|
Coffee 10:50 Wed Sep 14
Football and Big Data
|
|
Warning - long article.
"Big data" - the world of analytics, algorithms and statistical models - are increasingly part of our lives, and professional sports such as football are no different.
"Why did you pick him?" "Don't take it short, get it in the box!" "Put a striker on!"
Complaining about the manager's selections and questioning players' decisions on the pitch are time-honoured traditions of being a football fan.
Whether watching from the stands, on TV or listening to the match on the radio, offering full-throated advice to the coach and exhorting the players to try harder or do something different is one of the "joys" of supporting a team.
We football fans flatter ourselves that our alternative ideas would immediately improve the team's performance. Mostly they're based on intuition and a "feel for game", often nurtured over years of watching nil-nil draws in the freezing rain at uncovered away ends.
Nowadays when player acquisitions or formations strike us as baffling or obtuse, there is likely to be method in the madness. As data on player attributes, movements and positioning become more comprehensive and analytical models more sophisticated, football is relying much less on gut instincts.
It is still "the beautiful game", but it is one that increasingly resembles a game of chess.
Data is being created at historically incomparable rates, in all conceivable areas of life. We are living at the start of the age of so-called "big data". Analytics, algorithms and statistical models are increasingly part of our lives, whether we like it or not.
Professional sports are no different. This is an extraordinarily lucrative sector, where data has been identified as potentially giving athletes and teams a competitive edge.
The data revolution in sports is often traced to Billy Beane, general manager of the Oakland Athletics or A's, an unheralded team in Major League Baseball in the US. Beane employed a method that came to be known as Moneyball after Michael Lewis published a book about the A's in 2003.
During his time with the Oakland Athletics baseball team, Billy Beane used an analytical, evidence-based approach which came to be known as "Moneyball"
Beane used an analytical, evidence-based approach to identifying players who could meaningfully contribute to the team and offer good value for money. It drew on sabermetrics, a scientific method for analysing baseball performance pioneered by Bill James. The A's sustained success on a limited budget, later chronicled in a movie based on Moneyball starring Brad Pitt, turned the spotlight on data analysis in sports.
From baseball, these analytical methods for appraising players quickly spread to the NFL and NBA, and a number of sports in the UK. In cricket, former England coach Duncan Fletcher favoured statistical analysis of batting and bowling to identify the best way for players to score runs and to get batsmen out.
Clive Woodward's innovations in using player data helped the England rugby team to win the World Cup. Dave Brailsford's innovations in performance training data helped make Team Sky multiple Tour de France winners.
In football there were pioneers too. In fact, recording granular data on players and match events goes back further than you might think.
Charles Reep coded his first football match, counting passes and noting positions, in 1950. Valeriy Lobanovsky was doing the same in the Ukraine in the 1970s. Former England manager Graham Taylor also used a crude form of analytics to inform his long ball tactics with Watford in the 1980s.
With the launch of the Premier League in 1992, and the money and exposure brought by the Sky TV deal, a number of football data companies were launched, including Prozone in 1995 and Opta in 1996.
Recording the direction of a player's penalty shots might just help avoid this.
These early efforts were impressive for the time. For instance, the computer game Championship Manager (later renamed Football Manager), launched in 1992 with a database of 4,000 players and statistics on 30 attributes per player.
Speaking to the British Science Festival in Swansea last week, Dr Tom Markham, head of strategic business development at Sports Interactive - makers of Football Manager - said those numbers have exploded in the subsequent decades.
"The game now has a database with 319,726 current players. With former players, who may take other roles in football, it comes in at over 600,000."
Compiling that database, Dr Markham said, is a big job.
"We have people on the ground in 51 different countries covering 140 leagues. There are 2,250 fully researched clubs, with 250 statistics on each player - aggregated to 47 in the user interface.
"With 1,300 scouts, all the main clubs have one researcher, and top clubs like Chelsea have multiple experts."
Some Football Manager alumni have gone on to work as scouts with professional teams, he added.
As professional football revenues continue to grow, and leagues become increasingly competitive, the data industry has also expanded. Huge amounts of data from companies like Opta and Prozone underpin not only team tactics but also sophisticated media coverage.
Coaches employ wearable tech to monitor player fatigue on the pitch and in training, to prevent injuries resulting from physically overloading players. Recorded movements on the pitch inform models of formations and playing style, with simulations and in-game stats for coaches to make halftime adjustments.
Data analysis is about spotting patterns and making predictions. Recording the direction of a players' penalty shots can show which area he favours. Knowing this a goalkeeper can increase the probability of "guessing" right.
Last season's Premier League champions Leicester are comprehensive users of analytics.
One important metric is "expected goals", a key input in betting and analytical models. It is a predicted probability of a goal coming from a shot in a particular area of the pitch. How many shots a team has from those areas can be used to predict the likelihood of scoring.
When Leicester became Premier League champions, it was a huge shock. But it is no coincidence that their use of analytics was among the most comprehensive and forward-looking in the league.
Leicester's unusual style of play, with little possession and relying on fast attacks, took many opponents by surprise. The team suffered virtually no injuries, and relied on the emergence of unheralded players like N' Golo Kante and Jamie Vardy.
Those who believe in the data-driven approach would say this is exactly the kind of comparative advantage statistics can bring.
Another of the great rituals for football fans is speculating about transfers. Who are we going to buy? Who should we buy?
Buying and selling players is a huge business. In the recently concluded summer transfer window, Premier League teams combined to spend over £1bn, with Manchester United spending in excess of £80m on a single player.
Datasets like those compiled by Football Manager have become a resource for the scouting and recruitment operations of many teams.
Finding a low-cost, high impact player like Riyahd Mahrez or Dmitri Payet can have remarkable results on the pitch. For clubs with smaller budgets, finding a rough gem or talented youngster that they can later sell for a profit is a crucial form of revenue.
Rating Norwegian prodigy Martin Odegaard, who made his international debut aged 15, proved difficult
"£8 million a year is the average running cost for a tier 1 academy and teams have to find talented youngsters who they can nurture and sell on," Dr Markham said.
But assessing young talent is difficult, and not every talented youngster will become a Gareth Bale, who was discovered as a boy in Wales, nurtured by Southampton's youth academy and later signed for Real Madrid for a world record fee.
Markham told another story of young talent, Martin Odegaard, the Norwegian prodigy who signed for Real Madrid at the age of 16 after making his debut for the national team at just 15.
When Football Manager came out in Norway Odegaard wasn't in the game because he was a minor, causing a metaphorical riot among Norwegian fans. He was added to the game's database when his dad tweeted a picture giving parental consent.
But how to rate the prodigy?
When the Football Manager club scout sent his rankings to the head of Norway operations, it raised a red flag. How could a 15-year old score so highly? The Norwegian chief went to see Odegaard play a dozen times before corroborating the data sent through to the London HQ, where it was again rejected as improbable.
Dr Markham says Odegaard's stats went through a dozen different checks before his astonishing grades were accepted.
Recruitment is so important to professional clubs that the average Premier League team has 7 international scouts. But clubs don't have the resources to cover players in every country - and many teams use Football Manager to inform their own scouting strategies, Dr Markham said.
Other teams are creating their own datasets, and working with other companies to come up with bespoke solutions. Teams using analytics to thrive include Brentford and the Danish club Midtjylland, both with connections to Matthew Benham, a noted convert to the data-driven analytical approach.
Aside from clubs and gamers using simulations, data underpins many other aspects of the football industry, from TV coverage to betting models and fantasy football. Using analytics to spot patterns in match results is used to monitor match fixing. Books with titles like Soccernomics and Soccermatics allow fans to get close to the "action" of data analytics.
The relationship of gaming with professional football goes both ways. Players enjoy simulations like FIFA in their frequent downtime and many players are used to receiving data on their own performance.
A picture of Paul Pogba playing Football Manager and signing himself for Chelsea set off speculation that he might move from Juventus to Chelsea.
And according to Dr Markham, those involved in the beautiful game itself can be - perhaps unsurprisingly - fixated on their representation in virtual versions like Football Manager.
He often receives messages from players and agents, he said. "Sometimes they complain about their ratings in the game, or their agents try to get them put up."
- BBC (www.bbc.com/news/science-environment-37327939)
|
|
|
Replies - Newest Posts First ( Show In Chronological Order)
Alex V
1:17 Fri Sep 16
Re: Football and Big Data
|
Ah I thought you were talking about the stuff that tracks the position of every player at all times (which I think prozone and maybe others do). I guess you could say that any logged action from the field is positional in some sense.
|
gregan
11:15 Thu Sep 15
Re: Football and Big Data
|
It's just logging Alex. What players have done and where on the pitch they've done it. It's what you are logging that potentially have any real meaning. But so many things you can argue to devalue it. Sports Science is guesswork to a degree otherwise everyone would have the same approach.
I think it's good for scouting. E.g. I'm a manager. I want my full backs overlapping all the time. Stats. I want CB's that can play out and pass into midfield beyond a line. Packing stats. I want players who get into receiving positions in between the line. Packing stats. I want players that make continuous goalscoring runs beyond the back 4. Stats. I want players who can receive in between the lines and play a killer ball beyond the back 4. Stats/Packing. There will be stats on all these areas. Not conclusive but definitely a tool that could help shape your team the way you want it to play. No one says stats are gonna solve world peace. Love breaking down the game though.
|
Alex V
5:25 Thu Sep 15
Re: Football and Big Data
|
What I don't see with positional data is how it correlates with the abilities of a player. If you're looking to sign a player I think you'd be less interested in where they receive the ball on the pitch, as much as their technical ability, attitude, effectiveness. For example when Kouyate plays CB, DM or AM for us he is operating in different parts of the pitch looking to do entirely different things - it may be useful to evaluate them with positional data, but it doesn't really tell us much about him as a player and his ability. After all, positional play is much more likely to be the result of a coaches approach rather than a player's. Lampard Jnr was great at arriving late in the box, but if he hadn't been encouraged in that role we might never have known that potential (as potential buyers).
|
gregan
1:44 Thu Sep 15
Re: Football and Big Data
|
Ditcher, similar. I was actually commissioned to design the opta system for rugby league, but said no as it was pay on completion!
Stats will never cover everything, but you can focus on specific areas. Packing stats clearly show players who pass with purpose and receivers who pick up passing positions in pockets beyond defensive lines. Still variables but not as many as some general stats.
Heat maps. Who gives a flying flip. It tells me nothing without seeing where the ball is and where every other player is on the pitch at that moment.
Positional and tracking data is the new thing and has potential. As someone said stats largely ignore positional data as well as recieving data
|
Alex V
6:15 Wed Sep 14
Re: Football and Big Data
|
You're answering your own criticism though - clean sheet data clearly needs to be accompanied by info on the nature of the defensive record etc. There is literally no single stat that offers a failsafe measure of any player's quality - the whole business of dealing with stats is working with other bits of data (and scouting) to qualify and interpret those numbers.
|
El Scorchio
5:34 Wed Sep 14
Re: Football and Big Data
|
Clean sheets are tricky as well.
You could have a bang average keeper with a superb back four in front of him which doesn't give the oppo a sniff of goal, or an abortion of a defence with a superb goalkeeper, but he gets 6 put past him because the oppo have 6 point blank shots.
It's no reflection on the abilities of the goalkeeper either way and a great example of the flaws in recording stats for individuals in football.
In both of those instances, watching the players playing would be a far more accurate way of assessing the players rather than looking at a load of potentially misleading numbers on a sheet
|
Alex V
4:33 Wed Sep 14
Re: Football and Big Data
|
Goalkeepers can be credited with assists like any other player. Also their passes can be analysed like any other player's. I think a clean sheet for a keeper is generally a useful stat - it doesn't say everything, but it doesn't claim to. You could look at clean sheets for a keeper and combine it with the amount of activity they undertake in matches, then it might be more useful - that's not a weakness of analytics, it's a strength.
This thing about dummy runs or actual dummies by players is quite true - those are hard to measure in statistical terms. But nobody claims analytics solves everything in football. I think it's a bit unfair to criticise data for what it doesn't do - let's look at what it can measure and how useful that is.
|
geoffpikey
4:16 Wed Sep 14
Re: Football and Big Data
|
El Scorchio
"If one player for example makes a dummy run and pulls a defender out of position which leads to a goal for someone, that's never noted or analysed."
Exactly. That's why I HATE statistics for "Assists". Never seen a keeper credited with an "assist" when it happens often, even under FIFA guidelines, for a long kick. Even if it has been credited, "Assists" is Bollocks data. So, a lot of the time, is "clean sheet." Some keepers never even have to make a testing save in a game.
|
El Scorchio
4:09 Wed Sep 14
Re: Football and Big Data
|
Pulis had Palace playing exactly like Leicester when they went on that unbelieveable run a few seasons ago and he got manager of the season at the end of it.
Sit back and then break really quickly with long balls over the top or in the channels for quick players to run on to.
Clearly you need the right combination of players in your team to make it so ruthlessly effective.
|
Ditcher
4:03 Wed Sep 14
Re: Football and Big Data
|
gregan 2:29 Wed Sep 14
I worked at Opta 2000-2001 at the Threadneedle St office until Sky bought it.
|
Mart O
3:17 Wed Sep 14
Re: Football and Big Data
|
The interesting thing about Leicester is that their style of play went against what everybody was preaching at the time, other than Diego Simeone and to some extent Mourinho.
It's worth bearing in mind that the alternatives for lower level teams to be competitive seemed to be the possession based game Southampton, Spurs (and even Swansea) had all been playing and the Fat Sam, Tony Pulis style, respect the point mongfest.
I reckon that data analysis had more to with the development of the latter style than the former. This is not a good thing.
There's surely some place for data and analysis in the game but as Gregan says, it's difficult to define what is meaningful data in football. Look at the clusterfuck over at Fulham where that Craig Cline analyst chap seems to have the last word on transfers. Madness.
|
El Scorchio
2:29 Wed Sep 14
Re: Football and Big Data
|
Everyone knew what Leicester were going to do last season.
Not many people were able to stop them.
One thing which is really important that i think stats anaylsis simply doesn't take into account is chemistry between players and how playing for a certain team in a certain way, under a certain manager affects how they play.
Baseball is much easier to analyse a player's exact worth and what they bring as all the things a players does are in isolation and you can easily see exactly what each player has contributed obviously- pitching, batting and fielding.
With football, it's far more about how individuals fit together into a dynamic situation where one player's actions dramatically affect what the other players can and will do. If one player for example makes a dummy run and pulls a defender out of position which leads to a goal for someone, that's never noted or analysed.
|
gregan
2:29 Wed Sep 14
Re: Football and Big Data
|
I worked at opta when they first formed so practically invented football stats. So many variables in football make it harder to find that edge. Often lots of data, but people not knowing what is meaningful and relative.
Sometimes you will see a generalisation like crossing the ball from position x is pointless so don't attempt it. Now lets look at us, if Payet is crossing an inswinger from the left with Carroll attacking it, and Antonio behind then it's clearly worth doing at times. So, the generalisations I hate, I'd like to see more team/player specific data based on their strengths.
Most public stats you see on tv are pointless. Distance covered (headless chicken wins potentially), passes made (so what), possession % (Zzzzzz). PACKING is a meaningful stat that 2 German players came up with and is used on German TV sometimes. What is packing? It's the number of players you take out/bypass with a pass (line breaking). Few rules to it, but it produces meaningful numbers to passing and recieving. England's packing stats at the Euros were awful. We simply didn't bypass players with out passes or got players in between lines to receive. Packing stats generally reflect who has won the game, much more than any other tv stats do.
I'll shut up now. Goal!
|
Alex V
2:28 Wed Sep 14
Re: Football and Big Data
|
Hmm makes it seem like analytics is something to do with Football Manager. Not sure that's very helpful at all.
|
Marston Hammer
2:08 Wed Sep 14
Re: Football and Big Data
|
Baggins 2:01 Wed Sep 14
Yeah stuff like that annoys me. Lazy pundits come out with stuff like 'oh well, it will be tougher for them this year as teams know how they play now' as if it took everyone a whole year to figure out their style of play.
|
Baggins
2:01 Wed Sep 14
Re: Football and Big Data
|
Interesting article that Coffee, son. No doubt data analysis can be useful, especially in terms of sports science and fitness. I'm not entirely sure things like "areas of goals scored" are very useful though and I don't think football lends itself to "Big Data" as much as things like NFL, cricket and rugby.
People will keep trying to use such stats though, espcially those who struggle to understand the game when just watching it.
Oh and this part:
"Leicester's unusual style of play, with little possession and relying on fast attacks, took many opponents by surprise. "
...is a massive reach. Counter attacking is hardly a new tactic brought in by using data analysis.
|
|
|
|
