably include the ability to offload the ball if under intense pressure, which requires teammate options, which is also more or less how I've defined 'space control' in possession. Diniz-ball seems like a useful test case here, if positionism discourse is willing to stoop to the level of kindergarteners. Are the interchanges between players in close spaces about achieving high space control, high ball control, or both?

Secondly, is there an inherent advantage that space control has over ball control, or vice-versa? If you have to choose between optimising one or the other, is there a clear answer which it should be? If there is, does that affect the usefulness of the matrix?

I suspect that there are more. Are there holes in this you can find? Or, in the other direction, are there advantages that I've downplayed too much or have missed completely?

Assuming that this theory tracks, that means that football is:

- a game of space control and ball control, in and out of possession…

- …where the aim is to score more and concede fewer than the opponent (read the recent 'understanding AI' piece for that differentiation)

And everything goes from there.

Unclear at this stage how it leads to Premier League titles. Maybe that’ll be in the second half of the book.