What's Tackles? Defense is About Denying Space

A new pass prevention model measures how players defend off the ball.

You probably don’t remember it, but John Stones might have rescued Manchester City’s Champions League hopes yesterday. In the 60th minute of the second quarterfinal leg, tied 1-1, Dortmund broke through City’s press from deep and sent Erling Håland galumphing at an exposed back line. Stones had to time his tackle perfectly. If he committed too early, Håland would slide a ball past him to put Mahmoud Dahoud in on goal. If he waited too long, Håland would murder a shot from the Håland Zone. One rule of thumb in defense is that it’s generally bad to let opponents shoot from places named after them.

The reason the play’s not that memorable is that Stones picked the secret third option, none of the above. Instead of trying to win the ball, he eased out of the back line and got just close enough to usher Håland outside, cutting off the throughball angle to Dahoud every step of the way. Then he trotted back into the box and casually cut out the low cross while texting his assistant about where to order good delivery in Paris. City was going to the semis.

How fitting that the non-play of the game would come from Stones, who’d been the poster boy for everything supposedly wrong with Pep Guardiola’s preferred style of defending during the coach’s bumpy first season in England. “I am not a coach for the tackles,” Pep once told a reporter who asked if squad’s ballwinning stats showed they needed more fight. “What’s tackles?” Four years later, City has the best defense in Europe and Stones — who still can’t tackle for shit — is their rock.

The problem with defensive work that doesn’t involve winning the ball is that it’s hard to define, let alone measure. As former USMNT manager Bob Bradley told me in last week’s longform interview on how he sees the game: “You want to know if a guy is in a spot defensively and therefore the first three best options can’t occur just because he’s there, rather than a moment of that guy tackling.” Event data, Bradley said, “doesn’t take you anywhere” by telling you what’s happening on the ball.

But without data, the only way to compare players at scale is to chain a thousand video analysts to a thousand Wyscout accounts, and they’ll still probably come away arguing about what good individual defending even is. “Oh goodness,” my friend Mark Thompson wrote when I asked him to define defense, a subject he’s written thoughtfully about in his newsletter. “Have the ball. If you don’t have it, win it. If you can’t win it, keep them far from goal. If you can’t keep them far from goal, ensure bad quality chances. Or something.”

Most defense isn’t ballwinning — it’s a continuous fight for space. But how do we know who’s doing it right?

(Almost) Everything is Pass Prevention

At last month’s Stats Perform Pro Forum — probably soccer’s most exciting annual nerdfest, depending on how excited you get about this sort of thing — the Bengaluru-based analyst Aditya Kothari (@thecomeonman on Twitter) presented the best method I’ve seen for measuring how players defend space. Officially, his talk covered two models, one for pass prevention and another for shot prevention. But since the pass prevention part can also handle carries, Kothari told me that when you put the models together, they cover “really the three things that anyone’s doing when they have the ball.”

The basic idea behind pass prevention is easy enough: first you measure where a defender is keeping the ball from going, then you credit him with that space’s value to the attack.

The measuring part starts with something called “pitch control,” made famous by CERN physicist-turned-Liverpool FC analyst William Spearman. The math is fancy, but the soccer concept is as simple as it gets. Based on where they’re positioned and how they’re moving, every player “controls” some area of the field just by having a good chance of getting to a ball there. Defenses try to prevent goals by controlling the spaces that attackers want to play through. Makes sense, right?

Kothari came up with a trick to capture each defender’s contribution to the team effort: If you calculate which areas each player controls at any given moment, then delete one guy from the defense and measure everything again, the difference between the two scenarios — all that extra space where attackers would be more likely to receive a ball if the vanished defender weren’t there — is the part of the pitch where that guy is helping to prevent a pass.

The next step in the process is appraising what our defender’s real estate is worth. Here you need a possession value model to tell you how much more likely the opponent would be to score if they moved the ball into the area where he’s preventing a pass. To save time, Kothari used a simple estimate based on distance to goal, but there are other techniques out there that could probably give you a better number. The important thing was to assign these blobs of space some kind of value to show that preventing a pass to the penalty spot is more important than preventing one out to the sideline in the middle third.

If all the most valuable space is right in front of goal, why not just park the bus there? “Pass prevention is sort of an incremental value,” Kothari explained. “If all eleven players are standing at the goal, it’s going to be a really low-value space for those eleven because they’re all guarding the same territory.” By spacing out in an organized block, defenders control more of the pitch and avoid making themselves redundant. Even when you measure it individually, defense is a team effort.

Meet the Model

To show how pass prevention works in action, Kothari ran the model on a Kevin de Bruyne throughball against Leicester City the other week that produced what the Man City beat writer Sam Lee called “the most Guardiola City goal of the last five years.” Here, you’ll see what he means:

This video focuses on the nearest defenders’ pass prevention values on the throughball to Gabriel Jesus. Using the technique described above — measure pitch control, erase a defender, measure pitch control again, then calculate the possession value of the space that missing defender was covering — Kothari is able to show, frame by frame, how Jesus goes from being well marked by Leicester City’s left wingback, Timothy Castagne, to getting free in the channel so that he’s got access to high-value space behind the back line where neither Castagne nor the left center back, Wesley Fofana, are likely to beat him to the throughball.

In the freeze frames below, at the moment de Bruyne plays the pass, you can see how Castagne (green #27) is caught in a two-on-one against Jesus (white #9) and Ferran Torres (white #21). The white pitch control shading shows where each defender is preventing a pass to a given attacker. For example, if you look at the pitch diagram in No. 27’s column and No. 9’s row, you’ll see a ghostly triangle where Castagne is making it less likely that Jesus can receive a pass. Notice how the red plus sign showing where the actual throughball arrives falls just outside Castagne’s reach — and also goalkeeper Kasper Schmeichel’s (No. 1) and Fofana’s (No. 3). The model did a good job of predicting that the defense wasn’t likely to get to this one.

But anybody can watch a play and tell you whether a given attacker is likely to beat a given defender to a ball. When you repeat Kothari’s method for a lot of defenders and a lot of attackers and a lot of passes, that’s when you start to see the real juice. A video analyst would have to rewind a single clip a bunch of times to make careful judgments about what each player is or isn’t doing defensively. Tracking data can’t see quite as much soccer detail as a human, but it can measure a whole bunch of it at once. Suddenly Bradley’s vision of what the ideal defensive data should look like — “You want to know if a guy is in a spot defensively and therefore the first three best options can’t occur” — doesn’t sound so pie-in-the-sky.

Our Space-Age Future

As a tool to measure players’ defensive ability and tendencies, tracking data models like Kothari’s are light years ahead of tackles and interceptions per 90’. When we talked not long after his Pro Forum presentation (whose video should be made public soon), Kothari was already brainstorming how he could use pass prevention to measure stuff like the chemistry in teammates’ defensive handoffs, which would have sounded like sci-fi to data analysts a few years ago. Soccer’s explosion of off-ball data should bring you more of this kind of thing in the near future. Maybe it’ll even help turn up the next John Stones.

Still, Kothari’s realistic about his model’s limitations. “One of the questions at the Pro Forum was, ‘Can you reverse engineer this model in a way that you can calculate optimal positions or optimal actions they should have taken?’” he said. “I don’t think this model is the right way to address that, because that sort of gets into questions of cause and effect. Is the position of the defender preventing the run? Is the attacking player not making the run and therefore the defending player is not worried about a particular part of the pitch?”

Measuring space and movement is one thing. Game theory-ing your way through why players do what they do is another. “I’m not optimistic about being able to solve, like, causality for football,” Kothari said. But hey, maybe that’ll give Pep something to work on when he’s not coaching tackles. ❧

Thanks for reading space space space! This letter is free, so feel free to share it, and please consider becoming a paid subscriber (on sale right now).

Further reading:

William Spearman, Beyond Expected Goals (Sloan 2018)

Ricardo Tavares, Attributing xG Allowed to Individual Defenders

Mark Thompson, How Can Tracking Data Help Defending? (Get Goalside!)

Image:  Fish hook, ca. 1295–1070 B.C.