Are Short Goal Kicks Worth the Risk?

Plus: introducing goal kick chains!

Mikel Arteta was not super stoked over the weekend to talk about Arsenal’s flaming three-ring circus of a buildup that ended with Granit Xhaka banking a pass off a Burnley player into his own net. “We gave them a goal in a situation we train for,” he said in a Muscadet-dry monotone, visibly dreading the risk-reward discourse that bubbles up every time another blunder at the back makes headlines. “It can happen when you play long,” he pointed out. “The only chance Burnley had was from a long goal kick from Bernd.”

If you’re a regular space space space reader, you know we take the “Is building from the back really worth it?” question seriously around here. Sure, good teams tend to play shorter from the keeper, but is it the secret of their success or just a fad right now for clubs rich enough to get away with it? The latest and greatest possession value models like g+ and VAEP love field position and vertical velocity, the two qualities slow buildups sacrifice in hopes that greater control will pay off in the long run.

But models can’t tell you everything. The most important thing they get right is how they frame the question: In the foreseeable future, over however many seconds or actions or possessions, will the thing I’m about to do make it more likely that my team will score or concede?

You don’t need a fancy model to think about soccer in those terms. American Soccer Analysis’s Jared Young did the math on goal kicks years ago and reported that two-possession goal difference favored MLS teams that played short rather than long. Tom Worville surveyed the effect of the new goal kick rule at last year’s Opta Pro Forum and found that shots for and against Premier League teams following goal kicks under 40 yards came out even over the next two possessions. That sounded like a better deal than the negative shot difference for long goal kicks, though the quality of the teams involved, differences in xG per shot, and limited samples made it hard to draw firm conclusions.

To build on those guys’ work, I started by grouping goal kicks into ten types, using the same k-means clustering algorithm as our progressive pass survey a couple weeks ago. Then I used expected goal differences over two and four possessions to look at some questions about risk and reward. Which kinds of goal kick lead to better outcomes over hundreds or thousands of reps? Can smaller clubs with less talented players succeed at the short goal kicks rich clubs love? And what does a good short routine look like, anyway?

The Ten Types of Goal Kick, and Why Short Ones Are Generally Worth It

To define our pass groups, I ran the clustering algorithm on seven seasons’ worth of Premier League goal kicks, nearly 40,000 in all. There’s no right number of clusters, but after trying it out different ways I felt like these ten groups offered a good mix of detail, balance, interpretability (you can close your eyes and picture how each one plays out differently, right?), and big samples, with thousands of passes in each group.

The differences in outcome by pass type are pretty stark. The short clusters A, B, and C are the yellow-greenest, meaning teams taking those goal kicks create better chances than they allow over the next four possessions. Long kicks up the middle are next best, even though they’re completed just 29% of the time, while goal kick attempts that land a little shy of the halfway line are the purplest, meaning they’re most likely to end in a goal for the defense. There are so many passes in each of these clusters that I’m a little surprised by the difference between D on the left and E on the right (maybe it has to do with more keepers being right-footed?), but otherwise there’s a nice symmetry to the whole thing.

To be honest, measuring these things by expected goal difference over the next four possessions (xGD4) is probably opening the window a little too wide. At some point you’re no longer capturing the effect of the goal kick so much as the influence of complicating factors like team strength or the scoreline. But I wanted to try to include the structural and field position advantages of different kinds of buildup, which could conceivably take more than one turnover to produce a shot. Anyway, even if xGD4 is too much, two-possession xGD tells pretty much the same story. The three types of goal kick associated with the best outcomes—in fact, the only kinds of goal kick that aren’t a net negative over four possessions—are the three short groups.

I know what you’re thinking: short goal kicks look good because good teams take short goal kicks. That’s definitely true to some extent, and I don’t have the stats skills to control for it in any sophisticated way, but even when we look outside the Big Six at smaller clubs whose players are less likely to be exceptionally good on the ball, short goal kicks tend to turn out at least as well as long ones across four possessions.

There’s probably some self-selecting going on here. The teams that take short goal kicks, whether they’re rich or not, are the ones that train them enough to feel pretty good about their chances of success. Burnley, for example, have taken an overwhelming majority of their goal kicks long this season (shocker, I know). It hasn’t been going great—they’re considerably more likely to give up a goal off one of their own goal kicks than to score from one—but when you compare it to their even worse outcomes in the three short goal kick clusters A, B, and C, it looks like Sean Dyche knows exactly what his team is about.

Some of these clubs could maybe stand to be a little more adventurous, though. Are Crystal Palace and Southampton only getting better results from short goal kicks because they’re so selective about when to try them, or are they better than they think at playing out of the back? Liverpool loves short goal kicks more than any team in England, but looking at their awful numbers on long ones it’s fair to ask if a 75/25 blend isn’t enough. As for Arsenal, who actually play long a little more often than short, their much worse expected goal balance on long goal kicks shows Arteta wasn’t kidding when he talked about how risky it can be.

But okay, sure, say you’re a coach who wants to play more short goal kicks. How should you do it? What kind of patterns after that first short pass are most likely to get you to goal?

Introducing Goal Kick Chains

To take a crack at that question, I pushed the k-means trick a couple steps further, piling clusters on clusters like it was harvest season in Mendoza. The idea was to identify the best three-pass sequences, along the lines of what analysts used to call “motifs” and Opta has lately been calling “movement chains.”

The most successful of our ten main goal kick types was group B, but I felt like that was probably a situational thing where teams that pass short up the middle aren’t under much pressure to begin with. So I turned to the next best cluster, group A, which just so happens to have a corresponding group C on the other side. I decided to lump those two together. Mirroring the right-sided group so its field coordinates match the left-sided group may hide some interesting asymmetries, but it doubles our sample, allowing better xGD measurements from slices that get smaller and smaller the farther our chains get from goal.

Starting with completed goal kicks in groups A and C, I divided successful second passes into the five groups marked with a circled 2 in the viz below. There are no player positions associated with these clusters, but based on their average distance and direction it seems fair to call them “Midfield,” “Switch,” “Fullback,” “Forward,” and “Keeper,” a pretty commonsense menu of passing options for a defender who’s just received a goal kick somewhere around the corner of the box.

Next, from each of those five successful second pass types, I ran three more clusters of possible third passes. For this last group, I didn’t specify whether the pass was completed or not; I just wanted to know what the two-possession xGD looked like if the third pass attempt traveled in that direction. Basically, which three-pass goal kick sequence is best?

So we start with a completed short goal kick, then one of five completed second passes, then three more possible pass attempts sprouting from each of those, for a total of 15 possible combinations. Now you start to see a wider range of xGD2 possibilities:

I could write a thousand words on every single one of these pass chains (why is GK-CB-GK-same CB better than playing over or away from the pressure? why does playing the second pass out to the fullback lead to the two worst outcomes but also one of the best?). But for now I want to focus on the chains that look most promising for the goal kick team, those two bright yellow ones after a switch.

What makes these chains so effective? The second cluster, the pass from (1) to (2), includes various kinds of switches. Some are the big, heroic lateral switches I argued teams should do more of in the Big Book of Buildups. But here’s a different example of one of these chains that I love specifically because it’s so mundane:

Tottenham doesn’t do anything fancy in this clip. By receiving short on the left side of the box and turning back inside instead of kicking it out to Sergio Reguilón on the left, Eric Dier steers clear of the low-xGD2 Fullback goal kick chains, toward the higher-percentage Switch chains. You can see right away why it matters.

Crystal Palace’s standard 4-4-2 press steps forward in response to the short goal kick and the two forwards rotate in a conventional way: the right forward, Wilfried Zaha, closes down the ball carrier while the left forward, Christian Benteke, drops to prevent the forward pass to Pierre-Emile Højbjerg in defensive midfield. That leaves Spurs’ other center back, Toby Alderweireld, as the free man. The second-pass “switch” here that opens up the play is nothing but a 15-yard shovel from one defender to another in space.

But it does open up the play. Benteke can’t step to Alderweireld right away because he has to follow Spurs’ second pivot, Moussa Sissoko. Sissoko responds by swinging out to the fullback hole, sending the actual right back, Serge Aurier, up the wing and forcing Palace to switch defensive assignments: the left midfielder steps up to keep an eye on Sissoko while Benteke starts to close down Alderweireld. It’s too late. Alderweireld has had plenty of time to pick out Aurier over Palace’s distended midfield line.

Like I said, there’s nothing special going on here. It’s a totally normal sequence of actions and reactions from a 4-4-2 press defending a short goal kick taken in a 4-2-3-1. That there’s an almost logical necessity to how it plays out suggests, I think, why these Switch goal kick chains light up on the viz. When Tottenham’s shot finally comes, it’s not because they played over the press to gain field position, won a second ball, and broke for goal. It’s because a good short buildup gets them into the attacking half in a controlled way and lets them pick their point of attack. If only it weren’t Tottenham, Mikel Arteta would have been proud. ❧

Thanks for reading space space space! This letter is free, so feel free to share it, and please consider becoming a paid subscriber to get more letters like last week's deep dive on the recipe for a perfect throughball.

Further reading:

• Jared Young, Forget Everything You Think You Know About Goal Kicks (American Soccer Analysis)
• Tom Worville, Long or Short? How the New Goal Kick Rule is Impacting Football (Opta Pro Forum)
• Duncan Alexander, The Exploration of Space Through Goal Kicks (Stats Perform)
• Bill Connelly, Why Bayern, Liverpool, top clubs will never stop playing out from the back, even when it goes horribly wrong (ESPN)

Image: Design for the Embroidery of the Lower Front Part of an Overskirt of the "Fabrique de St. Ruf"