Football/Soccer/Futbol Analytics and Strategy

We all watch football, right? We see the goals, the tackles,

the final score. But what if we could dig a bit

deeper? Go beyond just the results.

Exactly. Really understand how teams

play. There's their underlying style.

Today, we're diving into football analytics to see how

numbers can reveal that hidden DNA.

Yeah, it's fascinating stuff. I mean, think about it.

Can you actually group teams based on what they do on the

pitch, statistically speaking? Forget wins and losses for a

second. So like using data to find teams

that genuinely play alike? Precisely.

There was this study out of of Sala University.

They took a massive amount of performance data from the Big 5

European leagues, Italy, Germany, France, Spain, England

for the 20/22/23 season. OK.

And they tried to see if meaningful groups or clusters of

teams would just naturally emerge from that data.

So the big question was, does the data itself reveal distinct

playing style? Yeah, it was the core idea,

yeah. Are there real statistically

backed categories of teams? That sounds like a huge

undertaking. How do you even begin to analyze

all that information? It must be incredibly complex.

Well, they used a specific technique called agglomerative

hierarchical cluster analysis. OK.

That sounds technical. Break that down a bit.

Right. Imagine starting with every

single team as its own little island.

Then step by step, you join. The two islands are teams that

are statistically most similar based on their performance data.

How do they measure similar? They use something called

Euclidean distance. It's basically a mathematical

way to calculate how close teams are in terms of their stats, and

then Ward's algorithm help decide which teams or groups to

merge at each step. Ward's algorithm.

Why that specific one? What's the benefit?

Good question. A key advantage is that you

don't have to decide beforehand. OK, we're looking for exactly

five types of teams. So it lets the data suggest the

number of groups. Exactly.

It builds this sort of family tree diagram, a dendrogram

showing how the teams link up. It makes it more intuitive to

see where the natural splits might be.

It's trying to keep the groups as tight and distinct as

possible as it builds them. That makes sense.

So you're not forcing the data into predefined boxes now?

Football matches involve so much, there must have been a ton

of different stats for each team.

Oh absolutely, and that brings up a common challenge in data

analysis dealing with lots of different measurements.

The so-called curse of dimensionality.

So before they started grouping they standardized the data.

Think of it like converting everything to the same scale so

you can compare say, possession stats fairly with tackling stats

even though so they're measured differently.

OK. Leveling the playing field

statistically. So how did they finally decide

on the number of distinct groups?

They looked closely at that dendrogram, that family tree.

You look for points where joining two groups requires a

really big jump in difference or dissimilarity.

Suggesting those two groups aren't that alike.

Precisely that suggests a natural place to stop merging.

They also used a statistical measure, the silhouette

coefficient, to check how well each team actually fit into the

cluster it ended up in, like how much does it belong there versus

the next closest cluster. Right.

And after all that number crunching, what did they find?

Did they reveal like 10 different tactical philosophies?

Actually, the main result pointed towards just two primary

clusters or groups of teams. Only two?

I might have expected more nuance maybe.

It was interesting the first of these two clusters was actually

quite clear cut. It mainly contained the top

performing teams from each of those five big league.

So the powerhouses, generally speaking.

Yeah, teams like Manchester City, Real Madrid, Bayern

Munich, PSG, they tended to group together.

That cluster was more consistent, easier to interpret.

OK, that makes intuitive sense. What about the second cluster

then? That one was, well, messier.

It was much larger, contained a much wider range of teams from

mid table to lower down, and it was harder to pin down a single

unifying playing style within that group.

More varied, less defined. Did they try to like map this

out, visualize it? You mentioned principal

components earlier. They did.

They tried to squash all that complex multi dimensional data

down onto a simple 2 dimensional graph.

Basically to see if you can visually see two distinct blobs

representing the clusters. And could you?

Not really, clearly, no. The separation wasn't super

obvious on that 2D plot, which actually aligns with that

silhouette measure we talked about.

It was moderate about .31, suggesting the clusters weren't

perfectly distinct. Because you lose information

when you simplify it that much. Exactly those first two

principal components. Those two dimensions on the

graph only captured about 62% of the total variance in the

original data, so it gives you a hint, but you have to be

cautious interpreting that kind of simplified picture.

OK. So maybe not perfectly defined

clusters, but were there still clear statistical differences

between the average team in cluster one versus cluster 20?

Definitely big differences. For instance, average possession

that the top team cluster had on average about 10 percentage

points more possession than the second cluster. 10% is quite a

lot in football terms. It is, and goals scored a huge

difference. The first cluster averaged

around 25 more goals over the season and then the second

cluster. Again, aligns with what you'd

expect from top teams versus the rest.

Generally more ball, more goals. Anything else stand out?

Yeah, tackling locations were interesting.

The second, more varied Cluster averaged about 30-4 more tackles

per team, way back in their own defensive third of the pitch.

So defending deeper. Right, while the first cluster

at the top, teams averaged 22 more tackles, way up in the

attacking third. Winning the ball back higher up

the pitch. Exactly.

So you see this data-driven picture emerging cluster one.

The top teams tend to have more possessions, score significantly

more goals, and are more active in winning the ball back near

the opponent's goal. Cluster 2 tends to defend deeper

and score less. It really does mirror that

general perception we have, but now it's backed by the numbers

across five major leagues. Precisely.

It's like the data confirms some of that intuition.

And the researchers did suggest that maybe other clustering

methods, perhaps ones that are left for more overlap or focus

on specific aspects, might uncover even finer grained

patterns in future work. It's fascinating, but of course

a team's approach isn't static, is it?

It changes during a match based on the opponent the score.

Absolutely. And that brings us to the

dynamics within a game. Researchers Lori Shaw and Mark

Lichtman have done work looking at this, highlighting how

crucial formations are as tactical tools for managers.

Like Jurgen Klopp adjusting Liverpool's shape depending on

whether they have the ball or not.

Exactly that. Klopp himself talked about

creating different formations for defense and attack, trying

to get players like, say, Muhammad Sal or Darwin Nunez

into the optimal positions for each phase of play.

Formations have evolved so much over time, it's wild to think

about those early days. Jonathan Wilson's book Inverting

the Pyramid mentions that 1872 Scotland versus England game.

Oh yeah, Scotland apparently playing a 226 and England A127.

Imagine that now. Worlds away from the complex

structures you see with players like Kevin De Bruyne or Antoine

Griezmann orchestrating things. So how did Shaw and Klickman

actually measure these in game formation shifts?

Their approach was quite clever. They focused on the 10 outfield

players as a kind of coherent block.

They analyzed possessions that lasted at least 5 seconds.

For one, starting from dead balls, they looked at the

average position of the ball, but only when it was outside the

defending team's defensive quarter, trying to capture

attacking or midfield shape. Then they aggregated this data

into short 2 minute Windows 2. Minutes, so quite granular.

Yeah, and crucially, if a substitution happened, that 2

minute window ended immediately starting a new one.

They also excluded the goalkeeper.

This gave him about 10 snapshots of offensive formation and 10 of

defensive formation per team per match.

So you get a sense of the tactical flow and adjustments

throughout the game. And this kind of detailed

tactical data is becoming more accessible now, right?

It really is. You've got platforms like

Playmaker, maker.aifbrafe.com. They offer quite detailed

formation data and tactical analysis.

Which naturally leads into how teams use this kind of analysis

for opponent scouting. It must be invaluable.

Usually valuable and this is where insights from people like

Sergio at Naxport, a sports analysis company, come in.

He really emphasizes that the specific stats you look at, the

key performance indicators or KPI's have to change depending

on the opponent. You wouldn't.

Analyze Pep Guardiola's Manchester City the same way

you'd analyze, say, A-Team known for counter attacking.

Exactly. It depends on their style, their

typical formation, Sergio says. The main goal is always twofold

Identify the problems the opponent could cause you and

figure out how to exploit their potential weaknesses.

How do they do that in practice? It's usually a mix.

You got the qualitative side watching video footage, getting

a feel for their patterns, and then the quantitative side using

specific KPIs to confirm those patterns and find

vulnerabilities. The aim is to neutralize their

threats while setting up your team to execute its own game

plan effectively. And focusing on relevant games.

Yes, especially looking at matches where the opponent

played against teams that use tactics similar to your own that

gives the most relevant insights.

What kind of defensive Kelty eyes does Sergio highlight as

being particularly important when scouting an opponent?

Several key things. First, where do they lose the

ball in their own half, especially looking at center

backs, maybe like Virgil Van Dyke or wingers like Venetius

Junior. Who loses it and how is it under

pressure? A bad pass.

OK, turnover is in dangerous areas. 2nd, aerial battles.

One, particularly if the opponent likes to play long

balls. Third, how do they boon up?

Play from the back using video linked maps to see pass types

and where they succeed or fail. Video linked map, so you see the

stat and the corresponding video clip.

Exactly. 4th areas where they attack frequently versus areas

where they make passing errors. Again, map based. 5th

identifying the zones and specific players who deliver the

most dangerous key passes, and finally mapping where they lose

possession in the final third. Can you give a concrete example

of how this might inform A tactical decision?

Sure. Let's say the data shows that

when an opponent's goalkeeper plays a short goal kick to a

specific center back, maybe enter Militone for instance,

that player struggles under immediate pressure and often

gives the ball away. Well, you might then instruct

your forward, maybe someone like Harry Kane, to position himself

specifically to press that center back instantly on goal

kicks, trying to force an error or make the keeper plate to

someone else who might be less comfortable.

That really shows the direct link between data insight and on

pitch action. What about using KPI's to plan

your own attack against an opponent?

For organizing your attack, understanding the opponent's

defensive spatial structure is crucial.

What does that mean? Spatial structure.

It's about the distances between their defenders, how high up the

pitch their defensive line sits, their overall positioning and,

crucially, where the exploitable spaces are between or behind

them. Finding the gaps.

Finding the gaps When analyzing how they defend goal kicks, you

look at their spatial setup again, how their center backs do

in the air, where they tend to recover the ball, and

importantly, areas where they don't make many tackles or

interceptions, the potential weak zones to target.

And an open play. Spatial structure again for both

initiating attacks and progressing the ball.

Then when it comes to the shooting phase, you'd look at

things like they're expected goals conceded XG.

How many goals would you expect them to concede based on the

shots they face? Also, what percentage of crosses

against them lead to shots? Where do they concede crosses

from most often? Where do opponents tend to shoot

from against them? Where are their most dangerous

key passes coming from that you need to shut down?

And where do they tend to lose the ball near their own penalty

area? Maps linked to video are really

useful here too. And.

Of course, set pieces are a whole other battleground.

Absolutely. For your attacking set pieces,

you analyze where the opponent is vulnerable, which areas,

which of their players are weaker targets, what types of

crosses work best against their setup.

Defensively, you need to know where they are most effective

from set pieces, frequent areas, dangerous shooters or targets.

It seems like everything connects.

Defense, Offense, Transitions. Sergio makes that point, too.

Understanding how a team is organized defensively and

offensively helps you predict how they'll likely react during

transitions, when they lose the ball and when they win it back.

The patterns become clearer. It's such a systematic approach,

but then you have the other side of football analysis, the real

time fan reaction, which can be quite different.

Totally different, and a great example is looking at

discussions around a big game like a recent Classico between

FC Barcelona and Real Madrid. Checking out fan forms like

Reddit gives you that immediate, often emotional pulse.

And that recent one was a high scoring affair, wasn't it?

It was ended 4/3 to Barcelona goals for Barsa from Eric

Garcia, Lamin Yamal, Rafina got 2 and for me in Lopez and for

Real Madrid a hat trick from Killian Boppe.

Wow and Boppe hat trick in a losing effort.

What did the underlying numbers, the expected goals XG say about

that game? Interesting contrast,

Barcelona's XG was 3.26 while Real Madrid's was 2.06.

Suggesting Barsa created the better quality chances overall

despite the close scoreline. That's what the XG implies,

yeah. So what were the Real Madrid

fans saying online after that? That one beyond just

acknowledging Mbappe's. Goal well, obviously praise for

Mbappe's individual brilliance, you know, even in defeat.

But quite a lot of criticism aimed at their play down the

right flank, particularly involving Lucas Vasquez.

There were also questions about Carlo Enchiladi's tactics,

specifically playing some players like Art of Guler,

supposedly out of position as a right winger.

Lots of debate too, about disallowed goals, VAR reviews,

handball decisions. The usual controversy is

amplified in the classic. Always drama there.

Always defensively there were real concerns voiced about the

team's performance and also about squad depth.

General frustration with tactics, player choices, you

know. The overall feeling was

disappointment, despite the hat trick suggesting fans saw wider

team problems. Any positives mentioned besides

Mcbuper? Some shouts for Luca Madrich,

still putting in shifts despite his age, and definitely calls

for defensive reinforcements in the next transfer window.

It really is a stark contrast, isn't it, that immediate

passionate fan reaction versus the kind of structured data

heavy analysis we were just talking.

About absolutely. Fans react to what they see, the

big moments, the result, often with a lot of emotion.

The analytical approach tries to step back, look at the bigger

picture, the underlying trends more objectively.

Both offer value, just different kinds.

And looking at Barcelona beyond that classical win, their

overall season in La Liga 202425 looked pretty formidable based

on the stats. It did across the 38 games they

banged in 100 and two goals only let in 39.

That's a goal difference of plus 63, which is very strong.

Very strong indeed. Any other notable stats?

Clean sheets. Yep, 13 clean sheets and they

only failed to score in two matches all season.

Very consistent offensively. What about the overall number of

goals in their games? Any tendencies there?

Yeah, the data suggested their games weren't always goal fests.

About 29% of their matches actually had under 2.5 goals in

total. So maybe indicates good

defensive control alongside the high scoring.

Interesting balance and you can get stats on pretty much

anything these days, right? Corners cards.

Oh yeah. The data covers average total

corners per match, corners, four corners against, same for cards,

average total cards, yellows, Reds, even broken down by first

and second-half, plus average cards for Barsa players versus

average cards for their opponents when playing them.

A complete statistical profile and that classical when it felt

like more than just three points for Barcelona, didn't it?

There was talk about it marking a shift under Hansi Flick.

There was some reports, like one in Sports Illustrated, suggested

that victory could be the catalyst for a new era of

success under Flick. They apparently came really

close to a perfect El Clasico record against Madrid that

season, which is almost unheard of.

Weren't there some doubts when Flick first took over about him,

about the squad? Yes, initially there were some

questions, but the narrative shifted.

The reports highlighted key players like Pedry and Robert

Levandowski finding form again, plus the huge impact of young

Lamezia talents like Lamine Yamal really stepping up.

So a combination of experience and youth clicking under new

management. That seems to be the story, a

real resurgence attributed to Fix leadership, the positive

atmosphere he created and that pipeline of young talent coming

through. Sounds like things are looking

up for them. The reports certainly paint a

picture of a promising future. Yeah, flicks guidance plus those

exciting youngsters. So when you put it all together,

from these broad statistical clusters identifying team styles

across leagues, to the nitty gritty KPI's used for scouting

opponents week to week, and even down to the real time fan

chatter after a massive game, data analysis really gives us so

many different lenses to view football through.

Absolutely. Each layer offers something

different, you know, whether you're just a fan wanting to

understand why your team plays a certain way, or you're a coach

looking for that tiny edge, the insights are there.

It does make you wonder, doesn't it?

How much are managers like, well, you're going to Clop or

Hansi Flick actually using these large scale data insights versus

their own intuition and observation when making big

tactical calls or evaluating players?

That's the $1,000,000 question, isn't it?

Yeah. And as the data gets even

richer, even more granular, what other hidden patterns are we

going to uncover? What new predictions might

become possible? Definitely food for thought.

Football/Soccer/Futbol Analytics and Strategy

About this episode