Recruitment is an area of particular interest to me. It’s here that data can have an enormous impact and the potential payoff is huge. Everyone wants to find the next Riyad Mahrez for £350,000. Data and data models cannot perfectly predict future performance or identify world class potential but what they can do is:
- Provide more information and a more holistic view
- Reduce risk. Passing the data test AND passing the eye test can give you more confidence that a player is good, as opposed to using only one or the other
- Save you a HUGE amount of time
Let’s use an applied example.
Villa fans will be desperate that this is not the case but let’s say Jack Grealish does move on this summer. How does the club begin to replace a player like that? Or maybe you are another club who looks at Jack Grealish and thinks… we could do with a player who does the things he does (on the pitch!).
There are tens of thousands of players in my database currently, clubs will likely have access to databases far greater in size. Where do you start?
Let’s do it.
Introducing a 3-stage approach
- Player Similarity Model
- Player iD Model
- Video Analysis
STEP 1: Player Similarity Model
First, I’d like to thank Eoin O’Brien (https://twitter.com/Eoin_OBrien_) for his help with this – I’ve been sat on this for a while and we joked that it appears a few people in the fanalytics community have been working on similar projects unbeknownst to each other!
My model is built in Python, and clusters players on a number of stylistical variables before using the Euclidean Distance to find the most similar players to the chosen player.
There has been some great work recently on this, for a more in-depth explanation of how these type of models work check out Eoin’s blog here: https://eoin-obrien.com/2020/08/24/factor-analysis-for-player-similarity/
ESSENTIALLY… this model skims through a database of thousands and thousands of players and shoots out the those most statistically similar to the chosen player. Importantly, this model is about style not quality. It’s less about how well or successfully they perform particular actions, and more about how frequent certain types of actions are for them, compared to others.
To find a player similar to Grealish, the variables used attempted to dig into player behaviours:
- When a player receives the ball how often do they pass/shoot/dribble/cross?
- When they pass it, how often do they go short vs medium vs long
- How often do they pass it forwards compared to backwards or sideways?
- How often are they involved in duels: defensive/offensive?
Essentially, the process for this model was: Build, test, tweak. It will continue to be an iterative process but after considerable testing and tweaking, the results it give seem fairly reasonable. Do let me know if any names jump out at you as being particularly off, on twitter @sanchoquinn !
RESULTS – Who is Most Similar to Jack Grealish?
So there are two more similar players to Jack Grealish than himself from 2018-19? This isn’t actually as bad as it seems. Grealish played more from the left in the Premier League as opposed to as a number 8 in the Championship the season previous, so his style is likely to have changed a bit in a different role. The fact that it still picked up Grealish 18/19 as highly similar despite the change in role, I was quite pleased about actually.
Canales has had another great season in Spain, James Maddison is in there, Rodrigo de Paul has turned some heads with his performances for Udinese. Marco Reus is the only real ‘superstar’ to appear but again, this is about style rather than quality.
Filtering by U25
Let’s imagine Villa are only looking for players who are young and will have some resale value, if we filter by U25 only, the list is as follows:
If we take out the top 2 as unfortunately not very realistic… let’s use STEP 2 of the process to see which of these players are any good.
STEP 2: Player iD Model
Some thanks again, Lorna Brown (https://twitter.com/_Lorna_Brown) and Sean Miller (https://twitter.com/HipsterVizNinja) helped me at different points when I was stuck in building this in Tableau – much appreciated both!
This tool ranks player performance on key position specific metrics. Players are ranked as percentiles compared to other players in their league and position.
A slimmed down version can be found here, for you to have a play with. If you’re interested in the full version with multiple leagues and multi-season data, get in touch (email@example.com)
Although some of the players below operate in slightly different roles, they were all analysed on the “Number 10” template to aid comparison.
For reference, here’s Grealish 19/20
Very impressive – high volume dribbles and progressive runs. Get’s a lot of shots, high xG but very creative numbers too. Only 22.
Fantastic numbers for a teenager in the Championship, albeit in very limited minutes. Looks more of a goal threat than assist threat.
Not particularly exciting
Hamed Junior Traore
Not bad in a good league, might be worth another look
Again, not particularly exciting.
His numbers are good, particularly in 18/19, but Everton would want to recoup much of the £30m they spent on him, so likely out of Villa’s price range.
Not Premier League level, based on his numbers
Interesting, outstanding dribble numbers, gets a lot of shots, low market value.
3 Options stand out to me as the most interesting:
- C. Ejuke – 22 year old LW for Heerenveen, valued at £5m
- Louie Sibley – 18 year old Number 10 for Derby
- I. Chouaref – 19 year old RW for LB Chateroux in Ligue 2, valued at £1.3m
STEP 3: VIDEO – Metrica Play
Time to check the tapes for our 3-man shortlist…
All of the video editing, and telestration is done solely with Metrica Play (https://metrica-sports.com/)
*Interestingly, part way through writing this blog it was announced that he has moved to CSKA Moscow for 12m euros. Keeping in as an interesting case study to see how he gets on.*
Dribbling and general technique
Ejuke is fantastic on the ball. Brilliant dribbler, who uses body feints superbly to freeze defenders and then is lightening quick over the first few yards to burst past his marker. Technically he is very sound, his touch is clean and rarely lets him down.
He knows he’s a superb dribbler and occasionally will take too many touches or try to do to much on his own, it’s here that his decision making can improve.
Some clips of Ejuke with the ball at his feet are below.
Although his starting position is wide left, Ejuke likes to come inside and combine with teammates. He’s shows good invention and creativity with his passing in the final third. Though he can’t always execute, he can certainly see the killer ball.
At 5”7 / 5”8 how he would translate to a more physical league would be a concern. He is a great athlete though, extremely agile, very explosive and a good engine. He looks incredibly quick and thanks to Metrica Play you can see just how quick he is in the below, reaching 34kmh+.
That’s quick. For reference the fastest sprints ever recorded in the Bundesliga can be seen here with Hakimi touching 36kmh, Aubamayeng 35.4kmh etc: https://www.bundesliga.com/en/bundesliga/news/fastest-players-ever-kingsley-coman-ehizibue-7503. And the Premier League here, albeit from a different tracking data provider: https://www.premierleague.com/news/650237
For clarity, this was Ejuke’s top speed in the limited games I watched, I imagine he could go slightly higher than that.
OUT OF POSSESSION
Naturally, he is a very attacking player. He does however show some promise without the ball, with good energy that could lend itself to a pressing system and some desire to get back into shape and recover possession for his team. His defensive effort can be slightly inconsistent, although this may be partly coach design to enable him to stay higher up the pitch to capitalise on transitions – where his pace and 1v1 ability make him a big offensive threat. He appears to have the engine to put in a real shift defensively if asked.
He wouldn’t be for every coach but he is a real handful. Some of his goals and assists are below. There’s some absolute belters in here. Bags of flair and really fun to watch.
Dribbling and general technique
Probably more so than Grealish, Derby’s Sibley is reminicist of Will Hughes – another former Ram. Might be partly aesthetics, the deft left foot, the bleach blonde hair… but there is a similarity in the style of play also.
Sibley is not a 1v1 specialist dribbler in the way Grealish but he does help his team progress the ball well in two ways
- He can run with it when he gets it
- He positions himself well between the lines to receive vertical passes
His touch is good, he’s often able to play on the half turn and then move the ball quickly.
Some of Sibley’s work in possession is evidenced below.
Though not a dribbler in the same way as Grealish, he does show good feet in tight spaces and is able to manoeuvre himself into shooting positions quickly, which can catch the keeper off guard.
Sibley has a real natural feel for the game. He has a good understanding of where to position himself on the pitch, when to carry the ball and when to lay it off quickly. He’s comfortable under pressure and his decision making, for his age, is excellent.
He isn’t a huge threat to unlock the door vs a packed defence currently. He can certainly see a killer ball in the final third, though the execution can be inconsistent – which is not a surprise at his age. Generally, he is very neat and tidy in possession and rarely wasteful.
He’d be an interesting option coming in off the right-hand side onto his left foot but he certainly looks more comfortable playing centrally, often dropping deep to help his team progress the ball.
Some clips of his use of the ball in possession are below.
It might be slightly to early to expect him to be able to step up to the Premier League level, but as an investment who you could perhaps loan back to Derby for a full season in the Championship, he would be an interesting option. A few of his goals/assists can be seen below.
He’s a decent athlete. He’s agile, he’s a natural mover, whilst not hugely quick, and he can cover distances well. As you would expect from a teenager, he could still do with developing more upper body strength and power in his running. He’s never going to be a powerhouse of a midfielder but he doesn’t appear out of his depth physically at all, which in itself is impressive for a teenager in a very physically demanding league.
OUT OF POSSESSION
Sibley is a committed and enthusiastic defender. He can press as part of Derby’s system and is willing to track back and put a foot it. One issue he has defensively is he can be easily dribbled at times, he can overcommit in situations where he might be better dropping a yard deeper towards his own goal.
The hair is even a bit Jack Grealish. +1 for the model
Very talented dribbler. Big 1v1 threat and you can see why his progressive runs numbers are so high. Really able to carry the ball long distances and maintain control. Good balance and agility, and a good turn of pace. He played more from the right last season but actually looks better from the left. He regularly has the beating of the full back coming inside but is often let down by his shooting technique.
This is evidenced below. The first clip looks a bit daft but I actually don’t mind it, he doesn’t have a passing option on and the keeper was off his line. At the very least, it shows he plays with his head up.
As evidenced above he has a real tendency to come inside from the right, beat his man but be let down by the shot. Interestingly, his crossing technique is a pretty good and pretty consistent, demonstrating that he can strike the ball well.
His passing is safe generally, but is capable of an incisive pass in the final third.
OUT OF POSSESSION
Looks a good athlete. TransferMarkt have him just under but he looks about 6”. Athletic frame, quick, good energy, looks fairly robust. At just 19 you’d expect he could develop further physically, too.
Without the ball he shows decent defensive instincts, screening passing lanes, recovering back into shape and tracking runners. He just needs to react/be aware that split second earlier, at times. He’s a willing defender and works hard for the team, albeit he can be a little agricultural with his tackling technique.
3 interesting players there. Louie Sibley, for those in the UK, is a player many will be aware of but I must confess the other two (Ejuke & Chouaref), I wasn’t familiar with before this work.
So. We’ve gone from thousands and thousands of players to 3 interesting options, by just starting with the idea of “I want *this* type of player”.
You could do Steps 1 and 2 in about 20 minutes.
The video analysis takes the most time and is still probably the most important. There is no replacement for watching players. And I struggle to imagine a world where that would be the case.
Using the 3 tools in conjunction seems a pretty decent method of filtering, time saving and triangulating information.
Hopefully you enjoyed this as much as I did putting it together. I hope to make this a bit of a series going forward.
Thanks for reading.
Sancho Quinn, 30/8/20