As a Spurs fan and tactics blogger I spend roughly 6 or 7 hours a day thinking about Mousa Dembele. The brilliance of his unique skillset but also the downside of being quite reliant on a unique skillset that belongs to an injury-prone 30 year old. The result is having the question “Who can replace Dembele?” kicking around inside my head at all times.
I’m a firm advocate of analytical player scouting but a significant portion of Mousa’s play is barely or poorly represented in current fanalytics. The closest we currently have is the ‘dispossessed’ stat. Here’s Ted Knutson’s radar for 16/17.
Dembele doesn’t score highly. He is dispossessed a moderately high number of times per game but that doesn’t capture the wonder of how frequently he doesn’t lose the ball when putting himself into a situation in which the average player would.
This is how we arrive at Press Adjusted Absolute Ball Retention or just Ball Retention. I’m reluctant to give up the exact formula but basically we’re taking all of a player’s individual possessions to find the ratio of them that results in their team retaining possession of the ball and providing bonuses for successful dribbles as a proxy for pressure.
Low and behold, through no deliberate effort (I promise), Mousa Dembele scores very highly with 97% for the 16/17 season and 97.7% for 15/16. But he’s not top (of population trialled so far), that honour belongs to Mateo Kovacic with a whopping 98.8%!
I also tried my hand at a passing model akin to Knutson’s Passing Ability. Pass % but giving extra weight for long balls, key passes and through-balls; reducing the weighting for successful short passes. Much like Ted’s metric Toni Kroos not only tops but dominates what we’re calling Pass Efficiency.
With both metrics taking significant input from – but also adding new flavours and perspectives to – Pass % we’re going to see some crossover between the two metrics. But despite that they’re also somewhat at odds with one another. A player who manages to score highly (relative to population) in both metrics is simultaneously failing to give the ball away while also not being afraid to pull off some difficult or creative passes.
Combine that with a more advanced version of Paul Riley’s xG For Everyone model that takes into account penalties, headers and general shot choice quality to proxy shot centralisation and we could create a pretty decent predictive player radar based purely from publicly available statistics, templates and tools.
So that’s exactly what we did.