We are coming up on the three-year anniversary of StatsBomb. During that time (and despite a lengthy absence due to that whole “working for football clubs” thing), I have published more than 200 items in the SB article database. Plenty of these are fluff, like gifolutions during the World Cup, or weekly follow-ups on basic predictive models from back in the day, but plenty of these are early prototypes for concepts I would later successfully apply inside of football clubs.
Given the fact that there are a lot of new followers on my Twitter account since I started writing, and the fact that so many people seem to have missed the early years of football stats writing, I figure now is a good time to review my recent past. I’m just going to do it on my own work for now, which will keep me out of trouble for criticizing others work, and probably keep this to a (barely) manageable load. This is also just StatsBomb based and skips over work I did for The Mirror, The Guardian, Opta, and various guest spots on blogs over the years.
Doing this completely chronologically is going to be a mess, so I’ll break them into subject headings and then work from there. I’ll also add some context about why pieces are included here, how I feel about them years later, and what stuff was just plain wrong about.
All of the material here is me learning publicly, and then writing about it on the fly. There is plenty in the early work that is just wrong. Given that knowledge, there is probably plenty in my current work that is also wrong, but hopefully it’s at least a little less wrong than I was before.
How Do Teams Create Better Chances? Three years later, almost all the ideas in this piece still guide my thinking about football as a whole.
Pace and Margin for Error Early on, there was a lot of application of hockey metrics to football, spurred on by really interesting work from Gabriel Desjardins, James Grayson, and StatsBomb co-founder Benjamin Pugsley. One of the primary metrics used was TSR or Total Shots Ratio.
As of this article, I was having a really weird time applying TSR to football outside of the Premier League and La Liga. There are a variety of reasons for this, but a big factor in that is this concept of pace. It’s around this point I stopped using ratios at all, even in basic shots work, and moved to differentials.
I mention this one because I think I was the first to publish about it (though I suspect some clubs used it ages ahead of this), and also because it’s not defined as “pressing.”
Pressing is a set of actions designed to put pressure on the ball and yield [things]. Elements of success can be found in defensive actions, but not all successful pressure yields a tackle or an interception. Without tracking data, teams need to collect explicit data outside of the normal Opta event data set to examine pressing in detail.
Lower opposition passing percentages are a result or outcome, and can be caused by a variety of factors including tactics, player ability, pitch, weather, etc. It’s correlated to pressing, but it’s not pressing in and of itself.
If you want to examine team pressing beyond the occurrence of defensive actions in specific places, I would look at PPDA from Colin Trainor and Defensive Distance from Garry Gelade as well as opposition passing percentages. Introducing Possession-Adjusted Player Stats I still use these, though some find them controversial (and others pointless). There are probably better ways to go about it now, but it gets complicated pretty quickly and you need to be a data ninja to do the analysis.
Long story short, it’s one of the few ways you can make defensive actions apply to things you care about with per90 data.
One from the post-Brentford and Midtjylland era, this piece is actually a chapter from a book I will likely never publish. It took a year of working on the ideas to get to this final incarnation. It’s also based off the work of countless other people who went before me, and is not groundbreaking. The focus here was instead on clarity and brutal practicality.
If you didn’t understand the concept before, it’s a good introduction to shot locations and expected goals and covers a lot of the various forms of pushback I received on these concepts inside of football. If you already knew about expected goals and the like – which at the point of this writing is not nearly as widespread in football as analysts and even some journalists seem to think – it provides practical, real-world examples on how to teach these concepts to players in training and through what behavioral economics would call a nudge (the shot rings).
Increased data use in football is inevitable. This piece uses a Big Short metaphor to explain why.
It started out with seeing an NBA All-Star poster by Ramimo. It ended with months of reading up on data visualization and learning Photoshop to finally produce player radar charts. This is the first, mostly awful piece that introduced them.
Iteration would then come fast and furious. I split the radars into positional archetypes and added two standard deviations for the boundaries (A.K.A. scienced the shit out of them), and voila – lovely little bite-size player evaluation pictures for the footballing world.
These have been improved in subsequent years, but the public versions are mostly the same as what I was making at the end of 2014. The new versions are only available to paying customers, largely because there are proprietary metrics in there and of which there are currently exactly none.
And if you hate them, you have my apologies, because they seem to have spawned a variety of imitators not only in football, but across various sports.
Called “MK” because Marek Kwiatkowski did a lot of the heavy lifting for these, and there are many different variations of shot maps out there. The design diary for how these were created is here. The link above is a more practical use demonstration for using them with teams. You can also use them in player evaluation.
In terms of taking stats and applying them to football in a useful, beautiful way, I think these are at least as successful as the radars, if not better.
Opening the Door to Player Analytics in Football The first article I ever wrote about player stats and football. It all started with Max Kruse, who would go on to
a) Play in the CL and EL
b) Make a final table at the World Series of Poker
c) Leave £60,000 cash in the back of a taxi.
d) Allegedly get in trouble for his late night poker life style and eating too much Nutella.
This is hugely important and hugely misunderstood. I stand by taking Manchester City to task for their transfer buys in 13 and 14 not because of the immediate impact, but because it cost them heavily in the latter years of those contracts. Good thing FFP is irrelevant or they would be in a world of hurt trying to rebuild for Pep this summer.
What if you wrote a piece that read like good analysis but was completely wrong? Well, I totally did that and it looks just like this one.
The reason why it was wrong was a fundamental misunderstanding of the game at this point on my part. You see, I had been sucked in by the “all shots are equal” mentality of hockey analysis. The problem here is that while hockey and football overlap in many places, shot quality is one where there is a massive divergence. I didn’t understand that yet, and because of that I thought Suarez kept killing Liverpool attacks by taking poor shots.
The good news is that being wrong about this forced me to rebuild how I evaluate the game from the ground up. The bad news is that this mis-step blew up some of my credibility (especially among Liverpool fans), and I got to hear about it constantly both on Twitter and on the podcast from Pugsley. *sad trombone noises*
It is January 2014 and Arsenal need a new forward. I poked around the data to look at Draxler (hot in the media), Griezmann, Lacazette, and… Aboubakar? This sort of statistical shopping (but using more advanced, modern metrics) is immediately applicable at the club level.
I am public with my work and I’m not afraid of being wrong. If you make a lot of bets, you will get winners and losers. If you make a lot of player and team predictions, you will get the same. However, when you are wrong in this big a fashion, you are going to have to eat it. It was a great learning experience, and obviously I would not make the same mistake again.
Life lesson: If you are going to have to eat shit, don’t nibble.
Can we take statistics and use them to find young players that are going to develop into Champions League players? This research would consume the rest of my summer until I was hired by Smartodds, and to some extent still does.
A huge challenge, but hugely rewarding if you pull it off.
I had Morata number 1 and Memphis number 2 that summer for very young players that would likely be future stars. Number 3 was Lucas Piazon, who has had some serious issues in his personal life and looks like a complete bust.
Trying to predict the future is tricky, and there are absolutely, positively going to be failures. On the other hand, something like half of all transfers of mature players “fail” as well, meaning you don’t have to raise the bar that much to improve a Premier League club’s transfer business enough to save tens of millions of pounds yearly.
You would think someone who is a huge proponent of statistics would have strong feelings about traditional scouting. You would be correct. The article is adapted from a presentation I gave to Science + Football that encompasses two years developing a lot of the initial concepts you see written about above, but applying the theories inside the world of football.
It also explains why we almost never scouted players live at BFC, and how use of stats is pretty much the same as incorporating video services into your scouting, something that isn’t remotely controversial in the modern day.
The unexpected take away at the end (which is the fault of the clickbait title) is actually that good scouts are extremely valuable, but most football teams can apply their skills far better than they currently do.
A synthesis piece about player analysis and young player development. The second half talks about what clubs can do to better evaluate prospects coming from their academy and into the first team, which is yet another area I think most football clubs can improve dramatically.
Attacking Wrinkles – Manchester City and Barcelona
I really like these types of articles, though they are a massive time sink to produce. I think the reason I get so excited by them is because it felt like doing real football work, and it definitely paid off once I needed to do presentations to coaches and Directors of Football inside the clubs.
Here’s the transition from football fan to coach-analyst: Start to see games in sequences of possession. What do smart teams repeatedly do with the ball, immediately after they win it back? In the final third? What do they do immediately after losing possession? How can we further break that down into potential ways to train it on the pitch?
Thoughts on Football Clubs
Richard Whittall had written a piece for 21st Club about Directors of Football, and I wanted to apply some consulting knowledge to football clubs in general and see what turned up.
The essential logic is this: it’s nearly impossible to know everything you need to know and have enough time to accomplish all the things a “traditional English manager” must do in modern football. And this is especially true when you realize the average life span for this role last approximately 12-15 months.
In short, it’s a recipe for failure.
What happens when you goof around with ideas about how to better train players, at the same time grabbing their attention and imagination, while subtly trying to explain analytical concepts? You get this.
It might be too dumb to be useful, but I would love to have license to test it enough to find out.
Introducing Manager Fingerprints Both of these articles are the building blocks for me trying to find a better way to evaluate managers than results in the league table. This work got much bigger and better once inside of the football clubs, and I’m pretty confident that – much like for player transfers – the data we analyze now will produce a better population of potential future head coaches for clubs than current hiring practices.
I hope you have enjoyed this look back through material both old and new, correct and horribly wrong. Football analytics is just like any other endeavour – mistakes will be made. The hope is that you also manage to get a lot smarter in the process.