No one did, I was just arguing on the other side of the spectrum. Tackling = bad positioning is not true but if someone would say that tackling = good defending isn't also necessarily true (but it's one of the arguments which is used when saying that AWB is great defensively).
Brentford, Liverpool and the coming of age of football analytics?
- Thread starter Teja
- Start date
-
In the latest episode of the United Hour Podcast, Nik @Rood and Imran @Annihilate Now! start with the good news by celebrating Manchester United Women's emphatic FA Cup win over Spurs at Wembley and the future of manager Marc Skinner.
Then it's the bad news and the ever worsening form of our Men with losses to Arsenal and Palace.
Finally the main debate - ETH Out now or keep him until the end of the season - if not longer. Are you Team Nik or Team Imran?
At the end of day, when all is said and done and taking all things into consideration. I don't believe I used a cliche in my original comment.
I don't know who uses that, I remember people saying AWB is arguably the best 1v1 defender, not overall best defender as he makes mistakes not defending the far post and makes it way too many times.
AWB wins many tackles as many wingers tries to take him on and around 90-95% of the time they will fail and we have seen that for 2 years.
Sure.
Exactly my point. People love to draw a conclusion based on one stat, when in fact if combined with another stat it'd give a completely different reading.
One of the best examples is Fred and passing under pressure. Looking at stats, you'd see that he's pretty good there. Most passes under pressure, but people forget that the reason he's being pressed that much is because opposition believes that they can draw an error from him which can give them an easy shot.
Just on the Ole point - The issue of how useful real time data is again relates to how the manager in question is able to contextualise this data and come up with the right response for it under the pressure of the game.
Glad this thread was made - just been getting into Python and Tableau - making basic data visualisations, so getting some info off guys more experienced would be great. Anyone know how much access to real time data costs? Or event data for current season?
Glad this thread was made - just been getting into Python and Tableau - making basic data visualisations, so getting some info off guys more experienced would be great. Anyone know how much access to real time data costs? Or event data for current season?
Yeah agree with that. Many stats in isolation doesn't give full picture.
My favourite was people abusing "never dribbled past" "lost possession" stats.
The Brentford owner is an Oxford Physics graduate who made his millions with a company called Smartodds which uses statisitics to calculate the outcomes of games to beat the bookies. Whilst most clubs use increasinging amounts of data to make decisions few have been as dedicated or successful as Brentford with the approach. They managed to get promoted to the Premier League for the first time since the 40s whilst making a profit from transfer market dealings, they're one of the poorest clubs in the Championship in terms of other revenue sources.
Good thread. When it comes to advanced analytics in football, yeah, lots of clubs have good internal models nowadays and use it extensively. It obviously gives you an edge. Real time stats can confirm or disprove eye-test impressions of what is happening on the pitch, give information that was missed, etc
In scouting and player development is useful to identify a player's strenghts and weaknesses, etc
I also want to point out that at barcelona bartomeu spent a lot of money to build one of the best departments in the sport, so good other clubs went to them to learn new stuff, entirely as a PR move, while the club did not actually use them at all
In scouting and player development is useful to identify a player's strenghts and weaknesses, etc
I also want to point out that at barcelona bartomeu spent a lot of money to build one of the best departments in the sport, so good other clubs went to them to learn new stuff, entirely as a PR move, while the club did not actually use them at all
For sure.
Really? If so, that's pretty smart of them and probably helped him get more bonuses, etc?
I'm defintely interested in how analystics have come into football so I'll be paying a lot of attention to this thread for more articles and information.
Sweet - does it have to be done through R or can it also be done via Python?
It's a python code.
In the 1st code block, you can see the google drive link where the data is saved. So, you can maybe download the data from there for the games
Hey all! It's been a bit of a struggle getting access to good data. As Anant says, scraping seems to be the way to go.
If you're interesting in mucking around with scraping tools, Understat seems to be a popular target. This is a quick tutorial. He also has a bunch of other useful tutorials on visualizing football data.
There's also some really nice free data from statsbomb but this is historical. You can still build and test algorithms on this data though, it's just not quite instant gratification because 2017/18 feels like a long time ago. I think you can use their python api client (R available as well) to fetch more recent data. I have to play around with it to see what's here.
This playlist has a bunch of talks on how to model higher level things like pressing, threat etc. The whole thing is several hours long so be prepared for that if you start watching
And folks are probably familiar already but just scraping from fbref.com can help make interesting visualizations as well. (They use statsbomb data under the hood).
For me the most frustrating thing about being a Man United fan is that we seem to disregard set pieces.
Most other top clubs have started paying attention analytically, some for a few years.
And in my opinion, even the best are many years away from optimal strategy, I genuinely think it could break the game at some point.
Most other top clubs have started paying attention analytically, some for a few years.
And in my opinion, even the best are many years away from optimal strategy, I genuinely think it could break the game at some point.
I agree fully, stats only become interesting when you start combining lots of different inputs. And my understanding is that, even after all that, clubs will still get scouts to look at actual footage of the player.
Kinda as has been implied here, I think the one thing to keep in mind when it comes to well-run analytics department, is that these people won't simplify anything beyond what's absolutely necessary to keep things manageable. We at home don't have the time to consider everything there is and develop analytical models and watch the player - but those people are professionals who do this full-time, with big money and indeed the future of their clubs at stake. They're not going to take shortcuts.
I think both R and Python have functions that allow you to insert code from the other and let it be used in the other environment. (I.e., insert Python code in R, within a function that does the conversion for you.) I do know R if ever that might be useful in this thread. I can't do much with its statistical modelling features, but I can handle data manipulation really well. (I love creating long tidyverse sequences.
)If I remember correctly, btw, a lot of the freely accessible data behind more complex indicators (like xG) is of relatively low-quality. The paid variant supposedly is better.
I agree. Analytics should be combined with scout preparing their report watching the player. It should be both, not one or other.
Love what Brentford is doing, just brilliant to read their story and what they have achieved.
Yeah, xG models of Statbomb uses a lot more variable including GK position, height at which contact was made, pressing at that time, etc. However, if someone wants, they can either model the xG weight themselves by running regressions on a million records or use Understat numbers. Shot over shot- numbers might vary by a significant margin (20% odd), but over a bigger sample, the difference wont be a lot, I believeI think both R and Python have functions that allow you to insert code from the other and let it be used in the other environment. (I.e., insert Python code in R, within a function that does the conversion for you.) I do know R if ever that might be useful in this thread. I can't do much with its statistical modelling features, but I can handle data manipulation really well. (I love creating long tidyverse sequences.
If I remember correctly, btw, a lot of the freely accessible data behind more complex indicators (like xG) is of relatively low-quality. The paid variant supposedly is better.
We are in the age of Big Data.
The acquisition of data.
Turning the data into information.
The analysis of that information.
The learning of what the information is telling you.
The information led improvements.
And those who are best at that will be the most successful.
The acquisition of data.
Turning the data into information.
The analysis of that information.
The learning of what the information is telling you.
The information led improvements.
And those who are best at that will be the most successful.
Regarding Midgetland, did they outspend other clubs in Denmark?
I'm sceptical. Didn't Liverpool buy Naby Keita because of his stats? He's been bang average. Didn't Arsenal buy Xhaka because of stats?
I'm sceptical. Didn't Liverpool buy Naby Keita because of his stats? He's been bang average. Didn't Arsenal buy Xhaka because of stats?
It is measurable. Probably not in a direct metric the layman would understand.
Comment from Reddit
Yeah I agree. Look how many times a full back is left one-on-one compared to a centre back. And look at the relative risk of a full back making a tackle and risking a foul, almost always on the side of a pitch, as compared to a centre back, much more likely in or just outside their own penalty area. Stats are great if, and only if, they are presented in the context of all the intricacies of a football match.
I’m sure there’s a way it could be measured, but it’s very difficult. There are a lot of factors at play
True.
Wouldn't put it beyond an advanced degree holder in the computational sciences to put something together. They obviously wouldn't be doing it for free, so the odds of me and you seeing it are nil for a while...
Definitely, in basic terms you’d be measuring interceptions and also looking at passes to that area of the pitch across a large sample size but have to consider other factors.
E.g. Percentage of balls played by Team A from Area C1 to area B3 across 10 games was 35% average but in game against Opposition X it was 15%, attributed to the positioning of Player Y making it an unusable option etc.
In layman’s terms that’s a basic way of creating a measurable metric but obviously there’s far more in-depth ways to conclude things like a full backs positioning.
100%.
Fbref is going for showing per 90 minute stats and comparing players as a percentile across players in the big leagues for that rough position/role, my favourite one to look at though of course you always need to know how to use it.
For sure. There's loads of variables at play in football but it's still useful to look at them and compare them with watching football and see how it stacks up. The problem is when people take it as gospel and especially when small differences are taken as meaningful, when it can easily be explained away by surrounding players/competition/team style. And in terms of scouting, it's not as simple as just even picking the perfect player. Mentality and adaptability of players is something that nobody can measure and realistically project how they'll go from one team to another with different pressures, different group of people and a different location.
But it for sure has its uses and i have no problem using data to compare players, especially as it's an easy way to gauge how a random players generally does over a season and their impact compared to watching them once in a big game a year ago.
I got round to reading this article now. It's very interesting - and covers a subjects that's probably as of yet underdeveloped in football.
The way I see it (and I don't think it's happening much yet), analysts should be working with match data in real-time to look at things like available areas of play and other things that help figure out what spaces are available or need to be plugged on the field. They can then communicate their observations to an assistant coach who can make recommendations to the main coach. Are there any rules against players have some kind of earpiece? (I suppose most are annoying physically, but probably something could be developed that players could wear that's not dangerous and does not interfere with their play.) Cause once you have that in place as well, then the tactical battle can really go up a notch, with coaches directing team shape and actions in real-time to react to what the opposition is and isn't doing - kinda like what's happening in American Football already, except without all the breaks in play.
Even with all the stats in the world in play and used correctly, not every transfer won't be a success. Football is a complex game, and stats anyway can't take into account how tactics and personnel might change at the player's new club and how that might affect him, or what else might happent to a player's form. (And in Keita's case, it doesn't help either that he is injured all the time.) Also, if Keita is the only high-profile failure among Liverpool's transfers since they went all-in on analytics, then their transfer business has become a lot more effective than it used to be. (Low-key transfer fit a different conversation, since more risk can be taken with those.)
I'll probably date myself a bitThe way I see it (and I don't think it's happening much yet), analysts should be working with match data in real-time to look at things like available areas of play and other things that help figure out what spaces are available or need to be plugged on the field. They can then communicate their observations to an assistant coach who can make recommendations to the main coach. Are there any rules against players have some kind of earpiece? (I suppose most are annoying physically, but probably something could be developed that players could wear that's not dangerous and does not interfere with their play.) Cause once you have that in place as well, then the tactical battle can really go up a notch, with coaches directing team shape and actions in real-time to react to what the opposition is and isn't doing - kinda like what's happening in American Football already, except without all the breaks in play.
but I remember Hansie Cronje (South African Cricket Captain) wearing bluetooth earphones against India back in '99 and getting instructions from the coaching staff. That blew up into scandal territory back then.If you put on your black mirror hat a bit, you can even imagine a world where you have coaches yelling instructions realtime into players' ears and almost controlling them like it's a FIFA game. Or why stop at a coach, you can have a game AI updating its model and yelling out these instructions. I imagine it's football nirvana for control freaks like Bielsa.
Too many nerds on this thread for me........
Don't worry, it will barely make it past three pages. See previous, 6 seems to be the record.
https://www.redcafe.net/threads/data-why-scoring-goals-is-more-important-than-conceding-less.447862/
https://www.redcafe.net/threads/premier-league-table-2019-20-as-predicted-using-big-data.449323/
https://www.redcafe.net/threads/dat...eed-in-the-league-now-vs-a-decade-ago.452406/
https://www.redcafe.net/threads/match-stats-data-perspective.459041/page-6#post-27046587
https://www.redcafe.net/threads/big-data-in-football.461275/
Thanks for your contribution.
The OP was a really good read, especially the Brentford twitter thread
Well, you don't buy a player because of 'stats', hopefully. You try to use data to create a model, that you think will make for good player assessment. It is guaranteed to give you some failures, since there's inputs you probably can't quantify that well and some you won't even try adding. Then ideally you'd analyze the failures to try and find what the original model didn't pick up.
We also bought Alisson, van dijk, Robertson, Fabinho, Salah and Mane because of stats.
If you're doing it for kicks and to learn, many sites have embedded tables which you can pull directly to whatever BI tools you're using. I did it from fbref to Power BI. The dataset is modeled once in Power Query and then you just click refresh after each game. But I've never tried it on live data, what would be very interesting to see especially if you can put some AI on top of the stream of data against historical data.
Is that true?I also want to point out that at barcelona bartomeu spent a lot of money to build one of the best departments in the sport, so good other clubs went to them to learn new stuff, entirely as a PR move, while the club did not actually use them at all
Do you work in Automotive? Because that sounds like 6 stage Industry 4.0 description we use. But it probably applies to all areas.
I'll probably date myself a bit
If you put on your black mirror hat a bit, you can even imagine a world where you have coaches yelling instructions realtime into players' ears and almost controlling them like it's a FIFA game. Or why stop at a coach, you can have a game AI updating its model and yelling out these instructions. I imagine it's football nirvana for control freaks like Bielsa.
With the development of bone conduction headphone technology, it doesn't even need to be in the ear. I'd expect that to happen in next 3 years.I got round to reading this article now. It's very interesting - and covers a subjects that's probably as of yet underdeveloped in football.
The way I see it (and I don't think it's happening much yet), analysts should be working with match data in real-time to look at things like available areas of play and other things that help figure out what spaces are available or need to be plugged on the field. They can then communicate their observations to an assistant coach who can make recommendations to the main coach. Are there any rules against players have some kind of earpiece? (I suppose most are annoying physically, but probably something could be developed that players could wear that's not dangerous and does not interfere with their play.) Cause once you have that in place as well, then the tactical battle can really go up a notch, with coaches directing team shape and actions in real-time to react to what the opposition is and isn't doing - kinda like what's happening in American Football already, except without all the breaks in play.
Even with all the stats in the world in play and used correctly, not every transfer won't be a success. Football is a complex game, and stats anyway can't take into account how tactics and personnel might change at the player's new club and how that might affect him, or what else might happent to a player's form. (And in Keita's case, it doesn't help either that he is injured all the time.) Also, if Keita is the only high-profile failure among Liverpool's transfers since they went all-in on analytics, then their transfer business has become a lot more effective than it used to be. (Low-key transfer fit a different conversation, since more risk can be taken with those.)
I'd say most people consider stats as a tool to compare player cards like in fifa. This isn't really what clubs are doing with it. Finding weaknesses in opponents, looking for patterns of play to counter them, in which circumstances there is acceptable risk of fullback attacking, players getting tired etc - I'd say at the very least this it the level in which pro football analysts operate. If they are not using AI to detect change in shape or playing style from opponent, I'd expect them to start doing it very soon.
@Teja as a ML specialist, where do you see applications for this technology in football?
Thanks @Cheimoon for the notification, I would've missed a very interesting thread.
I'd be shocked if any club across all top leagues doesn't have an analyst team in place. I mean, I have a friend who works as second coach/ analyst in 2nd tier woman volleyball team in Poland.
Last edited: