Ben Griffis Profile picture
Statistician | Visualizing & analyzing soccer from around the globe | Run & @BuLiAnalytics | Views my own | DMs open, contact for freelance
Aug 11 28 tweets 11 min read
The unique role of the goalkeeper in Christian Titz's Magdeburg tactics ✍️

Titz uses his GK, Dominik Reimann, in a revolutionary way… almost as a CB in buildup

This 🧵 will show the key elements, benefits, and risks of Titz's tactic

#EINMALIMMER #Bundesliga Before I get started, just want to say this won't be a complete breakdown of Magdeburg's buildup. Just a ton of examples showing the positions the GK takes up, the types of passes he's instructed, the patterns with those passes, and more.
Aug 10 7 tweets 2 min read
I'm assuming people who are saying how this is terrible, or how there's a lot wrong with this, didn't read the methodology section or don't want to ask questions on what they don't understand (fyi not everyone will understand clustering without asking & that's fine lol) Essentially, this study takes a ton of metrics from FBRef and groups players based on their specific numbers recorded in all those metrics. They do this in two steps to come up with these "buckets" that they subjectively named
Aug 10 4 tweets 1 min read
Honestly, All-Star week challenges like this should be an American sporting novelty that spreads around the world 😂

Would love to see like, a Kane vs Salah volley challenge, or a dribble obstacle course 🤣 Like… I’m sorry but European leagues are missing out so fucking much. Look at this beaut
Aug 10 4 tweets 2 min read
Amazing thread from @MarcoLai_23 read it all!

This part specifically is spot on. Sessegnon ( and presume Perisic will be too) are high, uninvolved in buildup because Davies is higher & wider than Romero, leading to a lopsided buildup shape.

Suits Sess, Perisic, Udogie, Davies Throughout this thread we learn a ton about his attacking qualities, which in Conte's Spurs (if he's still here next season lol) is probably more important for the LWB than defense, given Davies plays in the half-space and is a great defender
Jul 5 21 tweets 19 min read
Analyzing all #BarclaysWSL teams' 1st pass after their kickoffs from 18/19 to 20/21

🔑 Takeaway: Some teams do have different tactics overall &/or for different game states

🧵 is alphabetical, so starting with @ArsenalWFC

Always more long & less accurate passes when losing! For @AVWFCOfficial there's just the one season.

Ignore winning game states since there were only 2... but they exclusively played very long balls up the pitch when drawing, but were more conservative while losing (with some long balls as well, but not just long passes)

Jul 5 4 tweets 2 min read
Playing around with harmonic means in the #KLeague

Here, the darker green shows players who are both involved in attack & defense and also have a good balance between the two phases, not being exclusively involved in just one

Incheon's Kim Bo-sub has the largest harmonic mean Harmonic means focus on both the magnitude of two numbers and also the balance between their levels.

Players involved almost exclusively in attack or defense will have lower numbers regardless of the level of their main contribution, because they don't have balance
Jul 4 15 tweets 5 min read
Inspired by a post by @theOptamist:

Here's a quick analysis of Barcelona's first pass after their kickoffs, from 16/17 (when the rule changed & kickoff didn't have to go forward) to 20/21. Only games Messi played in

Data: @StatsBomb open data


See 🧵 for more 📊 Graph showing the locations... What's interesting here is that over these 5 seasons, the avg length of the pass after kickoff was typically quite a bit longer when they were winning than when they were losing...

And what's more, those passes were also much less accurate on avg as well
Jun 26 12 tweets 3 min read
We possession-adjust defensive metrics often, but never offensive ones...

What does it look like if we try to adjust goals & assists by the number of touches a players' team has on average per match? 🤔

All info on graphs 📊, more detail in thread 🧵 ImageImage The logic behind this:

The more possession a team has, the more opportunity their players have to both score or create goals.

Less possession typically means less opportunity.

Less opportunity will translate to fewer goals and assists, and vice versa.
May 14 14 tweets 7 min read
I've been researching a number of soccer data sites recently to see what their data use/scraping policies are.

Wanted to share what I've found so far in this thread.

A critical factor to consider when using third party (i.e. not your own created dataset) is use ethics & terms .@fbref, Starting w/ probably the biggest one.


FBRef allows you to scrape, but can block you if you send more than 1 request every 3 seconds.

You will be blocked for about half a day... Enough time to add some code to 🆙 request time 😉…
May 13 23 tweets 20 min read
I have some free time this afternoon, so I think I'll live tweet some of my tactics thoughts on the first leg of Luton Town - Huddersfield

I'll just keep adding tweets to this thread each time I've got something 👍

#LUTHUD | #COYH | #HTAFC Should be 3-5-2 vs 3-5-2, will be an interesting one.

The noise coming through my headphones is nuts. Something special about low capacity stadiums
May 13 4 tweets 5 min read
Big matches coming up in #LaLigaSmartBank today & tomorrow!

4 points between 1st & 3rd (@UDAlmeria_Eng , @SDEibar & @realvalladolidE ), 3 matches to go! Top 2 automatically promoted.

These images show simulated result probabilities & each match's scoreline probabilities #RealSociedadBAlmería

@U_D_Almeria are currently in 1st & a win here would guarantee promotion!

Most likely result is 0-1 to Almería.

Chances of Almería scoring at least 1: 79%
Chances of Real Sociedad B scoring at least 1: 55%

Chances of both teams scoring at least 1: 44%
May 12 4 tweets 2 min read
Massive #NLD today.

Wanted to see if Harry Kane performs any differently against Arsenal in the Premier League than against other teams...

Essentially, no 😂 Using a simulation of a normal distribution, his G-xG is pretty similar.

T-test also insignificant @ .05

#coys #TOTARS Image This basically simulates 100,000 matches of each, using his mean G-xG/90 & standard deviation.

Only looks at PL games where he played at least 45 minutes.

Performed a Monte Carlo simulation in R & also a quick unequal variance t-test in Excel which was non-significant at 95%