Introducing Manager Fingerprints
This article was originally published on the OptaPro Blog.
Managers get a bum rap. For guys allegedly only responsible for 15-20% of a team’s performance (Soccernomics), they tend to catch a ton of flack when things go wrong, while the players catch most of the praise when things go right. Most managers are compensated fairly handsomely for their burden, but for a job that some view as simple as choosing which players to play each day, it seems pretty harsh.
Personally, I’m unconvinced managers have so little influence. Players matter a lot, but so do tactical systems. Give Sam Allardyce an identical set of players as Pep Guardiola, and you will get substantially different statistical output. You might even get different game outcomes (win/lose/draw). In fact, that’s one of the things that bothers me about football - there is only one way of evaluating managers. Results.
Did you win enough, Y/N?
Obviously you can contextualize that a bit and investigate whether a manager won enough versus expectation, or versus monetary expenditure or whatever, but at the end of the day results are the only widely used method to determine whether a manager is doing a good job. This is a problem.
Think of it another way…
You are the owner of an English Premier League football team. Your manager has taken a new job and now you need to search for a new manager. How do you evaluate candidates for the position?
Points per game at past jobs? That’s it?
Considering that the person you hire has an enormous influence on a company/investment that is likely worth hundreds of millions of pounds, that’s a little terrifying. What if they were lucky before? What if they inherited a ton of talent (and money) compared to other teams in the league and rode that to positive results?
Additionally, how can an owner – someone who is probably not completely immersed in football – know about what style of play a particular manager uses? Or 20 managerial candidates? Remember when Owen Coyle was branded as playing expansive passing football at Bolton, and then Zonal Marking did a tactical review and discovered that Bolton were among the league leaders in long balls? That type of thing happens all the time. Figuring out signal from noise in that environment is nearly impossible.
A Statistical Framework
This is where stats come in to play. Whenever I want to evaluate something intelligently, I always come back to the data. Managers influence the way their teams play via tactics. Some teams sit deep and rely on set piece expertise to grind out wins (Stoke under Tony Pulis). Some teams pass their opponents into oblivion (Barcelona under Pep and Villanova). Other opponents play fairly normal, positive football until they score, at which point they shell and unleash lightning counter-attacks.
All of these different tactical variations affect the stats teams produce. By focusing on statistics that actually matter (ones that have a positive correlation to points produced), we can create a “managerial fingerprint” that tells you how future teams a particular person manages are likely to perform.
For now I want to focus on five simple stats, four of which involve shooting and one that provides context and hits heavily on passing.
The shooting stats are Shots For per game, Shots Against per game, and then the percentage of Shots on Target, for and against. These stats take a step back from goal differential, but still provide a fairly accurate picture of what is happening in the areas of the pitch that matter.
The contextual stat is possession. Some systems need heavy possession to create chances, while other ones prefer to set up a bunker defense and then attack quickly out of it. Possession tells you a lot about how a manager wants their team to play (and at least a little about what type of football you’ll be giving to fans).
All of these stats correlate with winning, but there’s one other thing you need even out, and that’s the league dynamic. Teams in the Bundesliga and Spain average a much higher percentage of shots on target than those in England. Therefore a manager whose team is allowing 33% Shots on Target against in Bundesliga, is actually 2-3% better than league average. The same number in England would be a half percent worse than the league average, so league context remains important.
I’m going to delve into additional specific managers here on OptaPro in the future, but today I wanted to look at the fingerprint for one of the most well-travelled managers in recent Premier League history:
This is Mark Hughes at Blackburn. At the bottom there you can see the 38 games prior to when Hughes came in, including a disastrous 5 game stretch under Souness before he upped sticks and moved to Newcastle. The Shot Dominance (ShDom) numbers under Hughes are solid, and they would generally produce a happy midtable side for just about any team. They were mostly indifferent toward possession, basically floating around 51% the entire time. [Note: There’s a legend that explains all the abbreviations at the bottom of this article.]
However, EPL Shot on Target percentage is usually around 32.5%, and the Shots on Target Conceded percentage (SOTC%) in the second two years of Hughes’s reign would be worryingly high and possibly indicate a need to invest in better personnel there. In fact, the year after Hughes left, Rovers finished 15th, while conceding the third most goals in the league.
Hughes was hired during the summer of 2008-09 to help take Manchester City on to bigger and better things. The personnel he had at City was considered much better than what he had at Blackburn, and obviously given the Abu Dhabi takeover, they spent much more money as well. His first season included players like Robinho, Joe Hart, Micah Richards, Vincent Kompany, Pablo Zabaleta, and Craig Bellamy. By the start of Hughes’ second season in charge, he had also had Gareth Barry, Carlos Tevez, Adebayor, Kolo Toure, and Joleon Lescott on the playing staff.
That’s just weird. Despite talent that was massively better than what Hughes had at Blackburn, City’s statistical profile looks almost identical to the last place Hughes managed with one major difference – the shots on target percentages on both sides of the ball were much better. That’s it though. The tempo they played was a little faster (a touch more shots for and against), but given the different talent, I would expect the ShDom and possession numbers to differ much more than they do. The ShDom numbers in particular though, are quite a bit different than the football City produced under Sven Goran Eriksson.
The next stop on the Mark Hughes Managerial Tour was London, where Hughes took over Fulham after Roy Hodgson bounced off to West Brom (via Liverpool *ducks*).
From Blackburn to the nouveau riche of Manchester City and back to Fulham, and the underlying numbers remain the same. The Shots on Target for the offense is lower (in my opinion, this is again probably a result of talent disparity), but that is it.
Finally, we have QPR. Hughes replaced Neil Warnock midseason and was there for 30 matches before being replaced himself by Harry Redknapp.
Neil Warnock didn’t look like he could save this club. Mark Hughes never really improved them during his 30 games in charge bar the possession numbers. Nor did Harry Redknapp, whose shot dominance and shots on target numbers were basically the same as Hughes. I can’t really explain this one except to say that QPR just couldn’t do anything well enough on either side of the ball to survive, and none of these managers was clever enough to figure out how to fix them/keep them in the Premier League.
Mark Hughes’ Statistical Fingerprint
Looking at all of this – a little over eight full seasons of data – I would conclude that Hughes is an average manager. He’s able to consistently produce a style of football that is likely to finish in the middle of the table. When presented with a hugely talented team at Manchester City, he was unable to find a way to further maximize their skills. The average shot dominance number for his career is bang on 1, and both shots on target for and against are around the league average as well.
Statistically, Mark Hughes is almost the epitome of the average Premier League manager.
Some people would probably roll their eyes at this. “Oh boy, you found someone average!” In reality, this is kind of a big deal. If you hire Mark Hughes, at least you know you aren’t making a mistake. Teams in the top half of the table that want to improve probably don’t want him, but if you are a bottom half team attempting to simply secure regular survival, Hughes seems fine.
Sometimes fine is good enough.
Managerial Fingerprint Conclusions
The point of all this is that teams need additional methods to evaluate managers beyond “did they win enough?” That sliver of information is too thin to make judgments where tens or hundreds of millions of pounds can ride on the outcome (especially given the cost of firing managers and coaching staffs these days).
Going beyond the points per game approach and looking at the underlying statistics that matter is the first step in building a new framework that will help football organizations evaluate current personnel as well as future managerial hires. It can also help currently employed managers know the areas they need to improve better. This information may seem fairly basic right now (and obviously plenty of additional statistics can be layered on top of it), but despite the simplicity, it captures stats that have a significant relationship to winning.
There are plenty of additional questions that this framework can help us answer. Here are just a few:
- Do elite managers tend to dramatically change the statistical output of teams when they take over?
- When teams hire new managers, does statistical performance improve in the second season? (Given the average manager only last something like 1.5 seasons these days, this is kind of a big deal.)
- Can new managers elevate or decrease the performance of identical talent?
- Review past managerial hirings and firings to determine whether they made solid statistical sense, or whether they were destined to fail from the start.
Using the managerial fingerprint framework, I will start to delve into some of these topics on future posts here and on StatsBomb.com. It’s far from finished, but it certainly provides additional data for making better managerial decisions.
Shot PG – Shots For, per game
SOTPG – Shots on Target For, per game
SOT% - Shots on Target percent
ShAgPG – Shots Against, per game
SOTCPG – Shots on Target Conceded, per game (sometimes written as SOTCON)
SOTC% - Shots on Target Conceded percent
ShDom – Shot Dominance. This is a metric that looks at how many shots a team takes versus concedes against average competition. Greater than 1 means a team takes more shots than average opponents, less than 1 means they take fewer than the opponent. This is a different way of working with TSR (Total Shots Ratio)
Poss – Possession
Avg_PPG – Average points per game in the league table for that season.