Stop bunching

My French league scouting has continued unabashed this week. There just seem to be an incredible amount of talented players at the moment. I've put together a free transfer XI of young talents from Ligue 1 and 2. I've left out the likes of Rabiot because they aren't realistic targets for Championship / lower Premier League clubs. Instead, I have a list of high potential young players who would probably be open to offers from teams at that level and whose wages would be affordable. GK - Abdoulaye Diallo Diallo, 26, is your classic reserve goalkeeper. He has been with Rennes his whole career, but with two long loan spells, one with Le Havre and one in Turkey. He is a full international with 17 caps for Senegal after a youth international career with France. He tends to play European and cup games for Rennes. A thigh injury has ruled him out at the moment, which is unfortunate timing with his contract set to end. RB -Patrick Burner Buner, a 22 year old, attacking rig

'Buying the dips' is a stock market strategy where you load up on share purchases after significant falls in value. The FTSE 100 share index hit 7000 points in 1999. It didn't reach those heights again until 2017. But at various times the market was as low 3600 points. Someone with £100k worth of shares who bought them in 1999 would only just be breaking even (excluding dividends) whilst someone who used the buying the dips strategy could have doubled their initial investment. The problem is telling the difference between a dip and a slump. And the same is true when buying players. Imagine 2 identical players: Player 1 has scored 15 goals from 10 xG in their debut season. Player 2 has scored 7 goals from 10 xG in their debut season. Who would be worth more at the end of that season? Player 1 for sure. In the second season, this situation is reversed. Who is worth more now? The steadily improving Player 2 who has gone from 7 to 15 goals. Or the overhyped Pla

I've been posting and tweeting a lot about Ligue 2 football recently. It is probably the best value market for smart clubs around. The reason for this can be neatly summarised in comparing the two major junior cup competitions in England and France. The FA Youth Cup and the Coupe Gambardella. The Coupe Gambardella is like the FA Youth Cup for U19 players. It is played on the same say as the Coupe de France and attracts large audiences. To demonstrate why I think Ligue 2 is the most exciting league for talent let's look at the finalists for the last 10 competitions in both England and France England: Chelsea x 8 Manchester City x 3 Arsenal x 2 Manchester United x 1 Fulham x 1 Norwich x 1 Aston Villa x 1 Sheffield United x 1 Blackburn x 1 Liverpool x 1 10 different finalists in 10 years France Montpellier x2 Nantes Metz Sochaux x 2 Monaco x2 St Etienne x2 Nice Bordeaux Sedan Auxerre Reims Lyon RC Lens Marseille Troyes Tours 16 different f

When Everton were strongly linked with Paolo Fonseca as a manager last year I made a point of watching Shakhtar Donetsk play as often as I could. This wasn't an arduous task as they played some of the most attractive football in Europe. A 4-2-3-1 formation with beautiful attacking interchanges and line breaking passes from deep. At the heart of this side was a deep laying Brazilian midfielder, Fred. His role was to take the ball from the back 4 and build play through clever passes into the feet of the 3 attacking midfielders. In the Ukranian league fixtures the defensive side of his game was rarely tested, so dominant were his team. But in European games, he showed he could cover ground really well and do his share of the dirty work. When Manchester United paid £52m for him in the summer I was a little surprised. Not because I didn't think he was a good player, but because I didn't think he was a Mourinho type of midfielder. And thus it has come to pass th

Ligue 2 isn't exactly a secret in football. France has a glut of talented players, there are only two fully professional leagues, and the talent is arguably more distributed than in other countries. The success of Guendouzi with Arsenal has shown it is possible for players to move straight from this league into the Premier League. Brentford, in particular, has benefitted from keeping a close eye on Ligue 2 with top scorer Neal Maupay and hattrick scorer at the weekend Said Benrahma both recruited from the division. The French leagues are particularly good at giving younger players playing opportunities. 123 players aged 20 or younger have played in Ligue 1 and 2 so far this season compared to 74 in the top two English divisions. This could be a reflection on greater talent in France or simply that clubs have less money to buy in players and rely more on youth development. Either way, there are a lot of young players getting minutes in a competitive league. In keeping with

Yesterday I posted a twitter poll. It was one of those "never going to happen" polls with an unrealistic question, but it gave some interesting outcomes. I added another caveat to make it even more life-like. Clearly, a lot of answerers will have built in their own interpretations around the length of time on the pitch, and the playing position, of each option. What I was really trying to tease out was whether people think outperforming xG is better than generating lots of xG but performing well below it. When Statsbomb posted the xG of Che Adams a few days ago it showed he was on 20 goals from 10xG which seemed a massive overperformance and probably unsustainable. But would I probably still scout him over a player on 20 goals from 20xG? Probably as I'd like to see if a truly elite finisher had been found. The most popular answer was to scout the player on 7 goals from 14 xG. This was a bit surprising to me. I expect most have taken on board the fact it

We've known, for a long time, that sample size is key to the accuracy of judging a player with data. The more minutes a player has played the more accurate our outcomes will be. Even then we need to be careful as a player with 900 minutes may have played 10 full games, or appeared as a substitute in 38 games, with the game state determining a lot of the type of actions they will be performing. And we also always have to take into account the style of play and system the player operates within. A wide attacker's output will be very different in a Warnock team to a Guardiola team. And also we have the problem that a team plays differently against each opponent. A full back will profile differently if only measured against the top 6 clubs than the rest. Actually, I'm not sure on that point, might be worth looking at. Oh, and the other problem is we only have data for those who play. So how can I judge the other 45 members of playing staff when I am only getting data fo

FOMO (Fear of missing out) seems to be the driving force in a lot of football club recruitment. There is no logical reason why Premier League squads should contain 70 professional players. In a typical week with 1x first team, 1x U23 and 1xU18 fixture only 33 players will start a game. Yet almost all PL clubs will continue to sign new players, or renew contracts of youth team players, long after they have shown they have little chance of being first team regulars. Take Chelsea, they already have Calum Hudson Odoi, Reece James, Ethan Ampadu, Ruben Loftus Cheek, Tammy Abraham and many more in their squad, not playing. Yet they will still bring in 3-4 first team players every season. They've already agreed to sign Pulisic for next season which further reduces the chances of some of their talented youngsters. And the problem is even worse at youth level with some clubs bringing 10+ youngsters from other clubs in at U15/U16/U17/U18 level every year. In one Everton vs Liverpool y

Recency bias means that we tend to put too much significance and value on things that have happened recently and underrate things from the past. You've probably seen this with a goalkeeper making a few high profile errors in a short period of time and being labeled as 'dodgy' forever more. I know many people who think Fabianski and Szczesny are useless because they let in some bad goals for Arsenal about 7 years ago. So how does data scouting get around the problem of recency bias? I've said many times that data scouting shows you how well a player is playing in their current role, in their current system. And this is actually one of the strengths of data scouting. Because I can go back in time and find when players have done well in their careers, and when they haven't. A player who thrived as a central striker in an attacking team with wingers may have spent two seasons on the bench since the new coach came in. That doesn't mean he became a terribl

One of the biggest problems in science is unreported negative or null findings. It leads to time wasted on replicating failed trials. So with that in mind I'm going to blog about expected assists as a percentage of expected goals and why it isn't really enough to tell us anything. My hypothesis was that we'd see the patient build up teams having relatively high xA as a percentage of xG and the longer ball teams with a low percentage. I took the stats from Understat, stuck them into a spreadsheet and got the following results: It looks OK. Arsenal and Chelsea are high possession teams. Cardiff and Crystal Palace more direct teams. They are about where I thought they would be. But then Manchester City are sandwiched between Huddersfield and Newcastle, and Liverpool and Manchester United are creating a lot more unassisted xG than my hypothesis would suggest. My theory now is that high pressing teams like Liverpool will generate unassisted xG from forcing mi

Can you really tell if someone is good at passing? One of the biggest criticisms of football stats is the use of pass completion as a metric. Anyone can complete 100% of passes if they only pass sideways or backwards. Would 40% completion from 5 passes be good? What if the two that succeed are defence splitting passes that result in two goals? Isn't that better than 50 accurate passes that create nothing? So how could you measure passing ability? Perhaps if you had the X/Y coordinates for every pass. You could then look at the completion rate of passes from X/Y to X/Y. The completion rate would give you a difficulty rating for that pass, under typical circumstances. So a pass from centre back to centre back would probably be completed 99/100 but a cut back pass within the opposition box may only be completed 20/100. You could then work out the expected passing rate of every pass played by a player, and the average difficulty of the passes they played. These pass

Football should be a simple game. The beauty of football is that it is anything but. Yes, the end objective is clearly defined. You need to score more goals than your opponents to win. The laws of the game are fairly straightforward too, defining concepts such as foul play, offside, and where a goalkeeper can handle the ball. But the endless variety of approaches is what makes the game fascinating. And perhaps the most fascinating area of the pitch is the midfield. Whilst there are defenders who attack, and attackers who defend, the true all rounders of the pitch are surely the central midfield players. Traditionally football talked of midfield partnerships; a pair of all round midfielders both capable in defence and attack. However as tactics have evolved we tend to find teams field three central midfielders. Now this is the point where slightly older football fans, like me, have to point out that nothing is really new and that successful teams have always dominated mi

A while ago I looked for how the principles of 'Moneyball' might be applied to football. One of my suggestions was to look for players with blocked pathways to the first team. This might be young players at elite clubs, or players where the first team spot is taken up by an excellent player. I randomly picked Lyon as my team to look at. They have a thriving youth academy, and lots of very good young midfielders. So is there anyone in their squad who is good but not quite Ndombele good? At the same time I've been using some good new data sets. And one name keeps coming up in my search for Ndombele like midfielders. In a happy coincidence that players happens to be on Lyon's books. Christopher Martins Pereira - Luxembourg and Troyes (on loan from Lyon) 21 year old Christopher is a 6ft 2 tall midfielder who can also play centre back. The majority of his games are as a defensive central midfield player. This season has been a breakout in terms of production with 2

Thinking about the Opta Pro Forum and new questions keep emerging. The main one surrounds data informed coaching. You had a room full of talented data scientists representing some of the biggest clubs in the world, lots of enthusiastic people with the skills to create something special. However, almost everyone I spoke to was more on the data/management side of things. I'm sure there are data enthused coaches, eager to see what data science can do to help, and indeed we heard from one on the day, but I'd love to see more coach generated questions. I had a quick chat with Mladen after his presentation with Dan, which had been about giving credit for space creating runs. When I said I had enjoyed the fact it seemed to be providing useful information the coaches could act upon, he said the coaching staff had given feedback about how to develop their model.  I think this is an important point. I know in my own career how annoying it can be when outside organisations use da

One thing that was mentioned time and again by the professionals at the Opta Pro Forum was the importance of character. You can be a skillful and productive player but if you don't have the right work ethic or have an abrasive personality you can end up as a net negative influence on the team. So how do we scout for personality? One of the slightly clich é d critisisms of data is "data can't show me heart", "data can't show me desire" But how would you quantify heart and desire? I'd probably look at actions performed at certain game states. Does a player "hide" when they are in losing positions or do they look for the ball as normal, or even more than usual? When I think of the likes of Gerrard or Rooney I remember players that would demand the ball more and more when their teams were losing. Often dropping deeper to get on the ball and make things happen. The boring analyst side of me wonders if this is a net positive or nega

As I approached the venue, a swanky conference hall near Euston station, I didn't know what to expect from a football analytics conference. I've been to plenty of conferences; from barcoding equipment (bottom end of the scale for glamour), to document management software, to management consultancy shindigs, where you generally have to feign an interest in the topics being discussed and struggle to stay awake in those dimly lit lecture theatres. I knew that this wouldn't be an issue today. As soon as I walked down the steps into the venue, passing a phalanx of young men with PSG and Arsenal logos on their badges I realised I was very fortunate to be invited to the conference. I collected my name badge and seeing only my real name on it knew I was cursed to stuttering introductions about how I blog and tweet under the name Stop Bunching. We began the day with a coffee and small talk with fellow attendees. Being the type of person who writes internet blogs this is the pa

In my set pieces article/rant last weekend I mentioned the set piece problems Everton were suffering from. But it is easy to get caught up in your own clubs problems. So I decided to do a bit more digging to see if we really were that bad. Every club thinks they never score from corners but let them in all the time. The problem is far worse than I thought. Figures are from Understat and Footstats. Understat provides us with a table of expected goals for and against broken down by situation. I have turned them into lovely graphs for you, which is really nice of me. First off corners: To do this I used the expected goals for and against figures from Understat, then the number of corners for and against from Footstats to calculate an expected goals per corner stat. Teams that are good at corners will have a blue bar above the red line. The worst expected goal difference per corner, by a mile, is Everton followed by Manchester City (who don't need to worry) and Hudder

Whenever I read, or write, an article about data scouting, someone, somewhere, will reply "data scouting isn't moneyball". Often this is under an article that doesn't even mention the m word. There is no getting away from Moneyball. Any football team that shows any indication of using "stats gurus" or "boffins" or "men with laptops" is said to be using a Moneyball strategy. Moneyball was a system where data was used to buy undervalued, or sell overvalued, baseball players. The system of analysing data is referred to as sabermetrics. I'm no expert on baseball, but from my one game watching the Toronto Blue Jays about 15 years ago, I feel confident enough to say it isn't as dynamic a game as football. You don't have 22 players moving about in different directions constantly. The data is easier to collect and far more established. So the cynicism about Moneyball in football is well founded? Well, no. The guiding prin