I have talked on Twitter pretty regularly about using data to profile managers, both from a performance perspective and to detail style of play. However, what I have done very little of since I first started writing about Manager Fingerprints back in 2013, is practical public analysis. Let’s change that.
Coaching a football team is a funny thing. Amidst all the media appearances, press conferences, and televised games where your body language on the sideline will be scrutinized like the Zapruder film, coaches are also responsible for making strategic choices about how their team are supposed to play. What do they do when they have the ball? When the opponent has the ball? How do you transition from attack to defense and back again?
Then, once you have the strategy bit, you have to teach your players how to execute that at top speed, against the best players money can buy. This requires excellent analytical and communication skills. What went wrong? How do we fix it? How can I explain to each individual player what they need to see, how to understand that, and how to do better in the future?
It’s no surprise then that like tigers, coaches almost never change their stripes. They typically stick with the same style of play, year after year, and because of that produce similar statistical output. As I’ve explained before, coaching is an apprentice-based learning process. Unlike book learning, where you learn by reading, coaches learn by doing. They learn from mentors they have worked with before, and perhaps by explicitly seeking out other coaches who coach a style they want to learn. And not only do they have to learn how to create the style, and evaluate it in motion on the pitch in training and in games, they also have to learn how to teach it. Coaching well is an extremely difficult job.
With this in mind, it should come as little surprise that coaches whose teams don’t currently display certain tactical traits (like an aggressive zonal press without the ball) are unlikely to be able to coach that style in the future. The same is true for coaches whose teams don’t currently generate high quality chances inside the box. If they don’t already know how to create this, it’s unlikely they will magically learn it by taking a new job, at least not without supplemental learning or bringing in help from assistants.
If coaches rarely change styles, this means that data suddenly becomes a great way to profile what you can expect from various head coaches when they change jobs. Player quality can vary wildly between teams, but the style is likely to remain the same. It also means that searching for coaching candidates via data is now at least as interesting as searching for players in the transfer market. Both are expensive endeavours, especially when you make mistakes, so it makes sense to gather as much information as possible before making decisions.
Today I’m going to use these principles to look at how manager changes have impacted the relegation battle in the Premier League.
The headline number here is that Marco Silva came in at midseason and improved Hull’s expected goal difference per game by .59 goals a game, from -.95 to -.36. This translates to a goal difference of -14 throughout a whole season, a total that hasn’t seen a team get relegated in at least the last four seasons*. Silva hasn’t made Hull a great team, but he’s at least given them a fighting chance. They were relegation certainties under Mike Phelan.
(* Middlesboro are testing this number right now, mostly because their defense first policy under Karanka resulted in output that gave up very few goals but scored even fewer.)
It’s all well and good to say Marco Silva has made Hull dramatically better, but we care about the details as well. HOW are they better?
In attack they take about the same number of shots – a touch over 10 a game, but Silva’s managed to boost the quality of these significantly, from a putrescent .075 under Phelan (Hull would score an average of 7.5% of their shots) to .106. On the defensive side of the ball, we actually see the average quality of opposition shots has gone up a touch, which would seem bad at first until we look at the volume numbers. Phelan’s group were giving up 18.6 shots per game. Under Silva (as of this writing) it’s 14.2, which is a huge swing.
Beyond the headline expected goal numbers, there are subtle differences. Silva’s tactics have them defending a bit higher, and destroying the central block of opposition attacks more consistently. Average shot distance is now inside the penalty box, even from open play. Under Phelan, distance from open play was nearly 3m outside the box, which is baaaaaad.
Would Silva’s tactics be the same with a better group of players? Possibly not, but his output here is a classic recipe for what you can do to improve a seemingly doomed team, and give them a shot to avoid relegation. He seems like at least a safe pair of hands to hand any Premier League club to in the future, with a lot more potential upside.
Swansea’s battle against relegation this season has included three different managers, all with different preferred styles. What’s interesting to me is that Guidolin’s 16 games in 15-16 produced far better underlying numbers than his stint this year, but I think a lot of that is down to strength of schedule. Sure, Swansea started the PL season with Burnley and Hull, but then faced Leicester, Chelsea, Southampton, Manchester City, and Liverpool. Good luck surviving that run with non-relegation numbers.
The killer for Guidolin is the shot quality numbers in attack. Shot differential of -4.4 isn’t terrible, but cross that with a huge gap in shot quality and you’re likely to struggle. Would that have improved as they progressed into a softer schedule? Possibly. I said at the time I probably would not have changed managers at that point, and I still mostly think that, though it depends a bit on whether Guidolin had input into the poor quality of the squad coming out of the summer.
Bradley’s issue was the catastrophic defense. The style he played at Stabaek had elite athletes with pace that destroyed the midfield and then counterattacked. Swansea’s squad is the exact opposite of that – highly technical and slow. They destroyed better than under Guidolin, but the cost was a 50% increase in shot quality conceded. Bradley’s done well with existing talent in the past, but this squad was troubled and proved to be a poor fit for his tactical preferences.
This is something that can be teased out with proper manager candidate evaluations, both from a statistical perspective and from video scouting.
Finally we get to Paul Clement. On our podcast about a month ago, I cautioned that I thought Swansea’s performances probably weren’t as good as their results. Since then, they have lost five and drawn one, plunging straight back down into the relegation zone. On the surface, it looks like Clement has righted the ship. The defense is far less leaky than under Bradley or Guidolin, but it’s come at the cost of attacking numbers.
It’s all gone a bit Karanka, if we’re honest. They are defending and attacking well enough… to draw or regularly lose 1-0. The problem here is that draws won’t save them – they desperately need three points from as many matches as possible, and you only get 3 by scoring goals. The numbers are remarkably similar to what Silva is producing across the country. But… there’s a catch.
The last seven matches are one of their softest runs of schedule on the year for Swansea, playing Burnley, Hull, Bournemouth, Boro, Spurs, West Ham, and Watford in succession. The home game against Burnley is the only one where they won the xG race. You expect a bit of luck here and there, but being worse than your opponents every single week is still a recipe for relegation, which is why they once again find themselves in the bottom three.
The fact that Hull beat Swansea in a close match (and also Boro) is the major difference in survival right now. Maybe across a whole season with Clement, Swansea would have been fine – the numbers certainly suggest he’s improved them. However, with five matches left, Clement has to find some way for his guys to get a couple of wins or Wales will lose their Premier League representative for at least a season.
This was a brief look at the impact different managers have had on two of the teams fighting for Premier League survival. By using a variety of different key performance indicators, we can start to evaluate manager impact on squad output and to profile their styles of play. I didn’t touch on it here, but we can profile how teams attack and defend, whether they are good at set pieces on both sides of the ball, what their possession tendencies look like, and a whole lot more. With some higher-powered maths, we can also adjust performance by strength of schedule to correct for things like Guidolin’s tough run at the start of the year, to better evaluate how the team is actually doing.
Hiring and firing managers and their staffs is expensive. Data provides a much clearer picture of current and future performance under different managers than league table results, especially mid-season. We can use it not only to predict likely future performance, but we can also use it to profile coaching style and then help clubs find coaches that match their preferences for the style they want to see on the pitch.
Next week I’ll take a look at a manager change that happened a bit further up the table, and we’ll talk about the difference different big names can have on one of the super clubs.
If you want help evaluating your current manager or in finding your next one, get in touch.