Brentford, Liverpool and the coming of age of football analytics?

[Teja]

Brentford

I ran into a great Twitter thread today that talked about how Brentford moneyballed their way to the PL. Very much worth a read in its entirety. But the summary is:

1. Owner with background in finance / math took over both Brentford and FC Midtjylland.
2. FC Midtjylland was a test club where they would experiment with different ideas and see what worked. The ideas that worked would be used in Brentford, the ideas that didn't would be discarded.
3. Ignore short term results entirely. Individual games are ruled by luck and randomness. Develop a series of team performance metrics (of which xG is one) to see if they were making progress.
4. Had lots of transfer success by finding undervalued players (Benrahma, Watkins, Maupay)

Tweet

— Twitter API (@user) date

Liverpool

NYTimes profile on Liverpool's moneyball approach: https://www.nytimes.com/2020/06/26/sports/soccer/liverpool-premier-league-title.html

I'm a software engineer by trade and work in the ML space and I see a lot of exciting things happening football data right now. This forum has picked up on some trends and folks are using player cards from Opta etc. and looking at metrics like xG, key passes. However, just in the past two years we've had tremendous advancements in the space.

1. Computer vision to estimate pose / detect events (passes, dribbles etc.)
2. Analytics to go back and estimate value of each player action. (See: expected threat / xT)
3. Pitch control estimation using things like Voroni diagrams.

Liverpool and Chelsea have been quite publicly working with the football analytics community. Liverpool for instance noticed a bunch of researchers doing analysis on broadcast video and just invited them to put some cameras at Anfield to get better quality video.

Liverpool talking about how they use realtime analytics to affect the outcome of games: https://www.liverpool.com/liverpool-fc-news/features/liverpool-transfer-news-jurgen-klopp-17569689

Pitch control visualizations. You can see how a live pitch control visualization can almost tell you where the spaces are. (Or on the flip side, spaces you need to plug)

So what's the point?

I'm not sure there is one tbh. If I had to make some:

As a forum, I think there's a lot of distrust towards football data (and probably for good reason). If you don't take away anything else, just take away the fact that humans are inherently very biased and data can help clear that clutter.

If there are likeminded people here, I'd love to stay in touch on the Caf / on Twitter to share research papers, interesting reads etc. There are a precious few folks doing analytics on United performances. Most people I run into on Twitter seem to do City / Barca / Arsenal etc.

I worry that traditional clubs like United, Barca, RM are falling behind in this regard, but have seen some chatter on Twitter around the Statsbomb guys consulting for Barcelona on transfers etc. I think it'll take teams 5+ years to catch up to where the Liverpool analytics department is at.

 

[Varun]

Definitely interested in learning more about this and reading whatever you have. Please share.

Re the forum's love hate relationship with data. I think there are 2 reasons, 1- what you said about a mistrust of how useful it is but more importantly 2- number of posters just picking data off a site and then using them to compare players without any context. I work with data for a living and clients would laugh you off the room if you used data the way some guys do here without applying any context on player or team's play style. Data can be sliced to basically tell pretty much any story you want. This kind of 'analysis' just puts other posters off even further and leads to discarding of numbers rather than good discussion on what it means.

 

[sglowrider]

We have Statman Dave.

Enuf said.

 

[roonster09]

UtdArena is very good for analyzing ManUtd, with lot of info that isn't available.

Yeah he are behind Liverpool and other clubs, apparently we have upped our game in that area

Also I think some 12=13 clubs use statsbomb or struck some sort of partnership with them.

 

[11101]

I'd be interested in seeing more on our club, for sure. Its worthy of it's own thread at least. I work loosely in that field too.

I think the mistrust of data on this forum is down to people using the wrong data. Stats from places like whoscored.com is too limited to be of any use but is posted as though its providing insight.

 

[Zen86]

Definitely interested in learning more about this and reading whatever you have. Please share.

Re the forum's love hate relationship with data. I think there are 2 reasons, 1- what you said about a mistrust of how useful it is but more importantly 2- number of posters just picking data off a site and then using them to compare players without any context. I work with data for a living and clients would laugh you off the room if you used data the way some guys do here without applying any context on player or team's play style. Data can be sliced to basically tell pretty much any story you want. This kind of 'analysis' just puts other posters off even further and leads to discarding of numbers rather than good discussion on what it means.

This. I don’t think anyone distrusts data, simply the way it’s used. It should be used objectively, which is very rarely the case here.

 

[KM]

I'm pretty sure there was news in 2019 that we had revamped our analytics department and had two or three new faces in. Maybe there's work going background to improve this department of our club.

 

[Maagge]

Also worth noting that Midtjylland has won their first three championships ('15, '18 and '20) and cup ('19) since they were bought. I think they've only placed outside the top 3 once in that time.

 

[ClaytonBlackmoorLeftPeg]

Definitely interested in learning more about this and reading whatever you have. Please share.

Re the forum's love hate relationship with data. I think there are 2 reasons, 1- what you said about a mistrust of how useful it is but more importantly 2- number of posters just picking data off a site and then using them to compare players without any context. I work with data for a living and clients would laugh you off the room if you used data the way some guys do here without applying any context on player or team's play style. Data can be sliced to basically tell pretty much any story you want. This kind of 'analysis' just puts other posters off even further and leads to discarding of numbers rather than good discussion on what it means.

Good post and I agree entirely.

 

[VivaObertan]

@Teja I'm capable of visualising data however my web scraping skills suck and I'm a little lazy. Are there sources for Manchester United data? I tried to find some a year back for a project but found it difficult, even accepting I'd need to cleanse the dataset

 

[Iker Quesadillas]

number of posters just picking data off a site and then using them to compare players without any context. I work with data for a living and clients would laugh you off the room if you used data the way some guys do here without applying any context on player or team's play style. Data can be sliced to basically tell pretty much any story you want. This kind of 'analysis' just puts other posters off even further and leads to discarding of numbers rather than good discussion on what it means.

Yeah this is part of my issue with data. I'm not a data analyst, but I work in science, and I know that any model usually has a long list of caveats, a lot of different elements with margins of error and possible weaknesses, and that you should be measured in your conclusions. So when these complicated models are used to churn out a single number with no margin of error, I'm going to find it hard to take it too seriously, and this is used to argue X player is better than Y player, I'm going to be a little skeptical.

 

[Water Melon]

I'm pretty sure there was news in 2019 that we had revamped our analytics department and had two or three new faces in. Maybe there's work going background to improve this department of our club.

Hopefully some significant improvement on Moyesie's super Ipads.

 

[Pogue Mahone]

You have to wonder if the “real time analytics” stuff mentioned in the OP is the reason for Ole’s much derided obsession with checking his iPad?

 

[tenpoless]

Professional sports gambler? what

 

[Classical Mechanic]

KDB's team used data analytics of his performances in his recent contract negotiation with City.

 

[Davie Moyes]

You have to wonder if the “real time analytics” stuff mentioned in the OP is the reason for Ole’s much derided obsession with checking his iPad?

Great point Pogue. Would make sense and now thinking about it what else could it be.

 

[sullydnl]

Regarding this part of the Brentford tweet:

Brentford FC started to look more closely at “expected goals” rather than how many goals a player actually scored.

Their theory? In a low-scoring sport that is skewed by randomness & luck, the quality & quantity of chances created during a match mattered more.

I would have thought the value of xG would be evident to all top level teams at this point. It's fairly basic stuff.

Absolutely agree with this though:

Definitely interested in learning more about this and reading whatever you have. Please share.

Re the forum's love hate relationship with data. I think there are 2 reasons, 1- what you said about a mistrust of how useful it is but more importantly 2- number of posters just picking data off a site and then using them to compare players without any context. I work with data for a living and clients would laugh you off the room if you used data the way some guys do here without applying any context on player or team's play style. Data can be sliced to basically tell pretty much any story you want. This kind of 'analysis' just puts other posters off even further and leads to discarding of numbers rather than good discussion on what it means.

In particular I see some posters directly comparing players statistically as if raw numbers will conclusively tell you which players are better at whatever aspect of the game (be it passing, tackling, aerial ability, whatever). As if all data analysts trying to identify a player who would improve us aerially would have to do is sort defenders by "percentage of aerial duels won" and work their way down the list. Football as seen through the lense of whoscored ratings and squawka comparisons.

When in reality there's a reason clubs hire professionals in this regard rather than any idiot who has access to fbref. The skill is in properly contextualising the data, not in looking it up. I don't expect football forum posters to have that skillset but it would make for better discussions if they understood the limitations in their ability to use stats.

 

[JPRouve]

You have to wonder if the “real time analytics” stuff mentioned in the OP is the reason for Ole’s much derided obsession with checking his iPad?

This may interest you, it concerns Rugby but it applies to all sports.

 

[anant]

@Teja I'm capable of visualising data however my web scraping skills suck and I'm a little lazy. Are there sources for Manchester United data? I tried to find some a year back for a project but found it difficult, even accepting I'd need to cleanse the dataset

There are ways to scrape the whoscored data. Think you'd have to use R-Selenium or some other way to automate scraping for all games, else it's going to be an extremely tiring process.

Having said that, whoscored data is limited in some ways - it can tell when a tackle was made, success and failure, etc. but things like press, jockey, closing down actions are unavailable as they don't capture the positions of all players at a given time on tv frame (something Statsbomb has now started providing to their clients)

 

[Pogue Mahone]

This may interest you, it concerns Rugby but it applies to all sports.

Very cool. I listened to an interview with the lead medic on the Irish rugby team the other day. The sport has really embraced technology. He was talking about data on concussive impacts taken from micro-chips in gumshields!

 

[pascell]

Definitely interested in learning more about this and reading whatever you have. Please share.

Re the forum's love hate relationship with data. I think there are 2 reasons, 1- what you said about a mistrust of how useful it is but more importantly 2- number of posters just picking data off a site and then using them to compare players without any context. I work with data for a living and clients would laugh you off the room if you used data the way some guys do here without applying any context on player or team's play style. Data can be sliced to basically tell pretty much any story you want. This kind of 'analysis' just puts other posters off even further and leads to discarding of numbers rather than good discussion on what it means.

Absolutely agree, that's why I just don't pay attention when people are comparing players that play in different leagues, teams etc. They all have different styles, some players play for an superior team in a league were they get more space to roam to effect a game, or for an inferior team that defend more, so their defensive stats are going to be skewed.

 

[NicolaSacco]

There are ways to scrape the whoscored data. Think you'd have to use R-Selenium or some other way to automate scraping for all games, else it's going to be an extremely tiring process.

Having said that, whoscored data is limited in some ways - it can tell when a tackle was made, success and failure, etc. but things like press, jockey, closing down actions are unavailable as they don't capture the positions of all players at a given time on tv frame (something Statsbomb has now started providing to their clients)

This is a key bit for me. Take Wan Bissaka for example. He’s praised for his tackles, a lot of them last ditch ones, and he is legitimately excellent at them, but would another player with different positioning (a completely unmeasured and perhaps unmeasurable metric) simply have positioned himself so as not to need to make those tackles?

 

[anant]

This is a key bit for me. Take Wan Bissaka for example. He’s praised for his tackles, a lot of them last ditch ones, and he is legitimately excellent at them, but would another player with different positioning (a completely unmeasured and perhaps unmeasurable metric) simply have positioned himself so as not to need to make those tackles?

Not just AWB, but it's true for literally every defensive player and DM. Very rarely will you see the best defensive players have the best stats. Fred's defensive stats are better than those of Casemiro's numbers from his best ever season. Maguire statistically comes across as just ok from stats and so on. I dont think Carrick would have had some exceptional numbers as well while playing for us

 

[Adam-Utd]

You have to wonder if the “real time analytics” stuff mentioned in the OP is the reason for Ole’s much derided obsession with checking his iPad?

Pretty sure he's just watching a TV stream of the game to get a better view of incidents / replays a bit closer up. Maybe he has other info being given to him via an APP but he's often looking at it after a foul / shot / ref decision etc.

 

[Varun]

Not just AWB, but it's true for literally every defensive player and DM. Very rarely will you see the best defensive players have the best stats. Fred's defensive stats are better than those of Casemiro's numbers from his best ever season. Maguire statistically comes across as just ok from stats and so on. I dont think Carrick would have had some exceptional numbers as well while playing for us

There's that famous screen grab which takes the key defensive stats we all love from bramble and puts them right against vidic. Guess who looks like prime Maldini?

We've all seen that, I'm sure even the posters who love to randomly pick stats off a site for a ligue 1 CM they've never seen, put them against Fred on one tab, an obscure mid from La Liga they've never seen on another and start making definitive statements about how one's better than the others.

 

[roonster09]

This is a key bit for me. Take Wan Bissaka for example. He’s praised for his tackles, a lot of them last ditch ones, and he is legitimately excellent at them, but would another player with different positioning (a completely unmeasured and perhaps unmeasurable metric) simply have positioned himself so as not to need to make those tackles?

Its hilarious how this myth is taken as gospel. FBs will always have great tackling numbers because they face wingers 1v1. For CBs its different. So FBs no matter how good their positioning is, they have to win the tackle to stop winger who more often that not tries to take on FBs.

This is the prime example of this thread, people don't know how to use stats or give context to them, just go with "more tackles = bad positioning".

 

[Cheimoon]

Great thread! I'm very interested in this stuff, although I probably won't have anything to contribute myself. I work with statisticians though, if that helps any!

I'm also thinking @Borys will find this interesting. He used to run a good thread about match stats when he had more time. (Link)

 

[RobinLFC]

Its hilarious how this myth is taken as gospel. FBs will always have great tackling numbers because they face wingers 1v1. For CBs its different. So FBs no matter how good their positioning is, they have to win the tackle to stop winger who more often that not tries to take on FBs.

This is the prime example of this thread, people don't know how to use stats or give context to them, just go with "more tackles = bad positioning".

More succesful tackles doesn't equate better defender either though, if defender A attempts 200 and wins 50 yet defender B attempts 100 and wins 40, he has had a better success rate and potentially has been dribbled past less than defender A who was beaten on 150 of his tackles. It's not that black and white and you simply can't take one stat such as "tackles" without looking at the context. A fullback facing more 1v1s doesn't even mean he has to make more tackles, he can force a winger backwards or out wide with good positioning, he can force him into a bad pass to intercept him, force him into a bad cross, ...

The statement "no matter how good their positioning is, they have to win the tackle to stop the winger" is just simply not true.

 

[do.ob]

Do Liverpool or Brentford really play such a special role here? Or are they just the ones who have been featured in the press? I know for example that Sven Mislintat (co) founded his own data analysis company and I once heard a podcast with him where he spoke about how big a role such analysis plays in identifying transfer targets. I imagine nearly all clubs who can afford it rely on data analysis these days, it's just such a no-brainer.

 

[roonster09]

More succesful tackles doesn't equate better defender either though, if defender A attempts 200 and wins 50 yet defender B attempts 100 and wins 40, he has had a better success rate and potentially has been dribbled past less than defender A who was beaten on 150 of his tackles. It's not that black and white and you simply can't take one stat such as "tackles" without looking at the context. A fullback facing more 1v1s doesn't even mean he has to make more tackles, he can force a winger backwards or out wide with good positioning, he can force him into a bad pass to intercept him, force him into a bad cross, ...

The statement "no matter how good their positioning is, they have to win the tackle to stop the winger" is just simply not true.

It's nice in theory, it won't work on the pitch. All the pass, movement happens in fraction of second, you can't always positioning your body in such a way to force wingers outwide or control their movements as you want.

Also not sure who argued more tackles - better defender thing. I was arguing against the point that "tackling = bad positioning" myth, especially for FBs.

 

[sullydnl]

Absolutely agree, that's why I just don't pay attention when people are comparing players that play in different leagues, teams etc. They all have different styles, some players play for an superior team in a league were they get more space to roam to effect a game, or for an inferior team that defend more, so their defensive stats are going to be skewed.

Plus even say you can successfully identify a weakness in a player using those stats, you'd then have to examine the cause of the weakness (i.e. is it something coaching can address relatively easily or a permanent feature of their game) and how that weakness fits into the structure and style of your team (i.e. is it something that will be exposed regularly or something that will largely have little impact due to the team's style of play and the other players in the team).

 

[rotherham_red]

You have to wonder if the “real time analytics” stuff mentioned in the OP is the reason for Ole’s much derided obsession with checking his iPad?

It is. It was noted by the Athletic and UWS (probably the same person for both - Andy Mitten) that what the iPad was showing was the real-time stats.

 

[roonster09]

It is. It was noted by the Athletic and UWS (probably the same person for both - Andy Mitten) that what the iPad was showing was the real-time stats.

Showing real time stats won't be of much use isn't it, if they can give him aerial view of how the team is set up and where the gaps are, it would be useful in making the decisions.

I think we should out McKenna or Carrick on the top of OT roof, to give them top view of how the game is going on and where we can take advantage by using the spaces

 

[anant]

It's nice in theory, it won't work on the pitch. All the pass, movement happens in fraction of second, you can't always positioning your body in such a way to force wingers outwide or control their movements as you want.

Also not sure who argued more tackles - better defender thing. I was arguing against the point that "tackling = bad positioning" myth, especially for FBs.

Don't think there is any strong correlation between tackling and positioning for even a FB. The stat needs to be combined with how many crosses/ cut backs came in from each wing, where are the shots coming from, where are attackers entering the box from, etc.

If the FB is making more tackles but still the shots are coming from that direction (talking about a general case, not AWB), it might mean they're targeting the flank because of poor positioning or some other weakness

 

[TrustInOle]

You have to wonder if the “real time analytics” stuff mentioned in the OP is the reason for Ole’s much derided obsession with checking his iPad?

Great point, far from being an Ole basher but you'd think if this was the case we would find the space more between these low block defences? I personally feel he just likes seeing the shapes of the teams from a different vantage point. But then surely AWB wouldn't be caught out of position so much?

Damn, I'm throwing my own brain hrough a loop here!

 

[roonster09]

Don't think there is any strong correlation between tackling and positioning for even a FB. The stat needs to be combined with how many crosses/ cut backs came in from each wing, where are the shots coming from, where are attackers entering the box from, etc.

If the FB is making more tackles but still the shots are coming from that direction (talking about a general case, not AWB), it might mean they're targeting the flank because of poor positioning or some other weakness

I wouldn't draw that conclusion. Take ManUtd as example, no matter which team we palyed, we always attacked from left wing, not because opponent RB was weak, it's because we had only one functioning side.

Lot of factors to consider to make any conclusion, for example AWB is a great defender 1v1, almost nothing goes past him but overall he is good defender but nowhere near as his one v one defending quality. Shaw as a defender is better. But if we face tricky wingers, I always want them to go against AWB as you know he will block that side completely. When the cross comes in from the Left side, AWB's weakness is exposed as he is poor header of the ball and lacks concentration to defend far post.

 

[Oranges038]

Its hilarious how this myth is taken as gospel. FBs will always have great tackling numbers because they face wingers 1v1. For CBs its different. So FBs no matter how good their positioning is, they have to win the tackle to stop winger who more often that not tries to take on FBs.

This is the prime example of this thread, people don't know how to use stats or give context to them, just go with "more tackles = bad positioning".

With Wan Bissaka you don't need stats on tackles to know his positioning is suspect. All you have to do is watch him play.

 

[roonster09]

With Wan Bissaka you don't need stats on tackles to know his positioning is suspect. All you have to do is watch him play.

Yeah, sorry I didn't watch any games in last 2 seasons. Will try to catch the first game of next season.

 

[Oranges038]

Yeah, sorry I didn't watch any games in last 2 seasons. Will try to catch the first game of next season.

That's great. Make sure you have your speeadsheet ready.

 

[roonster09]

That's great. Make sure you have your speeadsheet ready.

Sure.

You make sure you buy "cliche for football beginner" book.